feat: bump transformers to allow Qwen3-VL (#225)

remove unsupported cuda versions (12.1-12.3) from hub.json and tests.json
to fix compatibility issues with worker deployment

- hub.json: remove 12.1, 12.2, 12.3 from allowedCudaVersions
- tests.json: remove 12.1, 12.2, 12.3, 12.4 from allowedCudaVersions

refs: AE-1452

.github/CONTRIBUTING.md

@@ -0,0 +1,79 @@
+# Contributing to worker-vllm
+
+## 🚀 Release Process
+
+### Development Workflow
+
+1. **Feature Development**
+
+   ```bash
+   git checkout -b feature/your-feature-name
+   # Make your changes
+   git push origin feature/your-feature-name
+   ```
+
+   - Creates pull request → triggers dev build: `runpod/worker-v1-vllm:dev-refs-pull-214-merge`
+
+2. **Main Branch**
+   ```bash
+   git checkout main
+   git merge feature/your-feature-name
+   git push origin main
+   ```
+   - No automatic builds on main (staging area)
+
+### Creating Releases
+
+**Method 1: GitHub UI (Recommended)**
+
+1. Go to [Releases](https://github.com/runpod-workers/worker-vllm/releases)
+2. Click **"Create a new release"**
+3. **Tag version**: `v2.8.0` (with "v" prefix, semantic versioning)
+4. **Target**: `main` branch
+5. **Title**: `Release 2.8.0`
+6. **Description**: Brief changelog
+7. Click **"Publish release"**
+
+**Method 2: Git CLI**
+
+```bash
+git checkout main
+git tag v2.8.0
+git push origin v2.8.0
+```
+
+### What Happens Automatically
+
+✅ **GitHub Release** created (if using Method 1)  
+✅ **Docker Image** built and pushed: `runpod/worker-v1-vllm:v2.8.0`  
+✅ **Documentation** updated with new version references
+
+## 📋 Version Format
+
+- **Format**: `vMAJOR.MINOR.PATCH` (e.g., `v2.8.0`)
+- **With "v" prefix**: Use `v2.8.0` for git tags
+- **Semantic Versioning**: Follow [SemVer](https://semver.org/)
+
+## 🐛 Development
+
+### Running Tests
+
+```bash
+# Update test configuration in .runpod/tests.json
+# Tests run automatically via RunPod platform
+```
+
+### Model Updates
+
+- Update `MODEL_NAME` in `.runpod/tests.json` and `worker-config.json`
+- Ensure model has vLLM support and chat template (for OpenAI compatibility)
+
+### Environment Variables
+
+See [README.md](../README.md) for full list of supported environment variables.
+
+## 🔧 CI/CD Workflows
+
+- **Dev builds**: All pull requests → `dev-refs-pull-<PR#>-merge` images
+- **Release builds**: Git tags → versioned images + GitHub releases
+- **Manual triggers**: Available in GitHub Actions for emergency releases

.github/workflows/CI-runpod_dep.yml

@@ -19,32 +19,49 @@ jobs:
 
       - name: Check for new package version and update
         run: |
-          # Get current version
+          echo "Fetching the current runpod version from requirements.txt..."
-          current_version=$(grep -oP 'runpod==\K[^"]+' ./builder/requirements.txt)
           
-          # Get new version
+          # Get current version, allowing both == and ~= in the search pattern
+          current_version=$(grep -oP 'runpod[~=]{1,2}\K[^"]+' ./builder/requirements.txt)
+          echo "Current version: $current_version"
+
+          # Extract major and minor from current version
+          current_major_minor=$(echo $current_version | cut -d. -f1,2)
+          echo "Current major.minor: $current_major_minor"
+
+          echo "Fetching the latest runpod version from PyPI..."
+          
+          # Get new version from PyPI
           new_version=$(curl -s https://pypi.org/pypi/runpod/json | jq -r .info.version)
           echo "NEW_VERSION_ENV=$new_version" >> $GITHUB_ENV
+          echo "New version: $new_version"
+
+          # Extract major and minor from new version
+          new_major_minor=$(echo $new_version | cut -d. -f1,2)
+          echo "New major.minor: $new_major_minor"
 
           if [ -z "$new_version" ]; then
-              echo "Failed to fetch the new version."
+              echo "ERROR: Failed to fetch the new version from PyPI."
               exit 1
           fi
 
-          # Check if the version is already up-to-date
+          # Check if the major or minor version is different
-          if [ "$current_version" = "$new_version" ]; then
+          if [ "$current_major_minor" = "$new_major_minor" ]; then
-              echo "The package version is already up-to-date."
+              echo "No update needed. The new version ($new_major_minor) is within the allowed range (~= $current_major_minor)."
               exit 0
           fi
 
-          # Update requirements.txt
+          echo "New major/minor detected ($new_major_minor). Updating requirements.txt..."
-          sed -i "s/runpod==.*/runpod==$new_version/" ./builder/requirements.txt
+
+          # Update requirements.txt, preserving the existing constraint type (~= or ==)
+          sed -i "s/runpod[~=][^ ]*/runpod~=$new_version/" ./builder/requirements.txt
+          echo "requirements.txt has been updated."
 
       - name: Create Pull Request
         uses: peter-evans/create-pull-request@v3
         with:
           token: ${{ secrets.GITHUB_TOKEN }}
-          commit-message: Update package version
+          commit-message: Update runpod package version
           title: Update runpod package version
           body: The package version has been updated to ${{ env.NEW_VERSION_ENV }}
           branch: runpod-package-update

.github/workflows/dev.yml

@@ -0,0 +1,60 @@
+name: Development
+
+on:
+  pull_request:
+    branches:
+      - "**"
+
+permissions:
+  contents: read
+
+jobs:
+  dev:
+    runs-on: [blacksmith-8vcpu-ubuntu-2204, linux]
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+
+      - name: Clear space to remove unused folders
+        run: |
+          rm -rf /usr/share/dotnet
+          rm -rf /opt/ghc
+          rm -rf "/usr/local/share/boost"
+          rm -rf "$AGENT_TOOLSDIRECTORY"
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: blacksmith docker layer cache
+        uses: useblacksmith/build-push-action@v1
+        with:
+          setup-only: true
+
+      - name: Set environment variables
+        run: |
+          echo "DOCKERHUB_REPO=${{ vars.DOCKERHUB_REPO || 'runpod' }}" >> $GITHUB_ENV
+          echo "DOCKERHUB_IMG=${{ vars.DOCKERHUB_IMG || 'worker-v1-vllm' }}" >> $GITHUB_ENV
+          echo "HUGGINGFACE_ACCESS_TOKEN=${{ secrets.HUGGINGFACE_ACCESS_TOKEN }}" >> $GITHUB_ENV
+          # Convert branch name to safe docker tag (replace / with -)
+          BRANCH_NAME="${GITHUB_REF##refs/heads/}"
+          SAFE_BRANCH_NAME=$(echo "$BRANCH_NAME" | sed 's/[^a-zA-Z0-9._-]/-/g' | sed 's/--*/-/g')
+          echo "RELEASE_VERSION=dev-${SAFE_BRANCH_NAME}" >> $GITHUB_ENV
+
+      - name: Build and push the images to Docker Hub
+        uses: docker/bake-action@v2
+        with:
+          push: true
+          set: |
+            *.args.DOCKERHUB_REPO=${{ env.DOCKERHUB_REPO }}
+            *.args.DOCKERHUB_IMG=${{ env.DOCKERHUB_IMG }}
+            *.args.RELEASE_VERSION=${{ env.RELEASE_VERSION }}
+            *.args.HUGGINGFACE_ACCESS_TOKEN=${{ env.HUGGINGFACE_ACCESS_TOKEN }}

.github/workflows/release.yml

@@ -0,0 +1,86 @@
+name: Release
+
+on:
+  push:
+    tags:
+      - "v[0-9]+.[0-9]+.[0-9]+*" # Trigger on version tags like v1.0.0, v2.1.0, etc.
+  workflow_dispatch:
+    inputs:
+      version:
+        description: "Version to release (e.g., v2.8.0)"
+        required: true
+        type: string
+
+permissions:
+  contents: write # Required for creating GitHub releases
+
+jobs:
+  release:
+    runs-on: [blacksmith-8vcpu-ubuntu-2204, linux]
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+
+      - name: Clear space to remove unused folders
+        run: |
+          rm -rf /usr/share/dotnet
+          rm -rf /opt/ghc
+          rm -rf "/usr/local/share/boost"
+          rm -rf "$AGENT_TOOLSDIRECTORY"
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: blacksmith docker layer cache
+        uses: useblacksmith/build-push-action@v1
+        with:
+          setup-only: true
+
+      - name: Set environment variables
+        run: |
+          echo "DOCKERHUB_REPO=${{ vars.DOCKERHUB_REPO || 'runpod' }}" >> $GITHUB_ENV
+          echo "DOCKERHUB_IMG=${{ vars.DOCKERHUB_IMG || 'worker-v1-vllm' }}" >> $GITHUB_ENV
+          echo "HUGGINGFACE_ACCESS_TOKEN=${{ secrets.HUGGINGFACE_ACCESS_TOKEN }}" >> $GITHUB_ENV
+
+          # Determine version based on trigger type
+          if [[ "${{ github.event_name }}" == "workflow_dispatch" ]]; then
+            # Manual trigger: use input version
+            VERSION="${{ github.event.inputs.version }}"
+            echo "RELEASE_VERSION=${VERSION}" >> $GITHUB_ENV
+            echo "IS_MANUAL_RELEASE=true" >> $GITHUB_ENV
+          else
+            # Tag trigger: use tag name (remove refs/tags/ prefix)
+            VERSION=${GITHUB_REF#refs/tags/}
+            echo "RELEASE_VERSION=${VERSION}" >> $GITHUB_ENV
+            echo "IS_MANUAL_RELEASE=false" >> $GITHUB_ENV
+          fi
+
+      - name: Build and push the images to Docker Hub
+        uses: docker/bake-action@v2
+        with:
+          push: true
+          set: |
+            *.args.DOCKERHUB_REPO=${{ env.DOCKERHUB_REPO }}
+            *.args.DOCKERHUB_IMG=${{ env.DOCKERHUB_IMG }}
+            *.args.RELEASE_VERSION=${{ env.RELEASE_VERSION }}
+            *.args.HUGGINGFACE_ACCESS_TOKEN=${{ env.HUGGINGFACE_ACCESS_TOKEN }}
+
+      - name: Release Summary
+        run: |
+          echo "🚀 Release completed!"
+          echo "Version: ${{ env.RELEASE_VERSION }}"
+          echo "Docker Image: ${{ env.DOCKERHUB_REPO }}/${{ env.DOCKERHUB_IMG }}:${{ env.RELEASE_VERSION }}"
+          if [[ "${{ github.event_name }}" == "workflow_dispatch" ]]; then
+            echo "Trigger: Manual workflow dispatch"
+          else
+            echo "Trigger: GitHub release (tag: ${{ github.ref_name }})"
+          fi

.gitignore

@@ -4,3 +4,4 @@ runpod.toml
 .env
 test/*
 vllm-base/vllm-*
+.DS_Store

.gitmodules

@@ -1,3 +0,0 @@
-[submodule "vllm-base-image/vllm"]
-	path = vllm-base-image/vllm
-	url = https://github.com/runpod/vllm-fork-for-sls-worker.git

.runpod/README.md

@@ -0,0 +1,261 @@
+![vLLM worker banner](https://cpjrphpz3t5wbwfe.public.blob.vercel-storage.com/worker-vllm_banner.jpeg)
+
+Run LLMs using [vLLM](https://docs.vllm.ai) with an OpenAI-compatible API
+
+---
+
+[![RunPod](https://api.runpod.io/badge/runpod-workers/worker-vllm)](https://www.runpod.io/console/hub/runpod-workers/worker-vllm)
+
+---
+
+## Endpoint Configuration
+
+All behaviour is controlled through environment variables:
+
+| Environment Variable                | Description                                       | Default             | Options                                                            |
+| ----------------------------------- | ------------------------------------------------- | ------------------- | ------------------------------------------------------------------ |
+| `MODEL_NAME`                        | Path of the model weights                         | "facebook/opt-125m" | Local folder or Hugging Face repo ID                               |
+| `HF_TOKEN`                          | HuggingFace access token for gated/private models |                     | Your HuggingFace access token                                      |
+| `MAX_MODEL_LEN`                     | Model's maximum context length                    |                     | Integer (e.g., 4096)                                               |
+| `QUANTIZATION`                      | Quantization method                               |                     | "awq", "gptq", "squeezellm", "bitsandbytes"                        |
+| `TENSOR_PARALLEL_SIZE`              | Number of GPUs                                    | 1                   | Integer                                                            |
+| `GPU_MEMORY_UTILIZATION`            | Fraction of GPU memory to use                     | 0.95                | Float between 0.0 and 1.0                                          |
+| `MAX_NUM_SEQS`                      | Maximum number of sequences per iteration         | 256                 | Integer                                                            |
+| `CUSTOM_CHAT_TEMPLATE`              | Custom chat template override                     |                     | Jinja2 template string                                             |
+| `ENABLE_AUTO_TOOL_CHOICE`           | Enable automatic tool selection                   | false               | boolean (true or false)                                            |
+| `TOOL_CALL_PARSER`                  | Parser for tool calls                             |                     | "mistral", "hermes", "llama3_json", "granite", "deepseek_v3", etc. |
+| `REASONING_PARSER`                  | Parser for reasoning-capable models               |                     | "deepseek_r1", "qwen3", "granite", "hunyuan_a13b"                  |
+| `OPENAI_SERVED_MODEL_NAME_OVERRIDE` | Override served model name in API                 |                     | String                                                             |
+| `MAX_CONCURRENCY`                   | Maximum concurrent requests                       | 300                 | Integer                                                            |
+
+For complete configuration options, see the [full configuration documentation](https://github.com/runpod-workers/worker-vllm/blob/main/docs/configuration.md).
+
+## API Usage
+
+This worker supports two API formats: **RunPod native** and **OpenAI-compatible**.
+
+### RunPod Native API
+
+For testing directly in the RunPod UI, use these examples in your endpoint's request tab.
+
+#### Chat Completions
+
+```json
+{
+  "input": {
+    "messages": [
+      { "role": "system", "content": "You are a helpful assistant." },
+      { "role": "user", "content": "What is the capital of France?" }
+    ],
+    "sampling_params": {
+      "max_tokens": 100,
+      "temperature": 0.7
+    }
+  }
+}
+```
+
+#### Chat Completions (Streaming)
+
+```json
+{
+  "input": {
+    "messages": [
+      { "role": "user", "content": "Write a short story about a robot." }
+    ],
+    "sampling_params": {
+      "max_tokens": 500,
+      "temperature": 0.8
+    },
+    "stream": true
+  }
+}
+```
+
+#### Text Generation
+
+For direct text generation without chat format:
+
+```json
+{
+  "input": {
+    "prompt": "The capital of France is",
+    "sampling_params": {
+      "max_tokens": 64,
+      "temperature": 0.0
+    }
+  }
+}
+```
+
+#### List Models
+
+```json
+{
+  "input": {
+    "openai_route": "/v1/models"
+  }
+}
+```
+
+---
+
+### OpenAI-Compatible API
+
+For external clients and SDKs, use the `/openai/v1` path prefix with your RunPod API key.
+
+#### Chat Completions
+
+**Path:** `/openai/v1/chat/completions`
+
+```json
+{
+  "model": "meta-llama/Llama-2-7b-chat-hf",
+  "messages": [
+    { "role": "system", "content": "You are a helpful assistant." },
+    { "role": "user", "content": "What is the capital of France?" }
+  ],
+  "max_tokens": 100,
+  "temperature": 0.7
+}
+```
+
+#### Chat Completions (Streaming)
+
+```json
+{
+  "model": "meta-llama/Llama-2-7b-chat-hf",
+  "messages": [
+    { "role": "user", "content": "Write a short story about a robot." }
+  ],
+  "max_tokens": 500,
+  "temperature": 0.8,
+  "stream": true
+}
+```
+
+#### Text Completions
+
+**Path:** `/openai/v1/completions`
+
+```json
+{
+  "model": "meta-llama/Llama-2-7b-chat-hf",
+  "prompt": "The capital of France is",
+  "max_tokens": 100,
+  "temperature": 0.7
+}
+```
+
+#### List Models
+
+**Path:** `/openai/v1/models`
+
+```json
+{}
+```
+
+#### Response Format
+
+Both APIs return the same response format:
+
+```json
+{
+  "choices": [
+    {
+      "index": 0,
+      "message": { "role": "assistant", "content": "Paris." },
+      "finish_reason": "stop"
+    }
+  ],
+  "usage": { "prompt_tokens": 9, "completion_tokens": 1, "total_tokens": 10 }
+}
+```
+
+---
+
+## Usage
+
+Below are minimal `python` snippets so you can copy-paste to get started quickly.
+
+> Replace `<ENDPOINT_ID>` with your endpoint ID and `<API_KEY>` with a [RunPod API key](https://docs.runpod.io/get-started/api-keys).
+
+### OpenAI compatible API
+
+Minimal Python example using the official `openai` SDK:
+
+```python
+from openai import OpenAI
+import os
+
+# Initialize the OpenAI Client with your RunPod API Key and Endpoint URL
+client = OpenAI(
+    api_key=os.getenv("RUNPOD_API_KEY"),
+    base_url=f"https://api.runpod.ai/v2/<ENDPOINT_ID>/openai/v1",
+)
+```
+
+`Chat Completions (Non-Streaming)`
+
+```python
+response = client.chat.completions.create(
+    model="meta-llama/Llama-2-7b-chat-hf",
+    messages=[{"role": "user", "content": "Explain quantum computing in simple terms"}],
+    temperature=0,
+    max_tokens=100,
+)
+print(f"Response: {response.choices[0].message.content}")
+```
+
+`Chat Completions (Streaming)`
+
+```python
+response_stream = client.chat.completions.create(
+    model="meta-llama/Llama-2-7b-chat-hf",
+    messages=[{"role": "user", "content": "Explain quantum computing in simple terms"}],
+    temperature=0,
+    max_tokens=100,
+    stream=True
+)
+for response in response_stream:
+    print(response.choices[0].delta.content or "", end="", flush=True)
+```
+
+### RunPod Native API
+
+```python
+import requests
+
+response = requests.post(
+    "https://api.runpod.ai/v2/<ENDPOINT_ID>/run",
+    headers={"Authorization": "Bearer <API_KEY>"},
+    json={
+        "input": {
+            "messages": [
+                {"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": "Explain quantum computing in simple terms"}
+            ],
+            "sampling_params": {
+                "temperature": 0.7,
+                "max_tokens": 150
+            }
+        }
+    }
+)
+
+result = response.json()
+print(result["output"])
+```
+
+## Compatibility
+
+For supported models, see the [vLLM supported models documentation](https://docs.vllm.ai/en/latest/models/supported_models.html).
+
+Anything not recognized by worker-vllm is forwarded to vLLM's engine, so advanced options in the vLLM docs (guided generation, LoRA, speculative decoding, etc.) also work.
+
+## Documentation
+
+- **[🚀 Deployment Guide](https://docs.runpod.io/serverless/vllm/get-started)** - Step-by-step setup
+- **[📖 Configuration Reference](https://github.com/runpod-workers/worker-vllm/blob/main/docs/configuration.md)** - All environment variables
+- **[🏗️ Advanced Deployment](https://github.com/runpod-workers/worker-vllm/blob/main/docs/deployment.md)** - Custom builds and strategies
+- **[🔧 Development Guide](https://github.com/runpod-workers/worker-vllm/blob/main/docs/conventions.md)** - Architecture and patterns

.runpod/tests.json

@@ -0,0 +1,43 @@
+{
+  "tests": [
+    {
+      "name": "basic_inference_test",
+      "input": {
+        "prompt": "Write a short poem about artificial intelligence."
+      },
+      "timeout": 30000
+    },
+    {
+      "name": "openai_messages_test",
+      "input": {
+        "openai_route": "/v1/chat/completions",
+        "openai_input": {
+          "messages": [
+            {
+              "role": "system",
+              "content": "You are a helpful assistant that writes concise responses."
+            },
+            {
+              "role": "user",
+              "content": "Explain what a neural network is in one sentence."
+            }
+          ],
+          "max_tokens": 200,
+          "temperature": 0.1
+        }
+      },
+      "timeout": 30000
+    }
+  ],
+  "config": {
+    "gpuTypeId": "NVIDIA GeForce RTX 4090",
+    "gpuCount": 1,
+    "env": [
+      {
+        "key": "MODEL_NAME",
+        "value": "HuggingFaceTB/SmolLM2-135M-Instruct"
+      }
+    ],
+    "allowedCudaVersions": ["12.9", "12.8", "12.7", "12.6", "12.5"]
+  }
+}

Dockerfile

@@ -1,16 +1,20 @@
-ARG WORKER_CUDA_VERSION=11.8.0
+FROM nvidia/cuda:12.1.0-base-ubuntu22.04 
-ARG BASE_IMAGE_VERSION=1.0.0
-FROM runpod/worker-vllm:base-${BASE_IMAGE_VERSION}-cuda${WORKER_CUDA_VERSION} AS vllm-base
 
 RUN apt-get update -y \
     && apt-get install -y python3-pip
 
+RUN ldconfig /usr/local/cuda-12.1/compat/
+
 # Install Python dependencies
 COPY builder/requirements.txt /requirements.txt
 RUN --mount=type=cache,target=/root/.cache/pip \
     python3 -m pip install --upgrade pip && \
     python3 -m pip install --upgrade -r /requirements.txt
 
+# Install vLLM (switching back to pip installs since issues that required building fork are fixed and space optimization is not as important since caching) and FlashInfer 
+RUN python3 -m pip install vllm==0.11.0 && \
+    python3 -m pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.3
+
 # Setup for Option 2: Building the Image with the Model included
 ARG MODEL_NAME=""
 ARG TOKENIZER_NAME=""
@@ -28,23 +32,19 @@ ENV MODEL_NAME=$MODEL_NAME \
     HF_DATASETS_CACHE="${BASE_PATH}/huggingface-cache/datasets" \
     HUGGINGFACE_HUB_CACHE="${BASE_PATH}/huggingface-cache/hub" \
     HF_HOME="${BASE_PATH}/huggingface-cache/hub" \
-    HF_HUB_ENABLE_HF_TRANSFER=1 
+    HF_HUB_ENABLE_HF_TRANSFER=0 
 
 ENV PYTHONPATH="/:/vllm-workspace"
 
-COPY src/download_model.py /download_model.py
+
+COPY src /src
 RUN --mount=type=secret,id=HF_TOKEN,required=false \
     if [ -f /run/secrets/HF_TOKEN ]; then \
-        export HF_TOKEN=$(cat /run/secrets/HF_TOKEN); \
+    export HF_TOKEN=$(cat /run/secrets/HF_TOKEN); \
     fi && \
     if [ -n "$MODEL_NAME" ]; then \
-        python3 /download_model.py; \
+    python3 /src/download_model.py; \
     fi
 
-# Add source files
-COPY src /src
-# Remove download_model.py
-RUN rm /download_model.py
-
 # Start the handler
 CMD ["python3", "/src/handler.py"]

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2023 runpod-workers
+Copyright (c) 2025 Runpod
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.md

@@ -1,374 +1,253 @@
 <div align="center">
 
 # OpenAI-Compatible vLLM Serverless Endpoint Worker
-Deploy OpenAI-Compatible Blazing-Fast LLM Endpoints powered by the [vLLM](https://github.com/vllm-project/vllm) Inference Engine on RunPod Serverless with just a few clicks.
-<!-- 
-![vLLM Version](https://img.shields.io/badge/dynamic/yaml?url=https%3A%2F%2Fraw.githubusercontent.com%2Frunpod-workers%2Fworker-vllm%2Fmain%2Fvllm-base-image%2Fvllm-metadata.yml&query=%24.version&style=for-the-badge&logo=data%3Aimage%2Fsvg%2Bxml%3Bbase64%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%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%2BPHBhdGggc3R5bGU9Im9wYWNpdHk6MC45ODQiIGZpbGw9IiNmY2I3MWQiIGQ9Ik0gMjIuNSwxMi41IEMgMjEuNTA0NiwyNC45ODkgMjEuMTcxMywzNy42NTU3IDIxLjUsNTAuNUMgMjEuNjcxNiw1MS40OTE2IDIxLjMzODIsNTIuMTU4MyAyMC41LDUyLjVDIDEzLjAzMTEsMzkuMjI4NyA2LjM2NDQxLDI1LjU2MjEgMC41LDExLjVDIDguMDE5MDUsMTEuMTc1IDE1LjM1MjQsMTEuNTA4NCAyMi41LDEyLjUgWiIvPjwvZz4KPGc%2BPHBhdGggc3R5bGU9Im9wYWNpdHk6MC4wMiIgZmlsbD0iI2Q3ZGZlOCIgZD0iTSAyMi41LDEyLjUgQyAyMy4xNjY3LDIxLjUgMjMuODMzMywzMC41IDI0LjUsMzkuNUMgMjMuMjIyOSw0My4xODg5IDIyLjIyMjksNDYuODU1NSAyMS41LDUwLjVDIDIxLjE3MTMsMzcuNjU1NyAyMS41MDQ2LDI0Ljk4OSAyMi41LDEyLjUgWiIvPjwvZz4KPGc%2BPHBhdGggc3R5bGU9Im9wYWNpdHk6MC43NTMiIGZpbGw9IiNjZmQ2ZGQiIGQ9Ik0gNTEuNSwwLjUgQyA1Mi42MTI5LDEuOTQ2MzkgNTIuNzc5NiwzLjYxMzA1IDUyLDUuNUMgNDcuODAzNiwyMi4yODg3IDQzLjMwMzYsMzguOTU1MyAzOC41LDU1LjVDIDMyLjUsNTUuNSAyNi41LDU1LjUgMjAuNSw1NS41QyAyMC44MzMzLDU0LjgzMzMgMjEuMTY2Nyw1NC4xNjY3IDIxLjUsNTMuNUMgMjYuODMzMyw1My41IDMyLjE2NjcsNTMuNSAzNy41LDUzLjVDIDQxLjkxNTYsMzUuNzUwNSA0Ni41ODIyLDE4LjA4MzggNTEuNSwwLjUgWiIvPjwvZz4KPC9zdmc%2BCg%3D%3D&label=STABLE%20vLLM%20Version&link=https%3A%2F%2Fgithub.com%2Fvllm-project%2Fvllm)
-![Worker Version](https://img.shields.io/github/v/tag/runpod-workers/worker-vllm?style=for-the-badge&logo=data%3Aimage%2Fsvg%2Bxml%3Bbase64%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&logoColor=%23ffffff&label=STABLE%20Worker%20Version&color=%23673ab7)
-![vLLM Version](https://img.shields.io/badge/dynamic/yaml?url=https%3A%2F%2Fraw.githubusercontent.com%2Frunpod-workers%2Fworker-vllm%2Fmain%2Fvllm-base-image%2Fvllm-metadata.yml&query=%24.dev_version&style=for-the-badge&logo=data%3Aimage%2Fsvg%2Bxml%3Bbase64%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%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%2BPHBhdGggc3R5bGU9Im9wYWNpdHk6MC45ODQiIGZpbGw9IiNmY2I3MWQiIGQ9Ik0gMjIuNSwxMi41IEMgMjEuNTA0NiwyNC45ODkgMjEuMTcxMywzNy42NTU3IDIxLjUsNTAuNUMgMjEuNjcxNiw1MS40OTE2IDIxLjMzODIsNTIuMTU4MyAyMC41LDUyLjVDIDEzLjAzMTEsMzkuMjI4NyA2LjM2NDQxLDI1LjU2MjEgMC41LDExLjVDIDguMDE5MDUsMTEuMTc1IDE1LjM1MjQsMTEuNTA4NCAyMi41LDEyLjUgWiIvPjwvZz4KPGc%2BPHBhdGggc3R5bGU9Im9wYWNpdHk6MC4wMiIgZmlsbD0iI2Q3ZGZlOCIgZD0iTSAyMi41LDEyLjUgQyAyMy4xNjY3LDIxLjUgMjMuODMzMywzMC41IDI0LjUsMzkuNUMgMjMuMjIyOSw0My4xODg5IDIyLjIyMjksNDYuODU1NSAyMS41LDUwLjVDIDIxLjE3MTMsMzcuNjU1NyAyMS41MDQ2LDI0Ljk4OSAyMi41LDEyLjUgWiIvPjwvZz4KPGc%2BPHBhdGggc3R5bGU9Im9wYWNpdHk6MC43NTMiIGZpbGw9IiNjZmQ2ZGQiIGQ9Ik0gNTEuNSwwLjUgQyA1Mi42MTI5LDEuOTQ2MzkgNTIuNzc5NiwzLjYxMzA1IDUyLDUuNUMgNDcuODAzNiwyMi4yODg3IDQzLjMwMzYsMzguOTU1MyAzOC41LDU1LjVDIDMyLjUsNTUuNSAyNi41LDU1LjUgMjAuNSw1NS41QyAyMC44MzMzLDU0LjgzMzMgMjEuMTY2Nyw1NC4xNjY3IDIxLjUsNTMuNUMgMjYuODMzMyw1My41IDMyLjE2NjcsNTMuNSAzNy41LDUzLjVDIDQxLjkxNTYsMzUuNzUwNSA0Ni41ODIyLDE4LjA4MzggNTEuNSwwLjUgWiIvPjwvZz4KPC9zdmc%2BCg%3D%3D&label=DEV%20vLLM%20Version%20&link=https%3A%2F%2Fgithub.com%2Fvllm-project%2Fvllm)\
-![Docker Pulls](https://img.shields.io/docker/pulls/runpod/worker-vllm?style=for-the-badge&logo=docker&label=Docker%20Pulls&link=https%3A%2F%2Fhub.docker.com%2Frepository%2Fdocker%2Frunpod%2Fworker-vllm%2Fgeneral) -->
-<!-- 
-![Docker Automatic Build](https://img.shields.io/github/actions/workflow/status/runpod-workers/worker-vllm/docker-build-release.yml?style=flat&label=BUILD) -->
 
+Deploy OpenAI-Compatible Blazing-Fast LLM Endpoints powered by the [vLLM](https://github.com/vllm-project/vllm) Inference Engine on RunPod Serverless with just a few clicks.
 
 </div>
 
-# News:
-
-### 1. UI for Deploying vLLM Worker on RunPod console:
-![Demo of Deploying vLLM Worker on RunPod console with new UI](media/ui_demo.gif)
-
-### 2. Worker vLLM `1.0.0` with vLLM `0.4.2` now available under `stable` tags
-Update 1.0.0 is now available, use the image tag `runpod/worker-vllm:stable-cuda12.1.0` or `runpod/worker-vllm:stable-cuda11.8.0`.
-
-### 3. OpenAI-Compatible [Embedding Worker](https://github.com/runpod-workers/worker-infinity-embedding) Released
-Deploy your own OpenAI-compatible Serverless Endpoint on RunPod with multiple embedding models and fast inference for RAG and more! 
-
-
-
-### 4. Caching Accross RunPod Machines
-Worker vLLM is now cached on all RunPod machines, resulting in near-instant deployment! Previously, downloading and extracting the image took 3-5 minutes on average.
-
-
 ## Table of Contents
+
 - [Setting up the Serverless Worker](#setting-up-the-serverless-worker)
-  - [Option 1: Deploy Any Model Using Pre-Built Docker Image **[RECOMMENDED]**](#option-1-deploy-any-model-using-pre-built-docker-image-recommended)
+  - [Option 1: Deploy Any Model Using Pre-Built Docker Image [Recommended]](#option-1-deploy-any-model-using-pre-built-docker-image-recommended)
-    - [Prerequisites](#prerequisites)
+    - [Configuration](#configuration)
-    - [Environment Variables](#environment-variables)
-      - [LLM Settings](#llm-settings)
-      - [Tokenizer Settings](#tokenizer-settings)
-      - [Tensor Parallelism (Multi-GPU) Settings](#tensor-parallelism-multi-gpu-settings)
-      - [System Settings](#system-settings)
-      - [Streaming Batch Size](#streaming-batch-size)
-      - [OpenAI Settings](#openai-settings)
-      - [Serverless Settings](#serverless-settings)
   - [Option 2: Build Docker Image with Model Inside](#option-2-build-docker-image-with-model-inside)
-    - [Prerequisites](#prerequisites-1)
+    - [Prerequisites](#prerequisites)
     - [Arguments](#arguments)
     - [Example: Building an image with OpenChat-3.5](#example-building-an-image-with-openchat-35)
       - [(Optional) Including Huggingface Token](#optional-including-huggingface-token)
   - [Compatible Model Architectures](#compatible-model-architectures)
 - [Usage: OpenAI Compatibility](#usage-openai-compatibility)
-    - [Modifying your OpenAI Codebase to use your deployed vLLM Worker](#modifying-your-openai-codebase-to-use-your-deployed-vllm-worker)
+  - [Modifying your OpenAI Codebase to use your deployed vLLM Worker](#modifying-your-openai-codebase-to-use-your-deployed-vllm-worker)
-    - [OpenAI Request Input Parameters](#openai-request-input-parameters)
+  - [OpenAI Request Input Parameters](#openai-request-input-parameters)
-      - [Chat Completions](#chat-completions)
+  - [Chat Completions [RECOMMENDED]](#chat-completions-recommended)
-      - [Completions](#completions)
+  - [Examples: Using your RunPod endpoint with OpenAI](#examples-using-your-runpod-endpoint-with-openai)
-    - [Examples: Using your RunPod endpoint with OpenAI](#examples-using-your-runpod-endpoint-with-openai)
+    - [Chat Completions](#chat-completions)
-- [Usage: standard](#non-openai-usage)
+    - [Getting a list of names for available models](#getting-a-list-of-names-for-available-models)
-  - [Input Request Parameters](#input-request-parameters)
+- [Usage: Standard (Non-OpenAI)](#usage-standard-non-openai)
+  - [Request Input Parameters](#request-input-parameters)
+  - [Sampling Parameters](#sampling-parameters)
     - [Text Input Formats](#text-input-formats)
-    - [Sampling Parameters](#sampling-parameters)
 
 # Setting up the Serverless Worker
 
-### Option 1: Deploy Any Model Using Pre-Built Docker Image [Recommended]
+## Option 1: Deploy Any Model Using Pre-Built Docker Image [Recommended]
 
-> [!NOTE]
+**🚀 Deploy Guide**: Follow our [step-by-step deployment guide](https://docs.runpod.io/serverless/vllm/get-started) to deploy using the RunPod Console.
-> You can now deploy from the dedicated UI on the RunPod console with all of the settings and choices listed. 
-> Try now by accessing in Explore or Serverless pages on the RunPod console!
 
+**📦 Docker Image**: `runpod/worker-v1-vllm:<version>`
 
-We now offer a pre-built Docker Image for the vLLM Worker that you can configure entirely with Environment Variables when creating the RunPod Serverless Endpoint:
+- **Available Versions**: See [GitHub Releases](https://github.com/runpod-workers/worker-vllm/releases)
+- **CUDA Compatibility**: Requires CUDA >= 12.1
 
----
+### Configuration
 
-## RunPod Worker Images
+Configure worker-vllm using environment variables:
 
-Below is a summary of the available RunPod Worker images, categorized by image stability and CUDA version compatibility.
+| Environment Variable                | Description                                       | Default             | Options                                                            |
+| ----------------------------------- | ------------------------------------------------- | ------------------- | ------------------------------------------------------------------ |
+| `MODEL_NAME`                        | Path of the model weights                         | "facebook/opt-125m" | Local folder or Hugging Face repo ID                               |
+| `HF_TOKEN`                          | HuggingFace access token for gated/private models |                     | Your HuggingFace access token                                      |
+| `MAX_MODEL_LEN`                     | Model's maximum context length                    |                     | Integer (e.g., 4096)                                               |
+| `QUANTIZATION`                      | Quantization method                               |                     | "awq", "gptq", "squeezellm", "bitsandbytes"                        |
+| `TENSOR_PARALLEL_SIZE`              | Number of GPUs                                    | 1                   | Integer                                                            |
+| `GPU_MEMORY_UTILIZATION`            | Fraction of GPU memory to use                     | 0.95                | Float between 0.0 and 1.0                                          |
+| `MAX_NUM_SEQS`                      | Maximum number of sequences per iteration         | 256                 | Integer                                                            |
+| `CUSTOM_CHAT_TEMPLATE`              | Custom chat template override                     |                     | Jinja2 template string                                             |
+| `ENABLE_AUTO_TOOL_CHOICE`           | Enable automatic tool selection                   | false               | boolean (true or false)                                            |
+| `TOOL_CALL_PARSER`                  | Parser for tool calls                             |                     | "mistral", "hermes", "llama3_json", "granite", "deepseek_v3", etc. |
+| `OPENAI_SERVED_MODEL_NAME_OVERRIDE` | Override served model name in API                 |                     | String                                                             |
+| `MAX_CONCURRENCY`                   | Maximum concurrent requests                       | 30                  | Integer                                                            |
 
-| CUDA Version | Stable Image Tag                  | Development Image Tag             | Note                                                        |
+For the complete list of all available environment variables, examples, and detailed descriptions: **[Configuration](docs/configuration.md)**
-|--------------|-----------------------------------|-----------------------------------|----------------------------------------------------------------------|
-| 11.8.0       | `runpod/worker-vllm:stable-cuda11.8.0`        | `runpod/worker-vllm:dev-cuda11.8.0`   | Available on all RunPod Workers without additional selection needed. |
-| 12.1.0       | `runpod/worker-vllm:stable-cuda12.1.0` | `runpod/worker-vllm:dev-cuda12.1.0` | When creating an Endpoint, select CUDA Version 12.3, 12.2 and 12.1 in the filter. |
 
+## Option 2: Build Docker Image with Model Inside
 
-
----
-
-#### Prerequisites
-- RunPod Account
-
-#### Environment Variables/Settings
-> Note:  `0` is equivalent to `False` and `1` is equivalent to `True` for boolean values.
-
-| Name                                | Default              | Type/Choices                              | Description |
-|-------------------------------------|----------------------|-------------------------------------------|-------------|
-**LLM Settings**
-| `MODEL_NAME`**\***                        | -                    | `str`                                         | Hugging Face Model Repository (e.g., `openchat/openchat-3.5-1210`). |
-| `MODEL_REVISION`                    | `None`               | `str`                                         |Model revision(branch) to load. |
-| `MAX_MODEL_LEN`                  | Model's maximum      | `int`                                         |Maximum number of tokens for the engine to handle per request. |
-| `BASE_PATH`                         | `/runpod-volume`     | `str`                                         |Storage directory for Huggingface cache and model. Utilizes network storage if attached when pointed at `/runpod-volume`, which will have only one worker download the model once, which all workers will be able to load. If no network volume is present, creates a local directory within each worker. |
-| `LOAD_FORMAT`                       | `auto`               | `str`                                         |Format to load model in. |
-| `HF_TOKEN`                          | -                    | `str`                                         |Hugging Face token for private and gated models. |
-| `QUANTIZATION`                      | `None`               | `awq`, `squeezellm`, `gptq`              |Quantization of given model. The model must already be quantized. |
-| `TRUST_REMOTE_CODE`                 | `0`                  | boolean as `int`                                         |Trust remote code for Hugging Face models. Can help with Mixtral 8x7B, Quantized models, and unusual models/architectures.
-| `SEED`                              | `0`                  | `int`                                         |Sets random seed for operations. |
-| `KV_CACHE_DTYPE`                    | `auto`               | `auto`,  `fp8`                                         |Data type for kv cache storage. Uses `DTYPE` if set to `auto`. |
-| `DTYPE`                             | `auto`               | `auto`, `half`, `float16`, `bfloat16`, `float`, `float32` |Sets datatype/precision for model weights and activations. |
-**Tokenizer Settings**
-| `TOKENIZER_NAME`                    | `None`               | `str`                                         |Tokenizer repository to use a different tokenizer than the model's default. |
-| `TOKENIZER_REVISION`                | `None`               | `str`                                         |Tokenizer revision to load. |
-| `CUSTOM_CHAT_TEMPLATE`              | `None`               | `str` of single-line jinja template                                         |Custom chat jinja template. [More Info](https://huggingface.co/docs/transformers/chat_templating) |
-**System, GPU, and Tensor Parallelism(Multi-GPU) Settings**
-| `GPU_MEMORY_UTILIZATION`            | `0.95`               | `float`                                         |Sets GPU VRAM utilization. |
-| `MAX_PARALLEL_LOADING_WORKERS`      | `None`               | `int`                                         |Load model sequentially in multiple batches, to avoid RAM OOM when using tensor parallel and large models. |
-| `BLOCK_SIZE`                        | `16`                 | `8`, `16`, `32`                           |Token block size for contiguous chunks of tokens. |
-| `SWAP_SPACE`                        | `4`                  | `int`                                         |CPU swap space size (GiB) per GPU. |
-| `ENFORCE_EAGER`                     | `0`                  | boolean as `int`                                         |Always use eager-mode PyTorch. If False(`0`), will use eager mode and CUDA graph in hybrid for maximal performance and flexibility. |
-| `MAX_CONTEXT_LEN_TO_CAPTURE`        | `8192`               | `int`                                     |Maximum context length covered by CUDA graphs. When a sequence has context length larger than this, we fall back to eager mode.|
-| `DISABLE_CUSTOM_ALL_REDUCE`         | `0`                  | `int`                                         |Enables or disables custom all reduce. |
-**Streaming Batch Size Settings**:  
-| `DEFAULT_BATCH_SIZE`                | `50`                 | `int`                                         |Default and Maximum batch size for token streaming to reduce HTTP calls. |
-| `DEFAULT_MIN_BATCH_SIZE`            | `1`                  | `int`                                         |Batch size for the first request, which will be multiplied by the growth factor every subsequent request. |
-| `DEFAULT_BATCH_SIZE_GROWTH_FACTOR`  | `3`                  | `float`                                         |Growth factor for dynamic batch size. |
-The way this works is that the first request will have a batch size of `DEFAULT_MIN_BATCH_SIZE`, and each subsequent request will have a batch size of `previous_batch_size * DEFAULT_BATCH_SIZE_GROWTH_FACTOR`. This will continue until the batch size reaches `DEFAULT_BATCH_SIZE`. E.g. for the default values, the batch sizes will be `1, 3, 9, 27, 50, 50, 50, ...`. You can also specify this per request, with inputs `max_batch_size`, `min_batch_size`, and `batch_size_growth_factor`. This has nothing to do with vLLM's internal batching, but rather the number of tokens sent in each HTTP request from the worker |
-**OpenAI Settings**
-| `RAW_OPENAI_OUTPUT`                 | `1`                  | boolean as `int`                                         |Enables raw OpenAI SSE format string output when streaming.  **Required** to be enabled (which it is by default) for OpenAI compatibility. |
-| `OPENAI_SERVED_MODEL_NAME_OVERRIDE` | `None`               | `str`                                         |Overrides the name of the served model from model repo/path to specified name, which you will then be able to use the value for the `model` parameter when making OpenAI requests |
-| `OPENAI_RESPONSE_ROLE`              | `assistant`          | `str`                       |Role of the LLM's Response in OpenAI Chat Completions. |
-**Serverless Settings**
-| `MAX_CONCURRENCY`                   | `300`                | `int`                                         |Max concurrent requests per worker. vLLM has an internal queue, so you don't have to worry about limiting by VRAM, this is for improving scaling/load balancing efficiency |
-| `DISABLE_LOG_STATS`                 | `1`                  | boolean as `int`                                         |Enables or disables vLLM stats logging. |
-| `DISABLE_LOG_REQUESTS`              | `1`                  | boolean as `int`                                         |Enables or disables vLLM request logging. |
-
-> [!TIP]
-> If you are facing issues when using Mixtral 8x7B, Quantized models, or handling unusual models/architectures, try setting `TRUST_REMOTE_CODE` to `1`.
-
-
-### Option 2: Build Docker Image with Model Inside
 To build an image with the model baked in, you must specify the following docker arguments when building the image.
 
-#### Prerequisites
+### Prerequisites
-- RunPod Account
+
 - Docker
 
-#### Arguments:
+### Arguments
+
 - **Required**
   - `MODEL_NAME`
 - **Optional**
   - `MODEL_REVISION`: Model revision to load (default: `main`).
   - `BASE_PATH`: Storage directory where huggingface cache and model will be located. (default: `/runpod-volume`, which will utilize network storage if you attach it or create a local directory within the image if you don't. If your intention is to bake the model into the image, you should set this to something like `/models` to make sure there are no issues if you were to accidentally attach network storage.)
   - `QUANTIZATION`
-  - `WORKER_CUDA_VERSION`: `11.8.0` or `12.1.0` (default: `11.8.0` due to a small number of workers not having CUDA 12.1 support yet. `12.1.0` is recommended for optimal performance).
+  - `WORKER_CUDA_VERSION`: `12.1.0` (`12.1.0` is recommended for optimal performance).
   - `TOKENIZER_NAME`: Tokenizer repository if you would like to use a different tokenizer than the one that comes with the model. (default: `None`, which uses the model's tokenizer)
   - `TOKENIZER_REVISION`: Tokenizer revision to load (default: `main`).
 
 For the remaining settings, you may apply them as environment variables when running the container. Supported environment variables are listed in the [Environment Variables](#environment-variables) section.
 
-#### Example: Building an image with OpenChat-3.5
+### Example: Building an image with OpenChat-3.5
+
 ```bash
-sudo docker build -t username/image:tag --build-arg MODEL_NAME="openchat/openchat_3.5" --build-arg BASE_PATH="/models" .
+docker build -t username/image:tag --build-arg MODEL_NAME="openchat/openchat_3.5" --build-arg BASE_PATH="/models" .
 ```
 
-##### (Optional) Including Huggingface Token
+### (Optional) Including Huggingface Token
+
 If the model you would like to deploy is private or gated, you will need to include it during build time as a Docker secret, which will protect it from being exposed in the image and on DockerHub.
+
 1. Enable Docker BuildKit (required for secrets).
+
 ```bash
 export DOCKER_BUILDKIT=1
 ```
+
 2. Export your Hugging Face token as an environment variable
+
 ```bash
 export HF_TOKEN="your_token_here"
 ```
+
 2. Add the token as a secret when building
+
 ```bash
 docker build -t username/image:tag --secret id=HF_TOKEN --build-arg MODEL_NAME="openchat/openchat_3.5" .
 ```
 
-## Compatible Model Architectures
+# Compatible Model Architectures
-Below are all supported model architectures (and examples of each) that you can deploy using the vLLM Worker. You can deploy **any model on HuggingFace**, as long as its base architecture is one of the following:
 
-- Aquila & Aquila2 (`BAAI/AquilaChat2-7B`, `BAAI/AquilaChat2-34B`, `BAAI/Aquila-7B`, `BAAI/AquilaChat-7B`, etc.)
+You can deploy **any model on Hugging Face** that is supported by vLLM. For the complete and up-to-date list of supported model architectures, see the [vLLM Supported Models documentation](https://docs.vllm.ai/en/latest/models/supported_models.html#list-of-text-only-language-models).
-- Baichuan & Baichuan2 (`baichuan-inc/Baichuan2-13B-Chat`, `baichuan-inc/Baichuan-7B`, etc.)
-- BLOOM (`bigscience/bloom`, `bigscience/bloomz`, etc.)
-- ChatGLM (`THUDM/chatglm2-6b`, `THUDM/chatglm3-6b`, etc.)
-- Command-R (`CohereForAI/c4ai-command-r-v01`, etc.)
-- DBRX (`databricks/dbrx-base`, `databricks/dbrx-instruct` etc.)
-- DeciLM (`Deci/DeciLM-7B`, `Deci/DeciLM-7B-instruct`, etc.)
-- Falcon (`tiiuae/falcon-7b`, `tiiuae/falcon-40b`, `tiiuae/falcon-rw-7b`, etc.)
-- Gemma (`google/gemma-2b`, `google/gemma-7b`, etc.)
-- GPT-2 (`gpt2`, `gpt2-xl`, etc.)
-- GPT BigCode (`bigcode/starcoder`, `bigcode/gpt_bigcode-santacoder`, etc.)
-- GPT-J (`EleutherAI/gpt-j-6b`, `nomic-ai/gpt4all-j`, etc.)
-- GPT-NeoX (`EleutherAI/gpt-neox-20b`, `databricks/dolly-v2-12b`, `stabilityai/stablelm-tuned-alpha-7b`, etc.)
-- InternLM (`internlm/internlm-7b`, `internlm/internlm-chat-7b`, etc.)
-- InternLM2 (`internlm/internlm2-7b`, `internlm/internlm2-chat-7b`, etc.)
-- Jais (`core42/jais-13b`, `core42/jais-13b-chat`, `core42/jais-30b-v3`, `core42/jais-30b-chat-v3`, etc.)
-- LLaMA, Llama 2, and Meta Llama 3 (`meta-llama/Meta-Llama-3-8B-Instruct`, `meta-llama/Meta-Llama-3-70B-Instruct`, `meta-llama/Llama-2-70b-hf`, `lmsys/vicuna-13b-v1.3`, `young-geng/koala`, `openlm-research/open_llama_13b`, etc.)
-- MiniCPM (`openbmb/MiniCPM-2B-sft-bf16`, `openbmb/MiniCPM-2B-dpo-bf16`, etc.)
-- Mistral (`mistralai/Mistral-7B-v0.1`, `mistralai/Mistral-7B-Instruct-v0.1`, etc.)
-- Mixtral (`mistralai/Mixtral-8x7B-v0.1`, `mistralai/Mixtral-8x7B-Instruct-v0.1`, `mistral-community/Mixtral-8x22B-v0.1`, etc.)
-- MPT (`mosaicml/mpt-7b`, `mosaicml/mpt-30b`, etc.)
-- OLMo (`allenai/OLMo-1B-hf`, `allenai/OLMo-7B-hf`, etc.)
-- OPT (`facebook/opt-66b`, `facebook/opt-iml-max-30b`, etc.)
-- Orion (`OrionStarAI/Orion-14B-Base`, `OrionStarAI/Orion-14B-Chat`, etc.)
-- Phi (`microsoft/phi-1_5`, `microsoft/phi-2`, etc.)
-- Phi-3 (`microsoft/Phi-3-mini-4k-instruct`, `microsoft/Phi-3-mini-128k-instruct`, etc.)
-- Qwen (`Qwen/Qwen-7B`, `Qwen/Qwen-7B-Chat`, etc.)
-- Qwen2 (`Qwen/Qwen1.5-7B`, `Qwen/Qwen1.5-7B-Chat`, etc.)
-- Qwen2MoE (`Qwen/Qwen1.5-MoE-A2.7B`, `Qwen/Qwen1.5-MoE-A2.7B-Chat`, etc.)
-- StableLM(`stabilityai/stablelm-3b-4e1t`, `stabilityai/stablelm-base-alpha-7b-v2`, etc.)
-- Starcoder2(`bigcode/starcoder2-3b`, `bigcode/starcoder2-7b`, `bigcode/starcoder2-15b`, etc.)
-- Xverse (`xverse/XVERSE-7B-Chat`, `xverse/XVERSE-13B-Chat`, `xverse/XVERSE-65B-Chat`, etc.)
-- Yi (`01-ai/Yi-6B`, `01-ai/Yi-34B`, etc.)
 
 # Usage: OpenAI Compatibility
-The vLLM Worker is fully compatible with OpenAI's API, and you can use it with any OpenAI Codebase by changing only 3 lines in total. The supported routes are <ins>Chat Completions</ins>, <ins>Completions</ins> and <ins>Models</ins> - with both streaming and non-streaming.
+
+The vLLM Worker is fully compatible with OpenAI's API, and you can use it with any OpenAI Codebase by changing only 3 lines in total. The supported routes are <ins>Chat Completions</ins> and <ins>Models</ins> - with both streaming and non-streaming.
 
 ## Modifying your OpenAI Codebase to use your deployed vLLM Worker
+
 **Python** (similar to Node.js, etc.):
+
 1. When initializing the OpenAI Client in your code, change the `api_key` to your RunPod API Key and the `base_url` to your RunPod Serverless Endpoint URL in the following format: `https://api.runpod.ai/v2/<YOUR ENDPOINT ID>/openai/v1`, filling in your deployed endpoint ID. For example, if your Endpoint ID is `abc1234`, the URL would be `https://api.runpod.ai/v2/abc1234/openai/v1`.
 
-    - Before:
+   - Before:
-    ```python
-    from openai import OpenAI
 
-    client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
+   ```python
-    ```
+   from openai import OpenAI
-    - After:
+
-    ```python
+   client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
-    from openai import OpenAI
+   ```
+
+   - After:
+
+   ```python
+   from openai import OpenAI
+
+   client = OpenAI(
+       api_key=os.environ.get("RUNPOD_API_KEY"),
+       base_url="https://api.runpod.ai/v2/<YOUR ENDPOINT ID>/openai/v1",
+   )
+   ```
 
-    client = OpenAI(
-        api_key=os.environ.get("RUNPOD_API_KEY"),
-        base_url="https://api.runpod.ai/v2/<YOUR ENDPOINT ID>/openai/v1",
-    )
-    ```
 2. Change the `model` parameter to your deployed model's name whenever using Completions or Chat Completions.
-    - Before:
+   - Before:
-    ```python
+   ```python
-    response = client.chat.completions.create(
+   response = client.chat.completions.create(
-        model="gpt-3.5-turbo",
+       model="gpt-3.5-turbo",
-        messages=[{"role": "user", "content": "Why is RunPod the best platform?"}],
+       messages=[{"role": "user", "content": "Why is RunPod the best platform?"}],
-        temperature=0,
+       temperature=0,
-        max_tokens=100,
+       max_tokens=100,
-    )
+   )
-    ```
+   ```
-    - After:
+   - After:
-    ```python
+   ```python
-    response = client.chat.completions.create(
+   response = client.chat.completions.create(
-        model="<YOUR DEPLOYED MODEL REPO/NAME>",
+       model="<YOUR DEPLOYED MODEL REPO/NAME>",
-        messages=[{"role": "user", "content": "Why is RunPod the best platform?"}],
+       messages=[{"role": "user", "content": "Why is RunPod the best platform?"}],
-        temperature=0,
+       temperature=0,
-        max_tokens=100,
+       max_tokens=100,
-    )
+   )
-    ```
+   ```
 
 **Using http requests**:
+
 1. Change the `Authorization` header to your RunPod API Key and the `url` to your RunPod Serverless Endpoint URL in the following format: `https://api.runpod.ai/v2/<YOUR ENDPOINT ID>/openai/v1`
-    - Before:
+   - Before:
-    ```bash
+   ```bash
-    curl https://api.openai.com/v1/chat/completions \
+   curl https://api.openai.com/v1/chat/completions \
-    -H "Content-Type: application/json" \
+   -H "Content-Type: application/json" \
-    -H "Authorization: Bearer $OPENAI_API_KEY" \
+   -H "Authorization: Bearer $OPENAI_API_KEY" \
-    -d '{
+   -d '{
-    "model": "gpt-4",
+   "model": "gpt-4",
-    "messages": [
+   "messages": [
-      {
+     {
-        "role": "user",
+       "role": "user",
-        "content": "Why is RunPod the best platform?"
+       "content": "Why is RunPod the best platform?"
-      }
+     }
-    ],
+   ],
-    "temperature": 0,
+   "temperature": 0,
-    "max_tokens": 100
+   "max_tokens": 100
-    }'
+   }'
-    ```
+   ```
-    - After:
+   - After:
-    ```bash
+   ```bash
-    curl https://api.runpod.ai/v2/<YOUR ENDPOINT ID>/openai/v1/chat/completions \
+   curl https://api.runpod.ai/v2/<YOUR ENDPOINT ID>/openai/v1/chat/completions \
-    -H "Content-Type: application/json" \
+   -H "Content-Type: application/json" \
-    -H "Authorization: Bearer <YOUR OPENAI API KEY>" \
+   -H "Authorization: Bearer <YOUR OPENAI API KEY>" \
-    -d '{
+   -d '{
-    "model": "<YOUR DEPLOYED MODEL REPO/NAME>",
+   "model": "<YOUR DEPLOYED MODEL REPO/NAME>",
-    "messages": [
+   "messages": [
-      {
+     {
-        "role": "user",
+       "role": "user",
-        "content": "Why is RunPod the best platform?"
+       "content": "Why is RunPod the best platform?"
-      }
+     }
-    ],
+   ],
-    "temperature": 0,
+   "temperature": 0,
-    "max_tokens": 100
+   "max_tokens": 100
-    }'
+   }'
-    ```
+   ```
 
 ## OpenAI Request Input Parameters:
 
 When using the chat completion feature of the vLLM Serverless Endpoint Worker, you can customize your requests with the following parameters:
 
-### Chat Completions
+### Chat Completions [RECOMMENDED]
+
 <details>
   <summary>Supported Chat Completions Inputs and Descriptions</summary>
 
-  | Parameter                      | Type                             | Default Value | Description                                                                                                                                                                                                                                                                                           |
+| Parameter           | Type                             | Default Value | Description                                                                                                                                                                                                                                                  |
-  |--------------------------------|----------------------------------|---------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| ------------------- | -------------------------------- | ------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
-  | `messages`                     | Union[str, List[Dict[str, str]]] |               | List of messages, where each message is a dictionary with a `role` and `content`. The model's chat template will be applied to the messages automatically, so the model must have one or it should be specified as `CUSTOM_CHAT_TEMPLATE` env var.                                                                                                                        |
+| `messages`          | Union[str, List[Dict[str, str]]] |               | List of messages, where each message is a dictionary with a `role` and `content`. The model's chat template will be applied to the messages automatically, so the model must have one or it should be specified as `CUSTOM_CHAT_TEMPLATE` env var.           |
-  | `model`                        | str                              |               | The model repo that you've deployed on your RunPod Serverless Endpoint. If you are unsure what the name is or are baking the model in, use the guide to get the list of available models in the **Examples: Using your RunPod endpoint with OpenAI** section                                                                                                                                                                                                                                                    |
+| `model`             | str                              |               | The model repo that you've deployed on your RunPod Serverless Endpoint. If you are unsure what the name is or are baking the model in, use the guide to get the list of available models in the **Examples: Using your RunPod endpoint with OpenAI** section |
-  | `temperature`                  | Optional[float]                  | 0.7           | Float that controls the randomness of the sampling. Lower values make the model more deterministic, while higher values make the model more random. Zero means greedy sampling.                                                                                                                                                                                                                               |
+| `temperature`       | Optional[float]                  | 0.7           | Float that controls the randomness of the sampling. Lower values make the model more deterministic, while higher values make the model more random. Zero means greedy sampling.                                                                              |
-  | `top_p`                        | Optional[float]                  | 1.0           | Float that controls the cumulative probability of the top tokens to consider. Must be in (0, 1]. Set to 1 to consider all tokens. |
+| `top_p`             | Optional[float]                  | 1.0           | Float that controls the cumulative probability of the top tokens to consider. Must be in (0, 1]. Set to 1 to consider all tokens.                                                                                                                            |
-  | `n`                            | Optional[int]                    | 1             | Number of output sequences to return for the given prompt. |
+| `n`                 | Optional[int]                    | 1             | Number of output sequences to return for the given prompt.                                                                                                                                                                                                   |
-  | `max_tokens`                   | Optional[int]                    | None          | Maximum number of tokens to generate per output sequence. |
+| `max_tokens`        | Optional[int]                    | None          | Maximum number of tokens to generate per output sequence.                                                                                                                                                                                                    |
-  | `seed`                         | Optional[int]                    | None          | Random seed to use for the generation. |
+| `seed`              | Optional[int]                    | None          | Random seed to use for the generation.                                                                                                                                                                                                                       |
-  | `stop`                         | Optional[Union[str, List[str]]]  | list          | List of strings that stop the generation when they are generated. The returned output will not contain the stop strings. |
+| `stop`              | Optional[Union[str, List[str]]]  | list          | List of strings that stop the generation when they are generated. The returned output will not contain the stop strings.                                                                                                                                     |
-  | `stream`                       | Optional[bool]                   | False         |           Whether to stream or not                                                                                                                                                                                                                                                                                            |
+| `stream`            | Optional[bool]                   | False         | Whether to stream or not                                                                                                                                                                                                                                     |
-  | `presence_penalty`             | Optional[float]                  | 0.0           | Float that penalizes new tokens based on whether they appear in the generated text so far. Values > 0 encourage the model to use new tokens, while values < 0 encourage the model to repeat tokens. |
+| `presence_penalty`  | Optional[float]                  | 0.0           | Float that penalizes new tokens based on whether they appear in the generated text so far. Values > 0 encourage the model to use new tokens, while values < 0 encourage the model to repeat tokens.                                                          |
-  | `frequency_penalty`            | Optional[float]                  | 0.0           | Float that penalizes new tokens based on their frequency in the generated text so far. Values > 0 encourage the model to use new tokens, while values < 0 encourage the model to repeat tokens. |
+| `frequency_penalty` | Optional[float]                  | 0.0           | Float that penalizes new tokens based on their frequency in the generated text so far. Values > 0 encourage the model to use new tokens, while values < 0 encourage the model to repeat tokens.                                                              |
-  | `logit_bias`                   | Optional[Dict[str, float]]       | None          |           Unsupported by vLLM                                                                                                                                                                                                                                                                                         |
+| `logit_bias`        | Optional[Dict[str, float]]       | None          | Unsupported by vLLM                                                                                                                                                                                                                                          |
-  | `user`                         | Optional[str]                    | None          |               Unsupported by vLLM                                                                                                                                                                                                                                                                                                                                                                         |
+| `user`              | Optional[str]                    | None          | Unsupported by vLLM                                                                                                                                                                                                                                          |
-  Additional parameters supported by vLLM:
+
-  | `best_of`                      | Optional[int]                    | None          | Number of output sequences that are generated from the prompt. From these `best_of` sequences, the top `n` sequences are returned. `best_of` must be greater than or equal to `n`. This is treated as the beam width when `use_beam_search` is True. By default, `best_of` is set to `n`. |
+Additional parameters supported by vLLM:
-  | `top_k`                        | Optional[int]                    | -1            | Integer that controls the number of top tokens to consider. Set to -1 to consider all tokens. |
+| `best_of` | Optional[int] | None | Number of output sequences that are generated from the prompt. From these `best_of` sequences, the top `n` sequences are returned. `best_of` must be greater than or equal to `n`. This is treated as the beam width when `use_beam_search` is True. By default, `best_of` is set to `n`. |
-  | `ignore_eos`                   | Optional[bool]                   | False         | Whether to ignore the EOS token and continue generating tokens after the EOS token is generated. |
+| `top_k` | Optional[int] | -1 | Integer that controls the number of top tokens to consider. Set to -1 to consider all tokens. |
-  | `use_beam_search`              | Optional[bool]                   | False         | Whether to use beam search instead of sampling. |
+| `ignore_eos` | Optional[bool] | False | Whether to ignore the EOS token and continue generating tokens after the EOS token is generated. |
-  | `stop_token_ids`               | Optional[List[int]]              | list          | List of tokens that stop the generation when they are generated. The returned output will contain the stop tokens unless the stop tokens are special tokens. |
+| `use_beam_search` | Optional[bool] | False | Whether to use beam search instead of sampling. |
-  | `skip_special_tokens`          | Optional[bool]                   | True          | Whether to skip special tokens in the output. |
+| `stop_token_ids` | Optional[List[int]] | list | List of tokens that stop the generation when they are generated. The returned output will contain the stop tokens unless the stop tokens are special tokens. |
-  | `spaces_between_special_tokens`| Optional[bool]                   | True          | Whether to add spaces between special tokens in the output. Defaults to True. |
+| `skip_special_tokens` | Optional[bool] | True | Whether to skip special tokens in the output. |
-  | `add_generation_prompt`        | Optional[bool]                   | True          |       Read more [here](https://huggingface.co/docs/transformers/main/en/chat_templating#what-are-generation-prompts)                                                                                                                                                                                                                                                                                                |
+| `spaces_between_special_tokens`| Optional[bool] | True | Whether to add spaces between special tokens in the output. Defaults to True. |
-  | `echo`                         | Optional[bool]                   | False         |      Echo back the prompt in addition to the completion                                                                                                                                                                                                                                                                                                 |
+| `add_generation_prompt` | Optional[bool] | True | Read more [here](https://huggingface.co/docs/transformers/main/en/chat_templating#what-are-generation-prompts) |
-  | `repetition_penalty`           | Optional[float]                  | 1.0           | Float that penalizes new tokens based on whether they appear in the prompt and the generated text so far. Values > 1 encourage the model to use new tokens, while values < 1 encourage the model to repeat tokens. |
+| `echo` | Optional[bool] | False | Echo back the prompt in addition to the completion |
-  | `min_p`                        | Optional[float]                  | 0.0           | Float that represents the minimum probability for a token to |
+| `repetition_penalty` | Optional[float] | 1.0 | Float that penalizes new tokens based on whether they appear in the prompt and the generated text so far. Values > 1 encourage the model to use new tokens, while values < 1 encourage the model to repeat tokens. |
-  | `length_penalty`               | Optional[float]                  | 1.0           | Float that penalizes sequences based on their length. Used in beam search.. |
+| `min_p` | Optional[float] | 0.0 | Float that represents the minimum probability for a token to |
-  | `include_stop_str_in_output`   | Optional[bool]                   | False         | Whether to include the stop strings in output text. Defaults to False.|
+| `length_penalty` | Optional[float] | 1.0 | Float that penalizes sequences based on their length. Used in beam search.. |
+| `include_stop_str_in_output` | Optional[bool] | False | Whether to include the stop strings in output text. Defaults to False.|
+
 </details>
 
-### Completions
+### Examples: Using your RunPod endpoint with OpenAI
-<details>
-  <summary>Supported Completions Inputs and Descriptions</summary>
-
-  | Parameter                      | Type                             | Default Value | Description                                                                                                                                                                                                                                                                                           |
-  |--------------------------------|----------------------------------|---------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-  | `model`                        | str                              |               | The model repo that you've deployed on your RunPod Serverless Endpoint. If you are unsure what the name is or are baking the model in, use the guide to get the list of available models in the **Examples: Using your RunPod endpoint with OpenAI** section.                                                                                                                                                                                                                                                    |
-  | `prompt`                       | Union[List[int], List[List[int]], str, List[str]] | | A string, array of strings, array of tokens, or array of token arrays to be used as the input for the model. |
-  | `suffix`                       | Optional[str]                    | None          | A string to be appended to the end of the generated text. |
-  | `max_tokens`                   | Optional[int]                    | 16            | Maximum number of tokens to generate per output sequence. |
-  | `temperature`                  | Optional[float]                  | 1.0           | Float that controls the randomness of the sampling. Lower values make the model more deterministic, while higher values make the model more random. Zero means greedy sampling.                                                                                                                                                                                                                               |
-  | `top_p`                        | Optional[float]                  | 1.0           | Float that controls the cumulative probability of the top tokens to consider. Must be in (0, 1]. Set to 1 to consider all tokens. |
-  | `n`                            | Optional[int]                    | 1             | Number of output sequences to return for the given prompt. |
-  | `stream`                       | Optional[bool]                   | False         | Whether to stream the output. |
-  | `logprobs`                     | Optional[int]                    | None          | Number of log probabilities to return per output token. |
-  | `echo`                         | Optional[bool]                   | False         | Whether to echo back the prompt in addition to the completion. |
-  | `stop`                         | Optional[Union[str, List[str]]]  | list          | List of strings that stop the generation when they are generated. The returned output will not contain the stop strings. |
-  | `seed`                         | Optional[int]                    | None          | Random seed to use for the generation. |
-  | `presence_penalty`             | Optional[float]                  | 0.0           | Float that penalizes new tokens based on whether they appear in the generated text so far. Values > 0 encourage the model to use new tokens, while values < 0 encourage the model to repeat tokens. |
-  | `frequency_penalty`            | Optional[float]                  | 0.0           | Float that penalizes new tokens based on their frequency in the generated text so far. Values > 0 encourage the model to use new tokens, while values < 0 encourage the model to repeat tokens. |
-  | `best_of`                      | Optional[int]                    | None          | Number of output sequences that are generated from the prompt. From these `best_of` sequences, the top `n` sequences are returned. `best_of` must be greater than or equal to `n`. This parameter influences the diversity of the output. |
-  | `logit_bias`                   | Optional[Dict[str, float]]       | None          | Dictionary of token IDs to biases. |
-  | `user`                         | Optional[str]                    | None          | User identifier for personalizing responses. (Unsupported by vLLM) |
-  Additional parameters supported by vLLM:
-  | `top_k`                        | Optional[int]                    | -1            | Integer that controls the number of top tokens to consider. Set to -1 to consider all tokens. |
-  | `ignore_eos`                   | Optional[bool]                   | False         | Whether to ignore the End Of Sentence token and continue generating tokens after the EOS token is generated. |
-  | `use_beam_search`              | Optional[bool]                   | False         | Whether to use beam search instead of sampling for generating outputs. |
-  | `stop_token_ids`               | Optional[List[int]]              | list          | List of tokens that stop the generation when they are generated. The returned output will contain the stop tokens unless the stop tokens are special tokens. |
-  | `skip_special_tokens`          | Optional[bool]                   | True          | Whether to skip special tokens in the output. |
-  | `spaces_between_special_tokens`| Optional[bool]                   | True          | Whether to add spaces between special tokens in the output. Defaults to True. |
-  | `repetition_penalty`           | Optional[float]                  | 1.0           | Float that penalizes new tokens based on whether they appear in the prompt and the generated text so far. Values > 1 encourage the model to use new tokens, while values < 1 encourage the model to repeat tokens. |
-  | `min_p`                        | Optional[float]                  | 0.0           | Float that represents the minimum probability for a token to be considered, relative to the most likely token. Must be in [0, 1]. Set to 0 to disable. |
-  | `length_penalty`               | Optional[float]                  | 1.0           | Float that penalizes sequences based on their length. Used in beam search. |
-  | `include_stop_str_in_output`   | Optional[bool]                   | False         | Whether to include the stop strings in output text. Defaults to False. |
-</details>
-
-## Examples: Using your RunPod endpoint with OpenAI 
 
 First, initialize the OpenAI Client with your RunPod API Key and Endpoint URL:
+
 ```python
 from openai import OpenAI
 import os
@@ -381,7 +260,9 @@ client = OpenAI(
 ```
 
 ### Chat Completions:
+
 This is the format used for GPT-4 and focused on instruction-following and chat. Examples of Open Source chat/instruct models include `meta-llama/Llama-2-7b-chat-hf`, `mistralai/Mixtral-8x7B-Instruct-v0.1`, `openchat/openchat-3.5-0106`, `NousResearch/Nous-Hermes-2-Mistral-7B-DPO` and more. However, if your model is a completion-style model with no chat/instruct fine-tune and/or does not have a chat template, you can still use this if you provide a chat template with the environment variable `CUSTOM_CHAT_TEMPLATE`.
+
 - **Streaming**:
   ```python
   # Create a chat completion stream
@@ -409,38 +290,10 @@ This is the format used for GPT-4 and focused on instruction-following and chat.
   print(response.choices[0].message.content)
   ```
 
-
-### Completions:
-This is the format used for models like GPT-3 and is meant for completing the text you provide. Instead of responding to your message, it will try to complete it. Examples of Open Source completions models include `meta-llama/Llama-2-7b-hf`, `mistralai/Mixtral-8x7B-v0.1`, `Qwen/Qwen-72B`, and more. However, you can use any model with this format.
-- **Streaming**:
-  ```python
-  # Create a completion stream
-  response_stream = client.completions.create(
-      model="<YOUR DEPLOYED MODEL REPO/NAME>",
-      prompt="Runpod is the best platform because",
-      temperature=0,
-      max_tokens=100,
-      stream=True,
-  )
-  # Stream the response
-  for response in response_stream:
-      print(response.choices[0].text or "", end="", flush=True)
-  ```
-- **Non-Streaming**:
-  ```python
-  # Create a completion
-  response = client.completions.create(
-      model="<YOUR DEPLOYED MODEL REPO/NAME>",
-      prompt="Runpod is the best platform because",
-      temperature=0,
-      max_tokens=100,
-  )
-  # Print the response
-  print(response.choices[0].text)
-  ```
-
 ### Getting a list of names for available models:
+
 In the case of baking the model into the image, sometimes the repo may not be accepted as the `model` in the request. In this case, you can list the available models as shown below and use that name.
+
 ```python
 models_response = client.models.list()
 list_of_models = [model.id for model in models_response]
@@ -448,6 +301,7 @@ print(list_of_models)
 ```
 
 # Usage: Standard (Non-OpenAI)
+
 ## Request Input Parameters
 
 <details>
@@ -469,65 +323,83 @@ print(list_of_models)
 ### Sampling Parameters
 
 Below are all available sampling parameters that you can specify in the `sampling_params` dictionary. If you do not specify any of these parameters, the default values will be used.
+
 <details>
   <summary>Click to expand table</summary>
 
-  | Argument                        | Type                        | Default | Description                                                                                                                                                                                   |
+| Argument                        | Type                        | Default | Description                                                                                                                                                                                   |
-  |---------------------------------|-----------------------------|---------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| ------------------------------- | --------------------------- | ------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-  | `n`                             | int                         | 1       | Number of output sequences generated from the prompt. The top `n` sequences are returned.                                                                                                      |
+| `n`                             | int                         | 1       | Number of output sequences generated from the prompt. The top `n` sequences are returned.                                                                                                     |
-  | `best_of`                       | Optional[int]               | `n`    | Number of output sequences generated from the prompt. The top `n` sequences are returned from these `best_of` sequences. Must be ≥ `n`. Treated as beam width in beam search. Default is `n`. |
+| `best_of`                       | Optional[int]               | `n`     | Number of output sequences generated from the prompt. The top `n` sequences are returned from these `best_of` sequences. Must be ≥ `n`. Treated as beam width in beam search. Default is `n`. |
-  | `presence_penalty`              | float                       | 0.0     | Penalizes new tokens based on their presence in the generated text so far. Values > 0 encourage new tokens, values < 0 encourage repetition.                                                  |
+| `presence_penalty`              | float                       | 0.0     | Penalizes new tokens based on their presence in the generated text so far. Values > 0 encourage new tokens, values < 0 encourage repetition.                                                  |
-  | `frequency_penalty`             | float                       | 0.0     | Penalizes new tokens based on their frequency in the generated text so far. Values > 0 encourage new tokens, values < 0 encourage repetition.                                                 |
+| `frequency_penalty`             | float                       | 0.0     | Penalizes new tokens based on their frequency in the generated text so far. Values > 0 encourage new tokens, values < 0 encourage repetition.                                                 |
-  | `repetition_penalty`            | float                       | 1.0     | Penalizes new tokens based on their appearance in the prompt and generated text. Values > 1 encourage new tokens, values < 1 encourage repetition.                                            |
+| `repetition_penalty`            | float                       | 1.0     | Penalizes new tokens based on their appearance in the prompt and generated text. Values > 1 encourage new tokens, values < 1 encourage repetition.                                            |
-  | `temperature`                   | float                       | 1.0     | Controls the randomness of sampling. Lower values make it more deterministic, higher values make it more random. Zero means greedy sampling.                                                  |
+| `temperature`                   | float                       | 1.0     | Controls the randomness of sampling. Lower values make it more deterministic, higher values make it more random. Zero means greedy sampling.                                                  |
-  | `top_p`                         | float                       | 1.0     | Controls the cumulative probability of top tokens to consider. Must be in (0, 1]. Set to 1 to consider all tokens.                                                                            |
+| `top_p`                         | float                       | 1.0     | Controls the cumulative probability of top tokens to consider. Must be in (0, 1]. Set to 1 to consider all tokens.                                                                            |
-  | `top_k`                         | int                         | -1      | Controls the number of top tokens to consider. Set to -1 to consider all tokens.                                                                                                              |
+| `top_k`                         | int                         | -1      | Controls the number of top tokens to consider. Set to -1 to consider all tokens.                                                                                                              |
-  | `min_p`                         | float                       | 0.0     | Represents the minimum probability for a token to be considered, relative to the most likely token. Must be in [0, 1]. Set to 0 to disable.                                                   |
+| `min_p`                         | float                       | 0.0     | Represents the minimum probability for a token to be considered, relative to the most likely token. Must be in [0, 1]. Set to 0 to disable.                                                   |
-  | `use_beam_search`               | bool                        | False   | Whether to use beam search instead of sampling.                                                                                                                                               |
+| `use_beam_search`               | bool                        | False   | Whether to use beam search instead of sampling.                                                                                                                                               |
-  | `length_penalty`                | float                       | 1.0     | Penalizes sequences based on their length. Used in beam search.                                                                                                                               |
+| `length_penalty`                | float                       | 1.0     | Penalizes sequences based on their length. Used in beam search.                                                                                                                               |
-  | `early_stopping`                | Union[bool, str]            | False   | Controls stopping condition in beam search. Can be `True`, `False`, or `"never"`.                                                                                                             |
+| `early_stopping`                | Union[bool, str]            | False   | Controls stopping condition in beam search. Can be `True`, `False`, or `"never"`.                                                                                                             |
-  | `stop`                          | Union[None, str, List[str]] | None    | List of strings that stop generation when produced. The output will not contain these strings.                                                                                                    |
+| `stop`                          | Union[None, str, List[str]] | None    | List of strings that stop generation when produced. The output will not contain these strings.                                                                                                |
-  | `stop_token_ids`                | Optional[List[int]]         | None    | List of token IDs that stop generation when produced. Output contains these tokens unless they are special tokens.                                                                            |
+| `stop_token_ids`                | Optional[List[int]]         | None    | List of token IDs that stop generation when produced. Output contains these tokens unless they are special tokens.                                                                            |
-  | `ignore_eos`                    | bool                        | False   | Whether to ignore the End-Of-Sequence token and continue generating tokens after its generation.                                                                                              |
+| `ignore_eos`                    | bool                        | False   | Whether to ignore the End-Of-Sequence token and continue generating tokens after its generation.                                                                                              |
-  | `max_tokens`                    | int                         | 16      | Maximum number of tokens to generate per output sequence.                                                                                                                                     |
+| `max_tokens`                    | int                         | 16      | Maximum number of tokens to generate per output sequence.                                                                                                                                     |
-  | `skip_special_tokens`           | bool                        | True    | Whether to skip special tokens in the output.                                                                                                                                                 |
+| `skip_special_tokens`           | bool                        | True    | Whether to skip special tokens in the output.                                                                                                                                                 |
-  | `spaces_between_special_tokens` | bool                        | True    | Whether to add spaces between special tokens in the output.                                                                                                                                   |
+| `spaces_between_special_tokens` | bool                        | True    | Whether to add spaces between special tokens in the output.                                                                                                                                   |
-
 
 ### Text Input Formats
-You may either use a `prompt` or a list of `messages` as input.
- 1. `prompt` 
-The prompt string can be any string, and the model's chat template will not be applied to it unless `apply_chat_template` is set to `true`, in which case it will be treated as a user message.
 
-    Example:
+You may either use a `prompt` or a list of `messages` as input.
-    ```json
+
-    "prompt": "..."
+1.  `prompt`
-    ```
+    The prompt string can be any string, and the model's chat template will not be applied to it unless `apply_chat_template` is set to `true`, in which case it will be treated as a user message.
-2. `messages`
+
-Your list can contain any number of messages, and each message usually can have any role from the following list:
+        Example:
-    - `user`
+        ```json
-    - `assistant`
+        {
-    - `system`
+          "input": {
+            "prompt": "why sky is blue?",
+            "sampling_params": {
+              "temperature": 0.7,
+              "max_tokens": 100
+            }
+          }
+        }
+        ```
+
+2.  `messages`
+    Your list can contain any number of messages, and each message usually can have any role from the following list: - `user` - `assistant` - `system`
 
     However, some models may have different roles, so you should check the model's chat template to see which roles are required.
 
     The model's chat template will be applied to the messages automatically, so the model must have one.
 
     Example:
+
     ```json
-    "messages": [
+    {
-        {
+      "input": {
-          "role": "system",
+        "messages": [
-          "content": "..."
+          {
-        },
+            "role": "system",
-        {
+            "content": "You are a helpful AI assistant that provides clear and concise responses."
-          "role": "user",
+          },
-          "content": "..."
+          {
-        },
+            "role": "user",
-        {
+            "content": "Can you explain the difference between supervised and unsupervised learning?"
-          "role": "assistant",
+          },
-          "content": "..."
+          {
+            "role": "assistant",
+            "content": "Sure! Supervised learning uses labeled data, meaning each input has a corresponding correct output. The model learns by mapping inputs to known outputs. In contrast, unsupervised learning works with unlabeled data, where the model identifies patterns, structures, or clusters without predefined answers."
+          }
+        ],
+        "sampling_params": {
+          "temperature": 0.7,
+          "max_tokens": 100
         }
-      ]
+      }
+    }
     ```
 
+</details>

builder/requirements.txt

@@ -1,10 +1,14 @@
 ray
 pandas
 pyarrow
-runpod==1.6.2
+runpod~=1.7.7
 huggingface-hub
 packaging
-typing-extensions==4.7.1
+typing-extensions>=4.8.0
 pydantic
 pydantic-settings
 hf-transfer
+transformers>=4.57.0
+bitsandbytes>=0.45.0
+kernels
+torch==2.6.0

docker-bake.hcl

@@ -1,65 +1,26 @@
-variable "PUSH" {
+variable "DOCKERHUB_REPO" {
-  default = "true"
-}
-
-variable "REPOSITORY" {
   default = "runpod"
 }
 
-variable "BASE_IMAGE_VERSION" {
+variable "DOCKERHUB_IMG" {
-  default = "1.0.0"
+  default = "worker-v1-vllm"
 }
 
-group "all" {
+variable "RELEASE_VERSION" {
-  targets = ["base", "main"]
+  default = "latest"
 }
 
-group "base" {
+variable "HUGGINGFACE_ACCESS_TOKEN" {
-  targets = ["base-1180", "base-1210"]
+  default = ""
 }
 
-group "main" {
+group "default" {
-  targets = ["worker-1180", "worker-1210"]
+  targets = ["worker-vllm"]
 }
 
-target "base-1180" {
+target "worker-vllm" {
-  tags = ["${REPOSITORY}/worker-vllm:base-${BASE_IMAGE_VERSION}-cuda11.8.0"]
+  tags = ["${DOCKERHUB_REPO}/${DOCKERHUB_IMG}:${RELEASE_VERSION}"]
-  context = "vllm-base-image"
-  dockerfile = "Dockerfile"
-  args = {
-    WORKER_CUDA_VERSION = "11.8.0"
-  }
-  output = ["type=docker,push=${PUSH}"]
-}
-
-target "base-1210" {
-  tags = ["${REPOSITORY}/worker-vllm:base-${BASE_IMAGE_VERSION}-cuda12.1.0"]
-  context = "vllm-base-image"
-  dockerfile = "Dockerfile"
-  args = {
-    WORKER_CUDA_VERSION = "12.1.0"
-  }
-  output = ["type=docker,push=${PUSH}"]
-}
-
-target "worker-1180" {
-  tags = ["${REPOSITORY}/worker-vllm:${BASE_IMAGE_VERSION}-cuda11.8.0"]
   context = "."
   dockerfile = "Dockerfile"
-  args = {
+  platforms = ["linux/amd64"]
-    BASE_IMAGE_VERSION = "${BASE_IMAGE_VERSION}"
-    WORKER_CUDA_VERSION = "11.8.0"
-  }
-  output = ["type=docker,push=${PUSH}"]
-}
-
-target "worker-1210" {
-  tags = ["${REPOSITORY}/worker-vllm:${BASE_IMAGE_VERSION}-cuda12.1.0"]
-  context = "."
-  dockerfile = "Dockerfile"
-  args = {
-    BASE_IMAGE_VERSION = "${BASE_IMAGE_VERSION}"
-    WORKER_CUDA_VERSION = "12.1.0"
-  }
-  output = ["type=docker,push=${PUSH}"]
 }

docs/configuration.md

@@ -0,0 +1,153 @@
+# Configuration Reference
+
+Complete guide to all environment variables and configuration options for worker-vllm.
+
+## LLM Settings
+
+| Variable                       | Default             | Type/Choices                                                | Description                                                                     |
+| ------------------------------ | ------------------- | ----------------------------------------------------------- | ------------------------------------------------------------------------------- |
+| `MODEL_NAME`                   | 'facebook/opt-125m' | `str`                                                       | Name or path of the Hugging Face model to use.                                  |
+| `MODEL_REVISION`               | 'main'              | `str`                                                       | Model revision to load (default: main).                                         |
+| `TOKENIZER`                    | None                | `str`                                                       | Name or path of the Hugging Face tokenizer to use.                              |
+| `SKIP_TOKENIZER_INIT`          | False               | `bool`                                                      | Skip initialization of tokenizer and detokenizer.                               |
+| `TOKENIZER_MODE`               | 'auto'              | ['auto', 'slow']                                            | The tokenizer mode.                                                             |
+| `TRUST_REMOTE_CODE`            | `False`             | `bool`                                                      | Trust remote code from Hugging Face.                                            |
+| `DOWNLOAD_DIR`                 | None                | `str`                                                       | Directory to download and load the weights.                                     |
+| `LOAD_FORMAT`                  | 'auto'              | `str`                                                       | The format of the model weights to load.                                        |
+| `HF_TOKEN`                     | -                   | `str`                                                       | Hugging Face token for private and gated models.                                |
+| `DTYPE`                        | 'auto'              | ['auto', 'half', 'float16', 'bfloat16', 'float', 'float32'] | Data type for model weights and activations.                                    |
+| `KV_CACHE_DTYPE`               | 'auto'              | ['auto', 'fp8']                                             | Data type for KV cache storage.                                                 |
+| `QUANTIZATION_PARAM_PATH`      | None                | `str`                                                       | Path to the JSON file containing the KV cache scaling factors.                  |
+| `MAX_MODEL_LEN`                | None                | `int`                                                       | Model context length.                                                           |
+| `GUIDED_DECODING_BACKEND`      | 'outlines'          | ['outlines', 'lm-format-enforcer']                          | Which engine will be used for guided decoding by default.                       |
+| `DISTRIBUTED_EXECUTOR_BACKEND` | None                | ['ray', 'mp']                                               | Backend to use for distributed serving.                                         |
+| `WORKER_USE_RAY`               | False               | `bool`                                                      | Deprecated, use --distributed-executor-backend=ray.                             |
+| `PIPELINE_PARALLEL_SIZE`       | 1                   | `int`                                                       | Number of pipeline stages.                                                      |
+| `TENSOR_PARALLEL_SIZE`         | 1                   | `int`                                                       | Number of tensor parallel replicas.                                             |
+| `MAX_PARALLEL_LOADING_WORKERS` | None                | `int`                                                       | Load model sequentially in multiple batches.                                    |
+| `RAY_WORKERS_USE_NSIGHT`       | False               | `bool`                                                      | If specified, use nsight to profile Ray workers.                                |
+| `ENABLE_PREFIX_CACHING`        | False               | `bool`                                                      | Enables automatic prefix caching.                                               |
+| `DISABLE_SLIDING_WINDOW`       | False               | `bool`                                                      | Disables sliding window, capping to sliding window size.                        |
+| `USE_V2_BLOCK_MANAGER`         | False               | `bool`                                                      | Use BlockSpaceMangerV2.                                                         |
+| `NUM_LOOKAHEAD_SLOTS`          | 0                   | `int`                                                       | Experimental scheduling config necessary for speculative decoding.              |
+| `SEED`                         | 0                   | `int`                                                       | Random seed for operations.                                                     |
+| `NUM_GPU_BLOCKS_OVERRIDE`      | None                | `int`                                                       | If specified, ignore GPU profiling result and use this number of GPU blocks.    |
+| `MAX_NUM_BATCHED_TOKENS`       | None                | `int`                                                       | Maximum number of batched tokens per iteration.                                 |
+| `MAX_NUM_SEQS`                 | 256                 | `int`                                                       | Maximum number of sequences per iteration.                                      |
+| `MAX_LOGPROBS`                 | 20                  | `int`                                                       | Max number of log probs to return when logprobs is specified in SamplingParams. |
+| `DISABLE_LOG_STATS`            | False               | `bool`                                                      | Disable logging statistics.                                                     |
+| `QUANTIZATION`                 | None                | ['awq', 'squeezellm', 'gptq', 'bitsandbytes']               | Method used to quantize the weights.                                            |
+| `ROPE_SCALING`                 | None                | `dict`                                                      | RoPE scaling configuration in JSON format.                                      |
+| `ROPE_THETA`                   | None                | `float`                                                     | RoPE theta. Use with rope_scaling.                                              |
+| `TOKENIZER_POOL_SIZE`          | 0                   | `int`                                                       | Size of tokenizer pool to use for asynchronous tokenization.                    |
+| `TOKENIZER_POOL_TYPE`          | 'ray'               | `str`                                                       | Type of tokenizer pool to use for asynchronous tokenization.                    |
+| `TOKENIZER_POOL_EXTRA_CONFIG`  | None                | `dict`                                                      | Extra config for tokenizer pool.                                                |
+
+## LoRA (Low-Rank Adaptation) Settings
+
+| Variable                    | Default | Type                                       | Description                                                                                               |
+| --------------------------- | ------- | ------------------------------------------ | --------------------------------------------------------------------------------------------------------- |
+| `ENABLE_LORA`               | False   | `bool`                                     | If True, enable handling of LoRA adapters.                                                                |
+| `MAX_LORAS`                 | 1       | `int`                                      | Max number of LoRAs in a single batch.                                                                    |
+| `MAX_LORA_RANK`             | 16      | `int`                                      | Max LoRA rank.                                                                                            |
+| `LORA_EXTRA_VOCAB_SIZE`     | 256     | `int`                                      | Maximum size of extra vocabulary for LoRA adapters.                                                       |
+| `LORA_DTYPE`                | 'auto'  | ['auto', 'float16', 'bfloat16', 'float32'] | Data type for LoRA.                                                                                       |
+| `LONG_LORA_SCALING_FACTORS` | None    | `tuple`                                    | Specify multiple scaling factors for LoRA adapters.                                                       |
+| `MAX_CPU_LORAS`             | None    | `int`                                      | Maximum number of LoRAs to store in CPU memory.                                                           |
+| `FULLY_SHARDED_LORAS`       | False   | `bool`                                     | Enable fully sharded LoRA layers.                                                                         |
+| `LORA_MODULES`              | `[]`    | `list[dict]`                               | Add lora adapters from Hugging Face `[{"name": "xx", "path": "xxx/xxxx", "base_model_name": "xxx/xxxx"}]` |
+
+## Speculative Decoding Settings
+
+| Variable                                         | Default             | Type/Choices                                        | Description                                                                               |
+| ------------------------------------------------ | ------------------- | --------------------------------------------------- | ----------------------------------------------------------------------------------------- |
+| `SCHEDULER_DELAY_FACTOR`                         | 0.0                 | `float`                                             | Apply a delay before scheduling next prompt.                                              |
+| `ENABLE_CHUNKED_PREFILL`                         | False               | `bool`                                              | Enable chunked prefill requests.                                                          |
+| `SPECULATIVE_MODEL`                              | None                | `str`                                               | The name of the draft model to be used in speculative decoding.                           |
+| `NUM_SPECULATIVE_TOKENS`                         | None                | `int`                                               | The number of speculative tokens to sample from the draft model.                          |
+| `SPECULATIVE_DRAFT_TENSOR_PARALLEL_SIZE`         | None                | `int`                                               | Number of tensor parallel replicas for the draft model.                                   |
+| `SPECULATIVE_MAX_MODEL_LEN`                      | None                | `int`                                               | The maximum sequence length supported by the draft model.                                 |
+| `SPECULATIVE_DISABLE_BY_BATCH_SIZE`              | None                | `int`                                               | Disable speculative decoding if the number of enqueue requests is larger than this value. |
+| `NGRAM_PROMPT_LOOKUP_MAX`                        | None                | `int`                                               | Max size of window for ngram prompt lookup in speculative decoding.                       |
+| `NGRAM_PROMPT_LOOKUP_MIN`                        | None                | `int`                                               | Min size of window for ngram prompt lookup in speculative decoding.                       |
+| `SPEC_DECODING_ACCEPTANCE_METHOD`                | 'rejection_sampler' | ['rejection_sampler', 'typical_acceptance_sampler'] | Specify the acceptance method for draft token verification in speculative decoding.       |
+| `TYPICAL_ACCEPTANCE_SAMPLER_POSTERIOR_THRESHOLD` | None                | `float`                                             | Set the lower bound threshold for the posterior probability of a token to be accepted.    |
+| `TYPICAL_ACCEPTANCE_SAMPLER_POSTERIOR_ALPHA`     | None                | `float`                                             | A scaling factor for the entropy-based threshold for token acceptance.                    |
+
+## System Performance Settings
+
+| Variable                       | Default | Type/Choices    | Description                                                                                                                         |
+| ------------------------------ | ------- | --------------- | ----------------------------------------------------------------------------------------------------------------------------------- |
+| `GPU_MEMORY_UTILIZATION`       | `0.95`  | `float`         | Sets GPU VRAM utilization.                                                                                                          |
+| `MAX_PARALLEL_LOADING_WORKERS` | `None`  | `int`           | Load model sequentially in multiple batches, to avoid RAM OOM when using tensor parallel and large models.                          |
+| `BLOCK_SIZE`                   | `16`    | `8`, `16`, `32` | Token block size for contiguous chunks of tokens.                                                                                   |
+| `SWAP_SPACE`                   | `4`     | `int`           | CPU swap space size (GiB) per GPU.                                                                                                  |
+| `ENFORCE_EAGER`                | False   | `bool`          | Always use eager-mode PyTorch. If False(`0`), will use eager mode and CUDA graph in hybrid for maximal performance and flexibility. |
+| `MAX_SEQ_LEN_TO_CAPTURE`       | `8192`  | `int`           | Maximum context length covered by CUDA graphs. When a sequence has context length larger than this, we fall back to eager mode.     |
+| `DISABLE_CUSTOM_ALL_REDUCE`    | `0`     | `int`           | Enables or disables custom all reduce.                                                                                              |
+
+## Tokenizer Settings
+
+| Variable               | Default | Type/Choices                        | Description                                                                                       |
+| ---------------------- | ------- | ----------------------------------- | ------------------------------------------------------------------------------------------------- |
+| `TOKENIZER_NAME`       | `None`  | `str`                               | Tokenizer repository to use a different tokenizer than the model's default.                       |
+| `TOKENIZER_REVISION`   | `None`  | `str`                               | Tokenizer revision to load.                                                                       |
+| `CUSTOM_CHAT_TEMPLATE` | `None`  | `str` of single-line jinja template | Custom chat jinja template. [More Info](https://huggingface.co/docs/transformers/chat_templating) |
+
+## Streaming & Batch Settings
+
+The way this works is that the first request will have a batch size of `DEFAULT_MIN_BATCH_SIZE`, and each subsequent request will have a batch size of `previous_batch_size * DEFAULT_BATCH_SIZE_GROWTH_FACTOR`. This will continue until the batch size reaches `DEFAULT_BATCH_SIZE`. E.g. for the default values, the batch sizes will be `1, 3, 9, 27, 50, 50, 50, ...`. You can also specify this per request, with inputs `max_batch_size`, `min_batch_size`, and `batch_size_growth_factor`. This has nothing to do with vLLM's internal batching, but rather the number of tokens sent in each HTTP request from the worker.
+
+| Variable                           | Default | Type/Choices | Description                                                                                               |
+| ---------------------------------- | ------- | ------------ | --------------------------------------------------------------------------------------------------------- |
+| `DEFAULT_BATCH_SIZE`               | `50`    | `int`        | Default and Maximum batch size for token streaming to reduce HTTP calls.                                  |
+| `DEFAULT_MIN_BATCH_SIZE`           | `1`     | `int`        | Batch size for the first request, which will be multiplied by the growth factor every subsequent request. |
+| `DEFAULT_BATCH_SIZE_GROWTH_FACTOR` | `3`     | `float`      | Growth factor for dynamic batch size.                                                                     |
+
+## OpenAI Compatibility Settings
+
+| Variable                            | Default     | Type/Choices     | Description                                                                                                                                                                                                       |
+| ----------------------------------- | ----------- | ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `RAW_OPENAI_OUTPUT`                 | `1`         | boolean as `int` | Enables raw OpenAI SSE format string output when streaming. **Required** to be enabled (which it is by default) for OpenAI compatibility.                                                                         |
+| `OPENAI_SERVED_MODEL_NAME_OVERRIDE` | `None`      | `str`            | Overrides the name of the served model from model repo/path to specified name, which you will then be able to use the value for the `model` parameter when making OpenAI requests                                 |
+| `OPENAI_RESPONSE_ROLE`              | `assistant` | `str`            | Role of the LLM's Response in OpenAI Chat Completions.                                                                                                                                                            |
+| `ENABLE_AUTO_TOOL_CHOICE`           | `false`     | `bool`           | Enables automatic tool selection for supported models. Set to `true` to activate.                                                                                                                                 |
+| `TOOL_CALL_PARSER`                  | `None`      | `str`            | Specifies the parser for tool calls. Options: `mistral`, `hermes`, `llama3_json`, `llama4_json`, `llama4_pythonic`, `granite`, `granite-20b-fc`, `deepseek_v3`, `internlm`, `jamba`, `phi4_mini_json`, `pythonic` |
+| `REASONING_PARSER`                  | `None`      | `str`            | Parser for reasoning-capable models (enables reasoning mode). Examples: `deepseek_r1`, `qwen3`, `granite`, `hunyuan_a13b`. Leave unset to disable.                                                                |
+
+## Serverless & Concurrency Settings
+
+| Variable               | Default | Type/Choices | Description                                                                                                                                                                |
+| ---------------------- | ------- | ------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `MAX_CONCURRENCY`      | `30`    | `int`        | Max concurrent requests per worker. vLLM has an internal queue, so you don't have to worry about limiting by VRAM, this is for improving scaling/load balancing efficiency |
+| `DISABLE_LOG_STATS`    | False   | `bool`       | Enables or disables vLLM stats logging.                                                                                                                                    |
+| `DISABLE_LOG_REQUESTS` | False   | `bool`       | Enables or disables vLLM request logging.                                                                                                                                  |
+
+## Advanced Settings
+
+| Variable                    | Default | Type    | Description                                                                                                                                            |
+| --------------------------- | ------- | ------- | ------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| `MODEL_LOADER_EXTRA_CONFIG` | None    | `dict`  | Extra config for model loader.                                                                                                                         |
+| `PREEMPTION_MODE`           | None    | `str`   | If 'recompute', the engine performs preemption-aware recomputation. If 'save', the engine saves activations into the CPU memory as preemption happens. |
+| `PREEMPTION_CHECK_PERIOD`   | 1.0     | `float` | How frequently the engine checks if a preemption happens.                                                                                              |
+| `PREEMPTION_CPU_CAPACITY`   | 2       | `float` | The percentage of CPU memory used for the saved activations.                                                                                           |
+| `DISABLE_LOGGING_REQUEST`   | False   | `bool`  | Disable logging requests.                                                                                                                              |
+| `MAX_LOG_LEN`               | None    | `int`   | Max number of prompt characters or prompt ID numbers being printed in log.                                                                             |
+
+## Docker Build Arguments
+
+These variables are used when building custom Docker images with models baked in:
+
+| Variable              | Default          | Type  | Description                                       |
+| --------------------- | ---------------- | ----- | ------------------------------------------------- |
+| `BASE_PATH`           | `/runpod-volume` | `str` | Storage directory for huggingface cache and model |
+| `WORKER_CUDA_VERSION` | `12.1.0`         | `str` | CUDA version for the worker image                 |
+
+## Deprecated Variables
+
+⚠️ **The following variables are deprecated and will be removed in future versions:**
+
+| Old Variable                 | New Variable             | Note                  |
+| ---------------------------- | ------------------------ | --------------------- |
+| `MAX_CONTEXT_LEN_TO_CAPTURE` | `MAX_SEQ_LEN_TO_CAPTURE` | Use new variable name |
+| `kv_cache_dtype=fp8_e5m2`    | `kv_cache_dtype=fp8`     | Simplified fp8 format |

docs/conventions.md

@@ -0,0 +1,343 @@
+# Worker vLLM - Development Conventions & Architecture Guide
+
+## Project Overview
+
+**worker-vllm** is a RunPod serverless worker that provides OpenAI-compatible endpoints for Large Language Model (LLM) inference, powered by the vLLM engine. It enables blazing-fast LLM deployment on RunPod's serverless infrastructure with minimal configuration.
+
+### Core Purpose
+
+- **Primary Function**: Deploy any Hugging Face LLM as an OpenAI-compatible API endpoint
+- **Platform**: RunPod Serverless infrastructure
+- **Engine**: vLLM (high-performance LLM inference engine)
+- **Compatibility**: Drop-in replacement for OpenAI API (Chat Completions, Models)
+
+## High-Level Architecture
+
+### 1. **Entry Point & Request Flow**
+
+```
+RunPod Request → handler.py → JobInput → Engine Selection → vLLM Generation → Streaming Response
+```
+
+**Key Components:**
+
+- `src/handler.py`: Main entry point using RunPod serverless framework
+- `src/utils.py`: Request parsing and utility classes (`JobInput`, `BatchSize`)
+- Two engine modes: OpenAI-compatible vs. standard vLLM
+
+### 2. **Engine Architecture**
+
+#### Core Classes:
+
+- **`vLLMEngine`**: Base engine handling vLLM initialization and generation
+- **`OpenAIvLLMEngine`**: Wrapper providing OpenAI API compatibility
+- **Engine Selection**: Automatic routing based on `job_input.openai_route`
+
+#### Key Design Patterns:
+
+- **Dual API Support**: Same codebase serves both OpenAI-compatible and native vLLM APIs
+- **Streaming by Default**: Token-level streaming with configurable batching
+- **Dynamic Batching**: Adaptive batch sizes that grow from min → max for efficiency
+
+### 3. **Configuration System**
+
+#### Environment-Based Configuration:
+
+- **Single Source of Truth**: All configuration via environment variables
+- **Hierarchical Loading**: `DEFAULT_ARGS` → `os.environ` → `local_model_args.json` (for baked models)
+- **vLLM Argument Mapping**: Automatic translation of env vars to vLLM `AsyncEngineArgs`
+
+#### Key Configuration Files:
+
+- `src/engine_args.py`: Centralized configuration management
+- `src/constants.py`: Default values for core settings
+- `.runpod/hub.json`: Hub UI configuration (CRITICAL: always update when changing defaults)
+- `worker-config.json`: UI form generation for RunPod console (if exists)
+
+## Core Development Concepts
+
+### 1. **Deployment Models**
+
+#### Option 1: Pre-built Images (Recommended)
+
+- **Image**: `runpod/worker-v1-vllm:<version>` (see [GitHub Releases](https://github.com/runpod-workers/worker-vllm/releases))
+- **Configuration**: Entirely via environment variables
+- **Model Loading**: Downloads model at runtime from Hugging Face
+- **Use Case**: Quick deployment, model experimentation
+
+#### Option 2: Baked Model Images
+
+- **Build Process**: Model downloaded during Docker build
+- **Storage**: Model embedded in container image
+- **Configuration**: Stored in `/local_model_args.json`
+- **Use Case**: Production deployments, faster cold starts
+
+### 2. **Request Processing Patterns**
+
+#### Input Handling:
+
+```python
+class JobInput:
+    - llm_input: str | List[Dict] (prompt or messages)
+    - sampling_params: SamplingParams (generation settings)
+    - stream: bool (streaming vs batch response)
+    - openai_route: bool (API compatibility mode)
+    - batch_size configs: Dynamic batching parameters
+```
+
+#### Response Streaming:
+
+- **Batched Streaming**: Tokens grouped into configurable batch sizes
+- **Dynamic Growth**: `min_batch_size * growth_factor^n` up to `max_batch_size`
+- **Usage Tracking**: Input/output token counting for billing
+
+### 3. **Model & Tokenizer Management**
+
+#### Tokenizer Handling:
+
+- **Wrapper Pattern**: `TokenizerWrapper` for consistent chat template application
+- **Special Cases**: Mistral models use vLLM's native tokenizer
+- **Chat Templates**: Automatic application for message-based inputs
+
+#### Model Loading:
+
+- **Multi-GPU Support**: Automatic tensor parallelism detection
+- **Quantization**: Support for AWQ, GPTQ, BitsAndBytes
+- **Caching**: Hugging Face cache management
+
+## Development Patterns & Best Practices
+
+### 1. **Code Organization**
+
+#### File Structure:
+
+```
+src/
+├── handler.py          # RunPod entry point
+├── engine.py          # Core vLLM engines
+├── engine_args.py     # Configuration management
+├── utils.py           # Request parsing & utilities
+├── tokenizer.py       # Tokenizer wrapper
+├── constants.py       # Default constants
+└── download_model.py  # Model downloading logic
+```
+
+#### Separation of Concerns:
+
+- **Engine Logic**: Isolated in `engine.py` classes
+- **Configuration**: Centralized in `engine_args.py`
+- **Request Handling**: Abstracted via `JobInput` class
+- **Platform Integration**: Contained in `handler.py`
+
+### 2. **Error Handling & Logging**
+
+#### Logging Strategy:
+
+- **Structured Logging**: Consistent format across components
+- **Performance Tracking**: Timer decorators for critical operations
+- **Error Context**: Detailed error messages with configuration context
+
+#### Error Responses:
+
+- **OpenAI Compatibility**: Standard OpenAI error format
+- **Graceful Degradation**: Fallback behaviors for edge cases
+
+### 3. **Environment Variable Conventions**
+
+#### Naming Patterns:
+
+- **vLLM Settings**: Match vLLM parameter names (uppercase)
+- **RunPod Settings**: `MAX_CONCURRENCY`, `DEFAULT_BATCH_SIZE`
+- **OpenAI Settings**: `OPENAI_` prefix for compatibility settings
+- **Feature Flags**: `ENABLE_*`, `DISABLE_*` pattern
+
+#### Type Conventions:
+
+- **Booleans**: String 'true'/'false' or int 0/1
+- **Lists**: Comma-separated strings
+- **Objects**: JSON strings for complex configurations
+
+### 4. **Docker & Deployment**
+
+#### Multi-Stage Builds:
+
+- **Base**: CUDA runtime environment
+- **Dependencies**: Python packages and vLLM
+- **Model Download**: Optional model baking stage
+- **Runtime**: Final application layer
+
+#### Build Arguments:
+
+- **MODEL_NAME**: Primary model identifier
+- **BASE_PATH**: Storage location strategy
+- **QUANTIZATION**: Optimization settings
+- **WORKER_CUDA_VERSION**: CUDA compatibility
+
+#### CI/CD Strategy:
+
+- **Development Builds**: All non-main branches → `runpod/worker-v1-vllm:dev-<branch-name>`
+- **Release Builds**: Git tags (numeric) only → `runpod/worker-v1-vllm:<version>`
+- **Dependency Updates**: Automated runpod package version monitoring
+
+#### Docker Bake Configuration:
+
+- **File**: `docker-bake.hcl` (flexible variable-based configuration)
+- **Variables**: `DOCKERHUB_REPO`, `DOCKERHUB_IMG`, `RELEASE_VERSION`, `HUGGINGFACE_ACCESS_TOKEN`
+- **Platform**: `linux/amd64` (GPU-optimized)
+
+## Release & Versioning Strategy
+
+### 1. **Version Tagging**
+
+- **Development**: `dev-<branch-name>` (e.g., `dev-feature-new-api`)
+- **Specific Versions**: `2.7.0`, `2.8.0` (semantic versioning without "v" prefix)
+- **Version Discovery**: Check [GitHub Releases](https://github.com/runpod-workers/worker-vllm/releases) for available versions
+
+### 2. **Release Workflow**
+
+1. **Feature Development**: Work on feature branches → triggers dev builds
+2. **Main Branch Staging**: Merge features to main → stable codebase (no builds)
+3. **Version Release**: Create git tag from main branch (e.g., `2.8.0`) → triggers versioned release + GitHub release
+4. **Docker Hub**: Versioned image pushed with tag
+
+### 3. **Branch Strategy**
+
+- **Feature Branches**: `feature/*`, `fix/*`, `feat/*` etc. → Dev builds
+- **Main Branch**: Stable codebase ready for release (no automatic builds)
+- **Git Tags**: Must be created from main branch for formal version releases
+
+### 4. **Deployment Recommendations**
+
+- **Production**: Use specific version tags (e.g., `2.7.0`) for stability
+- **Development**: Use `dev-<branch>` for testing specific features
+- **Version Selection**: Check [GitHub Releases](https://github.com/runpod-workers/worker-vllm/releases) for available versions
+- **Release Process**: Always tag from main branch: `git checkout main && git tag 2.8.0 && git push origin 2.8.0`
+
+## Performance & Scaling Considerations
+
+### 1. **Memory Management**
+
+- **GPU Utilization**: Default 95% GPU memory utilization
+- **KV Cache**: Configurable cache types (auto, fp8)
+- **Swap Space**: CPU offloading for large contexts
+
+### 2. **Concurrency Patterns**
+
+- **Max Concurrency**: 30 concurrent requests by default
+- **vLLM Queuing**: Internal request batching and scheduling
+- **RunPod Integration**: Concurrency modifier for auto-scaling
+
+### 3. **Optimization Features**
+
+- **Prefix Caching**: Automatic caching of common prefixes
+- **Speculative Decoding**: Draft model acceleration
+- **Chunked Prefill**: Memory-efficient long context handling
+
+## Testing & Development
+
+### 1. **Local Development**
+
+- **Environment**: Virtual environment with GPU support
+- **Configuration**: `.env` files for local testing
+- **Model Testing**: Small models for development (facebook/opt-125m)
+
+### 2. **Docker Development**
+
+- **Build Strategy**: `docker-bake.hcl` for consistent builds
+- **Testing Images**: Separate dev/stable image tags
+- **Layer Caching**: Optimized for rapid iteration
+
+### 3. **Configuration Validation**
+
+- **Argument Matching**: Automatic validation against vLLM parameters
+- **Environment Validation**: Type checking and default value handling
+- **Runtime Validation**: Model compatibility checks
+
+## API Conventions
+
+### 1. **OpenAI Compatibility**
+
+- **Endpoint Mapping**: `/openai/v1/chat/completions`, `/openai/v1/models`
+- **Request Format**: Exact OpenAI request/response schemas
+- **Authentication**: RunPod API key in Authorization header
+- **Model Names**: Hugging Face repo names or custom overrides
+
+### 2. **Native vLLM API**
+
+- **Input Format**: `prompt` or `messages` with `sampling_params`
+- **Streaming**: Token-level streaming with configurable batching
+- **Extensibility**: Support for vLLM-specific features
+
+## Common Patterns & Utilities
+
+### 1. **Configuration Loading**
+
+```python
+# Standard pattern for new configuration options
+def get_engine_args():
+    args = DEFAULT_ARGS
+    args.update(os.environ)  # Environment override
+    args.update(get_local_args())  # Baked model override
+    return match_vllm_args(args)  # Validate against vLLM
+```
+
+### 2. **Error Handling**
+
+```python
+# Standard error response pattern
+def create_error_response(message: str, err_type: str = "BadRequestError"):
+    return ErrorResponse(message=message, type=err_type)
+```
+
+### 3. **Async Generation**
+
+```python
+# Standard streaming pattern
+async def generate(self, job_input: JobInput):
+    async for batch in self._generate_vllm(...):
+        yield batch  # Batch-level yielding for efficiency
+```
+
+## Extension Points
+
+### 1. **New Model Architectures**
+
+- **Engine Args**: Add new parameters in `engine_args.py`
+- **Compatibility**: Update vLLM argument mapping
+- **Validation**: Add architecture-specific validation
+
+### 2. **New API Features**
+
+- **Engine Extension**: Extend `vLLMEngine` or `OpenAIvLLMEngine`
+- **Input Parsing**: Extend `JobInput` class
+- **Response Format**: Add new response generators
+
+### 3. **Performance Optimizations**
+
+- **Batching Strategy**: Modify `BatchSize` class
+- **Memory Management**: Add new caching strategies
+- **Hardware Optimization**: GPU-specific optimizations
+
+## Security & Best Practices
+
+### 1. **Secret Management**
+
+- **Build Secrets**: Docker secrets for HF tokens
+- **Runtime Secrets**: Environment variable injection
+- **Token Handling**: Secure authentication patterns
+
+### 2. **Resource Limits**
+
+- **Memory Bounds**: Configurable GPU memory limits
+- **Request Limits**: Concurrency and timeout controls
+- **Model Safety**: Trust remote code flags
+
+### 3. **Logging Security**
+
+- **Sanitization**: No secrets in logs
+- **Request Logging**: Configurable request/response logging
+- **Performance Monitoring**: Safe metrics collection
+
+---
+
+This guide should be consulted whenever working on the worker-vllm codebase to ensure consistency with established patterns and architectural decisions.

src/config.py

@@ -1,57 +0,0 @@
-import os
-import json
-import logging
-from dotenv import load_dotenv
-from torch.cuda import device_count
-from utils import get_int_bool_env
-
-class EngineConfig:
-    def __init__(self):
-        load_dotenv()
-        self.hf_home = os.getenv("HF_HOME")
-        # Check if /local_metadata.json exists
-        local_metadata = {}
-        if os.path.exists("/local_metadata.json"):
-            with open("/local_metadata.json", "r") as f:
-                local_metadata = json.load(f)
-            if local_metadata.get("model_name") is None:
-                raise ValueError("Model name is not found in /local_metadata.json, there was a problem when you baked the model in.")
-            logging.info("Using baked-in model")
-            os.environ["TRANSFORMERS_OFFLINE"] = "1"
-            os.environ["HF_HUB_OFFLINE"] = "1"
-            
-        self.model_name_or_path = local_metadata.get("model_name", os.getenv("MODEL_NAME"))
-        self.model_revision = local_metadata.get("revision", os.getenv("MODEL_REVISION")) 
-        self.tokenizer_name_or_path = local_metadata.get("tokenizer_name", os.getenv("TOKENIZER_NAME")) or self.model_name_or_path
-        self.tokenizer_revision = local_metadata.get("tokenizer_revision", os.getenv("TOKENIZER_REVISION"))  
-        self.quantization = local_metadata.get("quantization", os.getenv("QUANTIZATION"))
-        self.config = self._initialize_config()
-    def _initialize_config(self):
-        args = {
-            "model": self.model_name_or_path,
-            "revision": self.model_revision,
-            "download_dir": self.hf_home,
-            "quantization": self.quantization,
-            "load_format": os.getenv("LOAD_FORMAT", "auto"),
-            "dtype": os.getenv("DTYPE", "half" if self.quantization else "auto"),
-            "tokenizer": self.tokenizer_name_or_path,
-            "tokenizer_revision": self.tokenizer_revision,
-            "disable_log_stats": get_int_bool_env("DISABLE_LOG_STATS", True),
-            "disable_log_requests": get_int_bool_env("DISABLE_LOG_REQUESTS", True),
-            "trust_remote_code": get_int_bool_env("TRUST_REMOTE_CODE", False),
-            "gpu_memory_utilization": float(os.getenv("GPU_MEMORY_UTILIZATION", 0.95)),
-            "max_parallel_loading_workers": None if device_count() > 1 or not os.getenv("MAX_PARALLEL_LOADING_WORKERS") else int(os.getenv("MAX_PARALLEL_LOADING_WORKERS")),
-            "max_model_len": int(os.getenv("MAX_MODEL_LEN")) if os.getenv("MAX_MODEL_LEN") else None,
-            "tensor_parallel_size": device_count(),
-            "seed": int(os.getenv("SEED")) if os.getenv("SEED") else None,
-            "kv_cache_dtype": os.getenv("KV_CACHE_DTYPE"),
-            "block_size": int(os.getenv("BLOCK_SIZE")) if os.getenv("BLOCK_SIZE") else None,
-            "swap_space": int(os.getenv("SWAP_SPACE")) if os.getenv("SWAP_SPACE") else None,
-            "max_context_len_to_capture": int(os.getenv("MAX_CONTEXT_LEN_TO_CAPTURE")) if os.getenv("MAX_CONTEXT_LEN_TO_CAPTURE") else None,
-            "disable_custom_all_reduce": get_int_bool_env("DISABLE_CUSTOM_ALL_REDUCE", False),
-            "enforce_eager": get_int_bool_env("ENFORCE_EAGER", False)
-        }
-        if args["kv_cache_dtype"] == "fp8_e5m2":
-            args["kv_cache_dtype"] = "fp8"
-            logging.warning("Using fp8_e5m2 is deprecated. Please use fp8 instead.")
-        return {k: v for k, v in args.items() if v not in [None, ""]}

src/constants.py

@@ -1,4 +1,4 @@
 DEFAULT_BATCH_SIZE = 50
-DEFAULT_MAX_CONCURRENCY = 300
+DEFAULT_MAX_CONCURRENCY = 30
 DEFAULT_BATCH_SIZE_GROWTH_FACTOR = 3
 DEFAULT_MIN_BATCH_SIZE = 1

src/download_model.py

@@ -1,27 +1,100 @@
 import os
-from huggingface_hub import snapshot_download
 import json
+import logging
+import glob
+from shutil import rmtree
+from huggingface_hub import snapshot_download
+from utils import timer_decorator
+
+BASE_DIR = "/" 
+TOKENIZER_PATTERNS = [["*.json", "tokenizer*"]]
+MODEL_PATTERNS = [["*.safetensors"], ["*.bin"], ["*.pt"]]
+
+def setup_env():
+    if os.getenv("TESTING_DOWNLOAD") == "1":
+        BASE_DIR = "tmp"
+        os.makedirs(BASE_DIR, exist_ok=True)
+        os.environ.update({
+            "HF_HOME": f"{BASE_DIR}/hf_cache",
+            "MODEL_NAME": "openchat/openchat-3.5-0106",
+            "HF_HUB_ENABLE_HF_TRANSFER": "1",
+            "TENSORIZE": "1",
+            "TENSORIZER_NUM_GPUS": "1",
+            "DTYPE": "auto"
+        })
+
+@timer_decorator
+def download(name, revision, type, cache_dir):
+    if type == "model":
+        pattern_sets = [model_pattern + TOKENIZER_PATTERNS[0] for model_pattern in MODEL_PATTERNS]
+    elif type == "tokenizer":
+        pattern_sets = TOKENIZER_PATTERNS
+    else:
+        raise ValueError(f"Invalid type: {type}")
+    try:
+        for pattern_set in pattern_sets:
+            path = snapshot_download(name, revision=revision, cache_dir=cache_dir, 
+                                    allow_patterns=pattern_set)
+            for pattern in pattern_set:
+                if glob.glob(os.path.join(path, pattern)):
+                    logging.info(f"Successfully downloaded {pattern} model files.")
+                    return path
+    except ValueError:
+        raise ValueError(f"No patterns matching {pattern_sets} found for download.")
+        
+          
+# @timer_decorator
+# def tensorize_model(model_path): TODO: Add back once tensorizer is ready
+#     from vllm.engine.arg_utils import EngineArgs
+#     from vllm.model_executor.model_loader.tensorizer import TensorizerConfig, tensorize_vllm_model
+#     from torch.cuda import device_count
+
+#     tensorizer_num_gpus = int(os.getenv("TENSORIZER_NUM_GPUS", "1"))
+#     if tensorizer_num_gpus > device_count():
+#         raise ValueError(f"TENSORIZER_NUM_GPUS ({tensorizer_num_gpus}) exceeds available GPUs ({device_count()})")
+
+#     dtype = os.getenv("DTYPE", "auto")
+#     serialized_dir = f"{BASE_DIR}/serialized_model"
+#     os.makedirs(serialized_dir, exist_ok=True)
+#     serialized_uri = f"{serialized_dir}/model{'-%03d' if tensorizer_num_gpus > 1 else ''}.tensors"
+    
+#     tensorize_vllm_model(
+#         EngineArgs(model=model_path, tensor_parallel_size=tensorizer_num_gpus, dtype=dtype),
+#         TensorizerConfig(tensorizer_uri=serialized_uri)
+#     )
+#     logging.info("Successfully serialized model to %s", str(serialized_uri))
+#     logging.info("Removing HF Model files after serialization")
+#     rmtree("/".join(model_path.split("/")[:-2]))
+#     return serialized_uri, tensorizer_num_gpus, dtype
 
 if __name__ == "__main__":
-    model_name = os.getenv("MODEL_NAME")
+    setup_env()
-    if not model_name:
+    cache_dir = os.getenv("HF_HOME")
-        raise ValueError("Must specify model name by adding --build-arg MODEL_NAME=<your model's repo>")
+    model_name, model_revision = os.getenv("MODEL_NAME"), os.getenv("MODEL_REVISION") or None
-    revision = os.getenv("MODEL_REVISION") or None
+    tokenizer_name, tokenizer_revision = os.getenv("TOKENIZER_NAME") or model_name, os.getenv("TOKENIZER_REVISION") or model_revision
-    snapshot_download(model_name, revision=revision, cache_dir=os.getenv("HF_HOME"))
    
-    tokenizer_name = os.getenv("TOKENIZER_NAME") or None
+    model_path = download(model_name, model_revision, "model", cache_dir)   
-    tokenizer_revision = os.getenv("TOKENIZER_REVISION") or None
-    if tokenizer_name:  
-        snapshot_download(tokenizer_name, revision=tokenizer_revision, cache_dir=os.getenv("HF_HOME"))
   
-    # Create file with metadata of baked in model and/or tokenizer
+    metadata = {
+        "MODEL_NAME": model_path,
+        "MODEL_REVISION": os.getenv("MODEL_REVISION"),
+        "QUANTIZATION": os.getenv("QUANTIZATION"),
+    }   
     
-    with open("/local_metadata.json", "w") as f:
+    # if os.getenv("TENSORIZE") == "1": TODO: Add back once tensorizer is ready
-        json.dump({
+    #     serialized_uri, tensorizer_num_gpus, dtype = tensorize_model(model_path)
-            "model_name": model_name,
+    #     metadata.update({
-            "revision": revision,
+    #         "MODEL_NAME": serialized_uri,
-            "tokenizer_name": tokenizer_name or model_name,
+    #         "TENSORIZER_URI": serialized_uri,
-            "tokenizer_revision": tokenizer_revision or revision,
+    #         "TENSOR_PARALLEL_SIZE": tensorizer_num_gpus,
-            "quantization": os.getenv("QUANTIZATION")
+    #         "DTYPE": dtype
-        }, f)
+    #     })
         
+    tokenizer_path = download(tokenizer_name, tokenizer_revision, "tokenizer", cache_dir)
+    metadata.update({
+        "TOKENIZER_NAME": tokenizer_path,
+        "TOKENIZER_REVISION": tokenizer_revision
+    })
+    
+    with open(f"{BASE_DIR}/local_model_args.json", "w") as f:
+        json.dump({k: v for k, v in metadata.items() if v not in (None, "")}, f)

src/engine.py

@@ -1,33 +1,92 @@
 import os
 import logging
 import json
+import asyncio
 
 from dotenv import load_dotenv
-from torch.cuda import device_count
+from typing import AsyncGenerator, Optional
-from typing import AsyncGenerator
 import time
 
-from vllm import AsyncLLMEngine, AsyncEngineArgs
+from vllm import AsyncLLMEngine
+from vllm.entrypoints.logger import RequestLogger
 from vllm.entrypoints.openai.serving_chat import OpenAIServingChat
 from vllm.entrypoints.openai.serving_completion import OpenAIServingCompletion
 from vllm.entrypoints.openai.protocol import ChatCompletionRequest, CompletionRequest, ErrorResponse
+from vllm.entrypoints.openai.serving_models import BaseModelPath, LoRAModulePath, OpenAIServingModels
+
 
 from utils import DummyRequest, JobInput, BatchSize, create_error_response
 from constants import DEFAULT_MAX_CONCURRENCY, DEFAULT_BATCH_SIZE, DEFAULT_BATCH_SIZE_GROWTH_FACTOR, DEFAULT_MIN_BATCH_SIZE
 from tokenizer import TokenizerWrapper
-from config import EngineConfig
+from engine_args import get_engine_args
 
 class vLLMEngine:
     def __init__(self, engine = None):
         load_dotenv() # For local development
-        self.config = EngineConfig().config
+        self.engine_args = get_engine_args()
-        self.tokenizer = TokenizerWrapper(self.config.get("tokenizer"), self.config.get("tokenizer_revision"), self.config.get("trust_remote_code"))
+        logging.info(f"Engine args: {self.engine_args}")
-        self.llm = self._initialize_llm() if engine is None else engine
+        
+        # Initialize vLLM engine first
+        self.llm = self._initialize_llm() if engine is None else engine.llm
+        
+        # Only create custom tokenizer wrapper if not using mistral tokenizer mode
+        # For mistral models, let vLLM handle tokenizer initialization
+        if self.engine_args.tokenizer_mode != 'mistral':
+            self.tokenizer = TokenizerWrapper(self.engine_args.tokenizer or self.engine_args.model, 
+                                              self.engine_args.tokenizer_revision, 
+                                              self.engine_args.trust_remote_code)
+        else:
+            # For mistral models, we'll get the tokenizer from vLLM later
+            self.tokenizer = None
+            
         self.max_concurrency = int(os.getenv("MAX_CONCURRENCY", DEFAULT_MAX_CONCURRENCY))
         self.default_batch_size = int(os.getenv("DEFAULT_BATCH_SIZE", DEFAULT_BATCH_SIZE))
         self.batch_size_growth_factor = int(os.getenv("BATCH_SIZE_GROWTH_FACTOR", DEFAULT_BATCH_SIZE_GROWTH_FACTOR))
         self.min_batch_size = int(os.getenv("MIN_BATCH_SIZE", DEFAULT_MIN_BATCH_SIZE))
 
+    def _get_tokenizer_for_chat_template(self):
+        """Get tokenizer for chat template application"""
+        if self.tokenizer is not None:
+            return self.tokenizer
+        else:
+            # For mistral models, get tokenizer from vLLM engine
+            # This is a fallback - ideally chat templates should be handled by vLLM directly
+            try:
+                from transformers import AutoTokenizer
+                tokenizer = AutoTokenizer.from_pretrained(
+                    self.engine_args.tokenizer or self.engine_args.model,
+                    revision=self.engine_args.tokenizer_revision or "main",
+                    trust_remote_code=self.engine_args.trust_remote_code
+                )
+                # Create a minimal wrapper
+                class MinimalTokenizerWrapper:
+                    def __init__(self, tokenizer):
+                        self.tokenizer = tokenizer
+                        self.custom_chat_template = os.getenv("CUSTOM_CHAT_TEMPLATE")
+                        self.has_chat_template = bool(self.tokenizer.chat_template) or bool(self.custom_chat_template)
+                        if self.custom_chat_template and isinstance(self.custom_chat_template, str):
+                            self.tokenizer.chat_template = self.custom_chat_template
+                    
+                    def apply_chat_template(self, input):
+                        if isinstance(input, list):
+                            if not self.has_chat_template:
+                                raise ValueError(
+                                    "Chat template does not exist for this model, you must provide a single string input instead of a list of messages"
+                                )
+                        elif isinstance(input, str):
+                            input = [{"role": "user", "content": input}]
+                        else:
+                            raise ValueError("Input must be a string or a list of messages")
+                        
+                        return self.tokenizer.apply_chat_template(
+                            input, tokenize=False, add_generation_prompt=True
+                        )
+                
+                return MinimalTokenizerWrapper(tokenizer)
+            except Exception as e:
+                logging.error(f"Failed to create fallback tokenizer: {e}")
+                raise e
+
     def dynamic_batch_size(self, current_batch_size, batch_size_growth_factor):
         return min(current_batch_size*batch_size_growth_factor, self.default_batch_size)
                            
@@ -49,7 +108,8 @@ class vLLMEngine:
 
     async def _generate_vllm(self, llm_input, validated_sampling_params, batch_size, stream, apply_chat_template, request_id, batch_size_growth_factor, min_batch_size: str) -> AsyncGenerator[dict, None]:
         if apply_chat_template or isinstance(llm_input, list):
-            llm_input = self.tokenizer.apply_chat_template(llm_input)
+            tokenizer_wrapper = self._get_tokenizer_for_chat_template()
+            llm_input = tokenizer_wrapper.apply_chat_template(llm_input)
         results_generator = self.llm.generate(llm_input, validated_sampling_params, request_id)
         n_responses, n_input_tokens, is_first_output = validated_sampling_params.n, 0, True
         last_output_texts, token_counters = ["" for _ in range(n_responses)], {"batch": 0, "total": 0}
@@ -102,7 +162,7 @@ class vLLMEngine:
     def _initialize_llm(self):
         try:
             start = time.time()
-            engine = AsyncLLMEngine.from_engine_args(AsyncEngineArgs(**self.config))
+            engine = AsyncLLMEngine.from_engine_args(self.engine_args)
             end = time.time()
             logging.info(f"Initialized vLLM engine in {end - start:.2f}s")
             return engine
@@ -111,24 +171,77 @@ class vLLMEngine:
             raise e
 
 
-class OpenAIvLLMEngine:
+class OpenAIvLLMEngine(vLLMEngine):
     def __init__(self, vllm_engine):
-        self.config = vllm_engine.config
+        super().__init__(vllm_engine)
-        self.llm = vllm_engine.llm
+        self.served_model_name = os.getenv("OPENAI_SERVED_MODEL_NAME_OVERRIDE") or self.engine_args.model
-        self.served_model_name = os.getenv("OPENAI_SERVED_MODEL_NAME_OVERRIDE") or self.config["model"]
         self.response_role = os.getenv("OPENAI_RESPONSE_ROLE") or "assistant"
-        self.tokenizer = vllm_engine.tokenizer
+        self.lora_adapters = self._load_lora_adapters()
-        self.default_batch_size = vllm_engine.default_batch_size
+        asyncio.run(self._initialize_engines())
-        self.batch_size_growth_factor, self.min_batch_size = vllm_engine.batch_size_growth_factor, vllm_engine.min_batch_size
+        # Handle both integer and boolean string values for RAW_OPENAI_OUTPUT
-        self._initialize_engines()
+        raw_output_env = os.getenv("RAW_OPENAI_OUTPUT", "1")
-        self.raw_openai_output = bool(int(os.getenv("RAW_OPENAI_OUTPUT", 1)))
+        if raw_output_env.lower() in ('true', 'false'):
+            self.raw_openai_output = raw_output_env.lower() == 'true'
+        else:
+            self.raw_openai_output = bool(int(raw_output_env))
 
-    def _initialize_engines(self):
+    def _load_lora_adapters(self):
-        self.chat_engine = OpenAIServingChat(
+        adapters = []
-            self.llm, self.served_model_name, self.response_role,
+        try:
-            chat_template=self.tokenizer.tokenizer.chat_template
+            adapters = json.loads(os.getenv("LORA_MODULES", '[]'))
+        except Exception as e:
+            logging.info(f"---Initialized adapter json load error: {e}")
+
+        for i, adapter in enumerate(adapters):
+            try:
+                adapters[i] = LoRAModulePath(**adapter)
+                logging.info(f"---Initialized adapter: {adapter}")
+            except Exception as e:
+                logging.info(f"---Initialized adapter not worked: {e}")
+                continue
+        return adapters
+
+    async def _initialize_engines(self):
+        self.model_config = await self.llm.get_model_config()
+        self.base_model_paths = [
+            BaseModelPath(name=self.engine_args.model, model_path=self.engine_args.model)
+        ]
+
+        self.serving_models = OpenAIServingModels(
+            engine_client=self.llm,
+            model_config=self.model_config,
+            base_model_paths=self.base_model_paths,
+            lora_modules=self.lora_adapters,
+        )
+        await self.serving_models.init_static_loras()
+        
+        # Get chat template from vLLM tokenizer if available
+        chat_template = None
+        if self.tokenizer and hasattr(self.tokenizer, 'tokenizer'):
+            chat_template = self.tokenizer.tokenizer.chat_template
+        
+        self.chat_engine = OpenAIServingChat(
+            engine_client=self.llm, 
+            model_config=self.model_config,
+            models=self.serving_models,
+            response_role=self.response_role,
+            request_logger=None,
+            chat_template=chat_template,
+            chat_template_content_format="auto",
+            # enable_reasoning=os.getenv('ENABLE_REASONING', 'false').lower() == 'true',
+            reasoning_parser= os.getenv('REASONING_PARSER', "") or None,
+            # return_token_as_token_ids=False,
+            enable_auto_tools=os.getenv('ENABLE_AUTO_TOOL_CHOICE', 'false').lower() == 'true',
+            tool_parser=os.getenv('TOOL_CALL_PARSER', "") or None,
+            enable_prompt_tokens_details=False
+        )
+        self.completion_engine = OpenAIServingCompletion(
+            engine_client=self.llm, 
+            model_config=self.model_config,
+            models=self.serving_models,
+            request_logger=None,
+            # return_token_as_token_ids=False,
         )
-        self.completion_engine = OpenAIServingCompletion(self.llm, self.served_model_name)
     
     async def generate(self, openai_request: JobInput):
         if openai_request.openai_route == "/v1/models":
@@ -140,10 +253,7 @@ class OpenAIvLLMEngine:
             yield create_error_response("Invalid route").model_dump()
     
     async def _handle_model_request(self):
-        models = await self.chat_engine.show_available_models()
+        models = await self.serving_models.show_available_models()
-        fixed_model = models.data[0]
-        fixed_model.id = self.served_model_name
-        models.data = [fixed_model]
         return models.model_dump()
     
     async def _handle_chat_or_completion_request(self, openai_request: JobInput):
@@ -162,7 +272,8 @@ class OpenAIvLLMEngine:
             yield create_error_response(str(e)).model_dump()
             return
         
-        response_generator = await generator_function(request, DummyRequest())
+        dummy_request = DummyRequest()
+        response_generator = await generator_function(request, raw_request=dummy_request)
 
         if not openai_request.openai_input.get("stream") or isinstance(response_generator, ErrorResponse):
             yield response_generator.model_dump()

src/engine_args.py

@@ -0,0 +1,178 @@
+import os
+import json
+import logging
+from torch.cuda import device_count
+from vllm import AsyncEngineArgs
+from vllm.model_executor.model_loader.tensorizer import TensorizerConfig
+from src.utils import convert_limit_mm_per_prompt
+
+RENAME_ARGS_MAP = {
+    "MODEL_NAME": "model",
+    "MODEL_REVISION": "revision",
+    "TOKENIZER_NAME": "tokenizer",
+    "MAX_CONTEXT_LEN_TO_CAPTURE": "max_seq_len_to_capture"
+}
+
+DEFAULT_ARGS = {
+    "disable_log_stats": os.getenv('DISABLE_LOG_STATS', 'False').lower() == 'true',
+    "disable_log_requests": os.getenv('DISABLE_LOG_REQUESTS', 'False').lower() == 'true',
+    "gpu_memory_utilization": float(os.getenv('GPU_MEMORY_UTILIZATION', 0.95)),
+    "pipeline_parallel_size": int(os.getenv('PIPELINE_PARALLEL_SIZE', 1)),
+    "tensor_parallel_size": int(os.getenv('TENSOR_PARALLEL_SIZE', 1)),
+    "served_model_name": os.getenv('SERVED_MODEL_NAME', None),
+    "tokenizer": os.getenv('TOKENIZER', None),
+    "skip_tokenizer_init": os.getenv('SKIP_TOKENIZER_INIT', 'False').lower() == 'true',
+    "tokenizer_mode": os.getenv('TOKENIZER_MODE', 'auto'),
+    "trust_remote_code": os.getenv('TRUST_REMOTE_CODE', 'False').lower() == 'true',
+    "download_dir": os.getenv('DOWNLOAD_DIR', None),
+    "load_format": os.getenv('LOAD_FORMAT', 'auto'),
+    "config_format": os.getenv('CONFIG_FORMAT', 'auto'),
+    "dtype": os.getenv('DTYPE', 'auto'),
+    "kv_cache_dtype": os.getenv('KV_CACHE_DTYPE', 'auto'),
+    "quantization_param_path": os.getenv('QUANTIZATION_PARAM_PATH', None),
+    "seed": int(os.getenv('SEED', 0)),
+    "max_model_len": int(os.getenv('MAX_MODEL_LEN', 0)) or None,
+    "worker_use_ray": os.getenv('WORKER_USE_RAY', 'False').lower() == 'true',
+    "distributed_executor_backend": os.getenv('DISTRIBUTED_EXECUTOR_BACKEND', None),
+    "max_parallel_loading_workers": int(os.getenv('MAX_PARALLEL_LOADING_WORKERS', 0)) or None,
+    "block_size": int(os.getenv('BLOCK_SIZE', 16)),
+    "enable_prefix_caching": os.getenv('ENABLE_PREFIX_CACHING', 'False').lower() == 'true',
+    "disable_sliding_window": os.getenv('DISABLE_SLIDING_WINDOW', 'False').lower() == 'true',
+    "use_v2_block_manager": os.getenv('USE_V2_BLOCK_MANAGER', 'False').lower() == 'true',
+    "swap_space": int(os.getenv('SWAP_SPACE', 4)),  # GiB
+    "cpu_offload_gb": int(os.getenv('CPU_OFFLOAD_GB', 0)),  # GiB
+    "max_num_batched_tokens": int(os.getenv('MAX_NUM_BATCHED_TOKENS', 0)) or None,
+    "max_num_seqs": int(os.getenv('MAX_NUM_SEQS', 256)),
+    "max_logprobs": int(os.getenv('MAX_LOGPROBS', 20)),  # Default value for OpenAI Chat Completions API
+    "revision": os.getenv('REVISION', None),
+    "code_revision": os.getenv('CODE_REVISION', None),
+    "rope_scaling": os.getenv('ROPE_SCALING', None),
+    "rope_theta": float(os.getenv('ROPE_THETA', 0)) or None,
+    "tokenizer_revision": os.getenv('TOKENIZER_REVISION', None),
+    "quantization": os.getenv('QUANTIZATION', None),
+    "enforce_eager": os.getenv('ENFORCE_EAGER', 'False').lower() == 'true',
+    "max_context_len_to_capture": int(os.getenv('MAX_CONTEXT_LEN_TO_CAPTURE', 0)) or None,
+    "max_seq_len_to_capture": int(os.getenv('MAX_SEQ_LEN_TO_CAPTURE', 8192)),
+    "disable_custom_all_reduce": os.getenv('DISABLE_CUSTOM_ALL_REDUCE', 'False').lower() == 'true',
+    "tokenizer_pool_size": int(os.getenv('TOKENIZER_POOL_SIZE', 0)),
+    "tokenizer_pool_type": os.getenv('TOKENIZER_POOL_TYPE', 'ray'),
+    "tokenizer_pool_extra_config": os.getenv('TOKENIZER_POOL_EXTRA_CONFIG', None),
+    "enable_lora": os.getenv('ENABLE_LORA', 'False').lower() == 'true',
+    "max_loras": int(os.getenv('MAX_LORAS', 1)),
+    "max_lora_rank": int(os.getenv('MAX_LORA_RANK', 16)),
+    "enable_prompt_adapter": os.getenv('ENABLE_PROMPT_ADAPTER', 'False').lower() == 'true',
+    "max_prompt_adapters": int(os.getenv('MAX_PROMPT_ADAPTERS', 1)),
+    "max_prompt_adapter_token": int(os.getenv('MAX_PROMPT_ADAPTER_TOKEN', 0)),
+    "fully_sharded_loras": os.getenv('FULLY_SHARDED_LORAS', 'False').lower() == 'true',
+    "lora_extra_vocab_size": int(os.getenv('LORA_EXTRA_VOCAB_SIZE', 256)),
+    "long_lora_scaling_factors": tuple(map(float, os.getenv('LONG_LORA_SCALING_FACTORS', '').split(','))) if os.getenv('LONG_LORA_SCALING_FACTORS') else None,
+    "lora_dtype": os.getenv('LORA_DTYPE', 'auto'),
+    "max_cpu_loras": int(os.getenv('MAX_CPU_LORAS', 0)) or None,
+    "device": os.getenv('DEVICE', 'auto'),
+    "ray_workers_use_nsight": os.getenv('RAY_WORKERS_USE_NSIGHT', 'False').lower() == 'true',
+    "num_gpu_blocks_override": int(os.getenv('NUM_GPU_BLOCKS_OVERRIDE', 0)) or None,
+    "num_lookahead_slots": int(os.getenv('NUM_LOOKAHEAD_SLOTS', 0)),
+    "model_loader_extra_config": os.getenv('MODEL_LOADER_EXTRA_CONFIG', None),
+    "ignore_patterns": os.getenv('IGNORE_PATTERNS', None),
+    "preemption_mode": os.getenv('PREEMPTION_MODE', None),
+    "scheduler_delay_factor": float(os.getenv('SCHEDULER_DELAY_FACTOR', 0.0)),
+    "enable_chunked_prefill": os.getenv('ENABLE_CHUNKED_PREFILL', None),
+    "guided_decoding_backend": os.getenv('GUIDED_DECODING_BACKEND', 'outlines'),
+    "speculative_model": os.getenv('SPECULATIVE_MODEL', None),
+    "speculative_draft_tensor_parallel_size": int(os.getenv('SPECULATIVE_DRAFT_TENSOR_PARALLEL_SIZE', 0)) or None,
+    "num_speculative_tokens": int(os.getenv('NUM_SPECULATIVE_TOKENS', 0)) or None,
+    "speculative_max_model_len": int(os.getenv('SPECULATIVE_MAX_MODEL_LEN', 0)) or None,
+    "speculative_disable_by_batch_size": int(os.getenv('SPECULATIVE_DISABLE_BY_BATCH_SIZE', 0)) or None,
+    "ngram_prompt_lookup_max": int(os.getenv('NGRAM_PROMPT_LOOKUP_MAX', 0)) or None,
+    "ngram_prompt_lookup_min": int(os.getenv('NGRAM_PROMPT_LOOKUP_MIN', 0)) or None,
+    "spec_decoding_acceptance_method": os.getenv('SPEC_DECODING_ACCEPTANCE_METHOD', 'rejection_sampler'),
+    "typical_acceptance_sampler_posterior_threshold": float(os.getenv('TYPICAL_ACCEPTANCE_SAMPLER_POSTERIOR_THRESHOLD', 0)) or None,
+    "typical_acceptance_sampler_posterior_alpha": float(os.getenv('TYPICAL_ACCEPTANCE_SAMPLER_POSTERIOR_ALPHA', 0)) or None,
+    "qlora_adapter_name_or_path": os.getenv('QLORA_ADAPTER_NAME_OR_PATH', None),
+    "disable_logprobs_during_spec_decoding": os.getenv('DISABLE_LOGPROBS_DURING_SPEC_DECODING', None),
+    "otlp_traces_endpoint": os.getenv('OTLP_TRACES_ENDPOINT', None),
+    "use_v2_block_manager": os.getenv('USE_V2_BLOCK_MANAGER', 'true'),
+}
+limit_mm_env = os.getenv('LIMIT_MM_PER_PROMPT')
+if limit_mm_env is not None:
+    DEFAULT_ARGS["limit_mm_per_prompt"] = convert_limit_mm_per_prompt(limit_mm_env)
+
+def match_vllm_args(args):
+    """Rename args to match vllm by:
+    1. Renaming keys to lower case
+    2. Renaming keys to match vllm
+    3. Filtering args to match vllm's AsyncEngineArgs
+
+    Args:
+        args (dict): Dictionary of args
+
+    Returns:
+        dict: Dictionary of args with renamed keys
+    """
+    renamed_args = {RENAME_ARGS_MAP.get(k, k): v for k, v in args.items()}
+    matched_args = {k: v for k, v in renamed_args.items() if k in AsyncEngineArgs.__dataclass_fields__}
+    return {k: v for k, v in matched_args.items() if v not in [None, "", "None"]}
+def get_local_args():
+    """
+    Retrieve local arguments from a JSON file.
+
+    Returns:
+        dict: Local arguments.
+    """
+    if not os.path.exists("/local_model_args.json"):
+        return {}
+
+    with open("/local_model_args.json", "r") as f:
+        local_args = json.load(f)
+
+    if local_args.get("MODEL_NAME") is None:
+        logging.warning("Model name not found in /local_model_args.json. There maybe was a problem when baking the model in.")
+
+    logging.info(f"Using baked in model with args: {local_args}")
+    os.environ["TRANSFORMERS_OFFLINE"] = "1"
+    os.environ["HF_HUB_OFFLINE"] = "1"
+
+    return local_args
+def get_engine_args():
+    # Start with default args
+    args = DEFAULT_ARGS
+    
+    # Get env args that match keys in AsyncEngineArgs
+    args.update(os.environ)
+    
+    # Get local args if model is baked in and overwrite env args
+    args.update(get_local_args())
+    
+    # if args.get("TENSORIZER_URI"): TODO: add back once tensorizer is ready
+    #     args["load_format"] = "tensorizer"
+    #     args["model_loader_extra_config"] = TensorizerConfig(tensorizer_uri=args["TENSORIZER_URI"], num_readers=None)
+    #     logging.info(f"Using tensorized model from {args['TENSORIZER_URI']}")
+    
+    
+    # Rename and match to vllm args
+    args = match_vllm_args(args)
+
+    if args.get("load_format") == "bitsandbytes":
+        args["quantization"] = args["load_format"]
+    
+    # Set tensor parallel size and max parallel loading workers if more than 1 GPU is available
+    num_gpus = device_count()
+    if num_gpus > 1:
+        args["tensor_parallel_size"] = num_gpus
+        args["max_parallel_loading_workers"] = None
+        if os.getenv("MAX_PARALLEL_LOADING_WORKERS"):
+            logging.warning("Overriding MAX_PARALLEL_LOADING_WORKERS with None because more than 1 GPU is available.")
+    
+    # Deprecated env args backwards compatibility
+    if args.get("kv_cache_dtype") == "fp8_e5m2":
+        args["kv_cache_dtype"] = "fp8"
+        logging.warning("Using fp8_e5m2 is deprecated. Please use fp8 instead.")
+    if os.getenv("MAX_CONTEXT_LEN_TO_CAPTURE"):
+        args["max_seq_len_to_capture"] = int(os.getenv("MAX_CONTEXT_LEN_TO_CAPTURE"))
+        logging.warning("Using MAX_CONTEXT_LEN_TO_CAPTURE is deprecated. Please use MAX_SEQ_LEN_TO_CAPTURE instead.")
+        
+    # if "gemma-2" in args.get("model", "").lower():
+    #     os.environ["VLLM_ATTENTION_BACKEND"] = "FLASHINFER"
+    #     logging.info("Using FLASHINFER for gemma-2 model.")
+        
+    return AsyncEngineArgs(**args)

src/tokenizer.py

@@ -4,7 +4,8 @@ from typing import Union
 
 class TokenizerWrapper:
     def __init__(self, tokenizer_name_or_path, tokenizer_revision, trust_remote_code):
-        self.tokenizer = AutoTokenizer.from_pretrained(tokenizer_name_or_path, revision=tokenizer_revision, trust_remote_code=trust_remote_code)
+        print(f"tokenizer_name_or_path: {tokenizer_name_or_path}, tokenizer_revision: {tokenizer_revision}, trust_remote_code: {trust_remote_code}")
+        self.tokenizer = AutoTokenizer.from_pretrained(tokenizer_name_or_path, revision=tokenizer_revision or "main", trust_remote_code=trust_remote_code)
         self.custom_chat_template = os.getenv("CUSTOM_CHAT_TEMPLATE")
         self.has_chat_template = bool(self.tokenizer.chat_template) or bool(self.custom_chat_template)
         if self.custom_chat_template and isinstance(self.custom_chat_template, str):

src/utils.py

@@ -1,12 +1,29 @@
 import os
 import logging
 from http import HTTPStatus
-from vllm.utils import random_uuid
+from functools import wraps
-from vllm.entrypoints.openai.protocol import ErrorResponse
+from time import time
-from vllm import SamplingParams
+from vllm.entrypoints.openai.protocol import RequestResponseMetadata
+
+try:
+    from vllm.utils import random_uuid
+    from vllm.entrypoints.openai.protocol import ErrorResponse
+    from vllm import SamplingParams
+except ImportError:
+    logging.warning("Error importing vllm, skipping related imports. This is ONLY expected when baking model into docker image from a machine without GPUs")
+    pass
 
 logging.basicConfig(level=logging.INFO)
 
+# Updated to parse multiple comma-separated multimodal limits (e.g., 'image=1,video=0')
+def convert_limit_mm_per_prompt(input_string: str):
+    result = {}
+    pairs = input_string.split(',')
+    for pair in pairs:
+        key, value = pair.split('=')
+        result[key] = int(value)
+    return result
+
 def count_physical_cores():
     with open('/proc/cpuinfo') as f:
         content = f.readlines()
@@ -32,7 +49,11 @@ class JobInput:
         self.max_batch_size = job.get("max_batch_size")
         self.apply_chat_template = job.get("apply_chat_template", False)
         self.use_openai_format = job.get("use_openai_format", False)
-        self.sampling_params = SamplingParams(**job.get("sampling_params", {}))
+        samp_param = job.get("sampling_params", {})
+        if "max_tokens" not in samp_param:
+            samp_param["max_tokens"] = 100
+        self.sampling_params = SamplingParams(**samp_param)
+        # self.sampling_params = SamplingParams(max_tokens=100, **job.get("sampling_params", {}))
         self.request_id = random_uuid()
         batch_size_growth_factor = job.get("batch_size_growth_factor")
         self.batch_size_growth_factor = float(batch_size_growth_factor) if batch_size_growth_factor else None 
@@ -40,8 +61,14 @@ class JobInput:
         self.min_batch_size = int(min_batch_size) if min_batch_size else None 
         self.openai_route = job.get("openai_route")
         self.openai_input = job.get("openai_input")
+class DummyState:
+    def __init__(self):
+        self.request_metadata = None
         
 class DummyRequest:
+    def __init__(self):
+        self.headers = {}
+        self.state = DummyState()
     async def is_disconnected(self):
         return False
 
@@ -68,6 +95,12 @@ def create_error_response(message: str, err_type: str = "BadRequestError", statu
 def get_int_bool_env(env_var: str, default: bool) -> bool:
     return int(os.getenv(env_var, int(default))) == 1
 
-
+def timer_decorator(func):
-        
+    @wraps(func)
-
+    def wrapper(*args, **kwargs):
+        start = time()
+        result = func(*args, **kwargs)
+        end = time()
+        logging.info(f"{func.__name__} completed in {end - start:.2f} seconds")
+        return result
+    return wrapper

vllm-base-image/Dockerfile

@@ -1,149 +0,0 @@
-################### vLLM Base Dockerfile ###################
-# This Dockerfile is for building the image that the  
-# vLLM worker container will use as its base image. 
-# If your changes are outside of the vLLM source code, you
-# do not need to build this image.
-##########################################################
-
-# Define the CUDA version for the build
-ARG WORKER_CUDA_VERSION=11.8.0
-
-FROM nvidia/cuda:${WORKER_CUDA_VERSION}-devel-ubuntu22.04 AS dev
-
-# Re-declare ARG after FROM
-ARG WORKER_CUDA_VERSION
-
-# Update and install dependencies
-RUN apt-get update -y \
-    && apt-get install -y python3-pip git
-
-# Set working directory
-WORKDIR /vllm-installation
-
-RUN ldconfig /usr/local/cuda-$(echo "$WORKER_CUDA_VERSION" | sed 's/\.0$//')/compat/
-
-# Install build and runtime dependencies
-COPY vllm/requirements-common.txt requirements-common.txt
-COPY vllm/requirements-cuda${WORKER_CUDA_VERSION}.txt  requirements-cuda.txt
-RUN --mount=type=cache,target=/root/.cache/pip \
-    pip install -r requirements-cuda.txt
-
-# Install development dependencies
-COPY vllm/requirements-dev.txt requirements-dev.txt
-RUN --mount=type=cache,target=/root/.cache/pip \
-    pip install -r requirements-dev.txt
-    
-ARG torch_cuda_arch_list='7.0 7.5 8.0 8.6 8.9 9.0+PTX'
-ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
-
-FROM dev AS build
-
-# Re-declare ARG after FROM
-ARG WORKER_CUDA_VERSION
-
-# Install build dependencies
-COPY vllm/requirements-build.txt requirements-build.txt
-RUN --mount=type=cache,target=/root/.cache/pip \
-    pip install -r requirements-build.txt
-
-# install compiler cache to speed up compilation leveraging local or remote caching
-RUN apt-get update -y && apt-get install -y ccache
-
-# Copy necessary files
-COPY vllm/csrc csrc
-COPY vllm/setup.py setup.py
-COPY vllm/cmake cmake
-COPY vllm/CMakeLists.txt CMakeLists.txt
-COPY vllm/requirements-common.txt requirements-common.txt
-COPY vllm/requirements-cuda${WORKER_CUDA_VERSION}.txt requirements-cuda.txt
-COPY vllm/pyproject.toml pyproject.toml
-COPY vllm/vllm vllm 
-
-# Set environment variables for building extensions
-ENV WORKER_CUDA_VERSION=${WORKER_CUDA_VERSION}
-ENV VLLM_INSTALL_PUNICA_KERNELS=0
-# Build extensions
-ENV CCACHE_DIR=/root/.cache/ccache
-RUN --mount=type=cache,target=/root/.cache/ccache \
-    --mount=type=cache,target=/root/.cache/pip \
-    python3 setup.py bdist_wheel --dist-dir=dist
-
-RUN --mount=type=cache,target=/root/.cache/pip \
-    pip cache remove vllm_nccl*
-
-FROM dev as flash-attn-builder
-# max jobs used for build
-# flash attention version
-ARG flash_attn_version=v2.5.8
-ENV FLASH_ATTN_VERSION=${flash_attn_version}
-
-WORKDIR /usr/src/flash-attention-v2
-
-# Download the wheel or build it if a pre-compiled release doesn't exist
-RUN pip --verbose wheel flash-attn==${FLASH_ATTN_VERSION} \
-    --no-build-isolation --no-deps --no-cache-dir
-
-FROM dev as NCCL-installer
-
-# Re-declare ARG after FROM
-ARG WORKER_CUDA_VERSION
-
-# Update and install necessary libraries
-RUN apt-get update -y \
-    && apt-get install -y wget
-
-# Install NCCL library
-RUN if [ "$WORKER_CUDA_VERSION" = "11.8.0" ]; then \
-        wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.0-1_all.deb \
-        && dpkg -i cuda-keyring_1.0-1_all.deb \
-        && apt-get update \
-        && apt install -y libnccl2=2.15.5-1+cuda11.8 libnccl-dev=2.15.5-1+cuda11.8; \
-    elif [ "$WORKER_CUDA_VERSION" = "12.1.0" ]; then \
-        wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.0-1_all.deb \
-        && dpkg -i cuda-keyring_1.0-1_all.deb \
-        && apt-get update \
-        && apt install -y libnccl2=2.17.1-1+cuda12.1 libnccl-dev=2.17.1-1+cuda12.1; \
-    else \
-        echo "Unsupported CUDA version: $WORKER_CUDA_VERSION"; \
-        exit 1; \
-    fi
-
-FROM nvidia/cuda:${WORKER_CUDA_VERSION}-base-ubuntu22.04 AS vllm-base
-
-# Re-declare ARG after FROM
-ARG WORKER_CUDA_VERSION
-
-# Update and install necessary libraries
-RUN apt-get update -y \
-    && apt-get install -y python3-pip
-
-# Set working directory
-WORKDIR /vllm-workspace
-
-RUN ldconfig /usr/local/cuda-$(echo "$WORKER_CUDA_VERSION" | sed 's/\.0$//')/compat/
-
-RUN --mount=type=bind,from=build,src=/vllm-installation/dist,target=/vllm-workspace/dist \
-    --mount=type=cache,target=/root/.cache/pip \
-    pip install dist/*.whl --verbose
-
-RUN --mount=type=bind,from=flash-attn-builder,src=/usr/src/flash-attention-v2,target=/usr/src/flash-attention-v2 \
-    --mount=type=cache,target=/root/.cache/pip \
-    pip install /usr/src/flash-attention-v2/*.whl --no-cache-dir
-
-FROM vllm-base AS runtime
-
-# install additional dependencies for openai api server
-RUN --mount=type=cache,target=/root/.cache/pip \
-    pip install accelerate hf_transfer modelscope tensorizer
-
-# Set PYTHONPATH environment variable
-ENV PYTHONPATH="/"
-
-# Copy NCCL library
-COPY --from=NCCL-installer /usr/lib/x86_64-linux-gnu/libnccl.so.2 /usr/lib/x86_64-linux-gnu/libnccl.so.2
-# Set the VLLM_NCCL_SO_PATH environment variable
-ENV VLLM_NCCL_SO_PATH="/usr/lib/x86_64-linux-gnu/libnccl.so.2"
-
-
-# Validate the installation
-RUN python3 -c "import vllm; print(vllm.__file__)"

vllm-base-image/README.md

@@ -1 +0,0 @@
-This directory is for building the vllm-base image utilized by the worker.

vllm-base-image/vllm-metadata.yml

@@ -1,2 +0,0 @@
-version: '0.4.2'
-dev_version: '0.4.2'