Update README.md

[HF]: set default max_token size

Dockerfile

@@ -12,7 +12,7 @@ RUN --mount=type=cache,target=/root/.cache/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==0.6.6.post1 && \
+RUN python3 -m pip install vllm==0.7.3 && \
     python3 -m pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.3
 
 # Setup for Option 2: Building the Image with the Model included
@@ -32,7 +32,7 @@ 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"
 
@@ -40,10 +40,10 @@ ENV PYTHONPATH="/:/vllm-workspace"
 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 /src/download_model.py; \
+    python3 /src/download_model.py; \
     fi
 
 # Start the handler

README.md

@@ -18,9 +18,9 @@ Deploy OpenAI-Compatible Blazing-Fast LLM Endpoints powered by the [vLLM](https:
 ### 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 `v1.8.0` with vLLM `0.6.6` now available under `stable` tags 
+### 2. Worker vLLM `v2.1.0` with vLLM `0.7.3` now available under `stable` tags 
 
-Update v1.8.0 is now available, use the image tag `runpod/worker-v1-vllm:v1.8.0stable-cuda12.1.0`.
+Update v2.0.0 is now available, use the image tag `runpod/worker-v1-vllm:v2.1.0stable-cuda12.1.0`.
 
 ### 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! 
@@ -82,7 +82,7 @@ Below is a summary of the available RunPod Worker images, categorized by image s
 
 | CUDA Version | Stable Image Tag                  | Development Image Tag             | Note                                                        |
 |--------------|-----------------------------------|-----------------------------------|----------------------------------------------------------------------|
-| 12.1.0       | `runpod/worker-v1-vllm:v1.8.0stable-cuda12.1.0` | `runpod/worker-v1-vllm:v1.8.0dev-cuda12.1.0` | When creating an Endpoint, select CUDA Version 12.3, 12.2 and 12.1 in the filter. |
+| 12.1.0       | `runpod/worker-v1-vllm:v2.1.0stable-cuda12.1.0` | `runpod/worker-v1-vllm:v2.1.0dev-cuda12.1.0` | When creating an Endpoint, select CUDA Version 12.3, 12.2 and 12.1 in the filter. |
 
 
 

docker-bake.hcl

@@ -7,7 +7,7 @@ variable "REPOSITORY" {
 }
 
 variable "BASE_IMAGE_VERSION" {
-  default = "stable"
+  default = "v2.0.0stable"
 }
 
 group "all" {

src/engine.py

@@ -4,14 +4,15 @@ import json
 import asyncio
 
 from dotenv import load_dotenv
-from typing import AsyncGenerator
+from typing import AsyncGenerator, Optional
 import time
 
 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_engine import BaseModelPath, LoRAModulePath
+from vllm.entrypoints.openai.serving_models import BaseModelPath, LoRAModulePath, OpenAIServingModels
 
 
 from utils import DummyRequest, JobInput, BatchSize, create_error_response
@@ -138,28 +139,35 @@ class OpenAIvLLMEngine(vLLMEngine):
             except:
                 lora_modules = None
 
-
+        self.serving_models = OpenAIServingModels(
+            engine_client=self.llm,
+            model_config=self.model_config,
+            base_model_paths=self.base_model_paths,
+            lora_modules=None,
+            prompt_adapters=None,
+        )
 
         self.chat_engine = OpenAIServingChat(
             engine_client=self.llm, 
             model_config=self.model_config,
-            base_model_paths=self.base_model_paths,
+            models=self.serving_models,
             response_role=self.response_role,
+            request_logger=None,
             chat_template=self.tokenizer.tokenizer.chat_template,
+            chat_template_content_format="auto",
+            # enable_reasoning=os.getenv('ENABLE_REASONING', 'false').lower() == 'true',
+            # reasoning_parser=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,
-            lora_modules=lora_modules,
-            prompt_adapters=None,
-            chat_template_content_format="auto",
-            request_logger=None
+            enable_prompt_tokens_details=False
         )
         self.completion_engine = OpenAIServingCompletion(
             engine_client=self.llm, 
             model_config=self.model_config,
-            base_model_paths=self.base_model_paths,
-            lora_modules=lora_modules,
-            prompt_adapters=None,
-            request_logger=None
+            models=self.serving_models,
+            request_logger=None,
+            # return_token_as_token_ids=False,
         )
     
     async def generate(self, openai_request: JobInput):
@@ -172,7 +180,7 @@ class OpenAIvLLMEngine(vLLMEngine):
             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()
         return models.model_dump()
     
     async def _handle_chat_or_completion_request(self, openai_request: JobInput):

src/engine_args.py

@@ -4,6 +4,7 @@ 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",
@@ -89,7 +90,7 @@ DEFAULT_ARGS = {
     "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')
+    "use_v2_block_manager": os.getenv('USE_V2_BLOCK_MANAGER', 'true'),
 }
 
 def match_vllm_args(args):

src/utils.py

@@ -15,6 +15,10 @@ except ImportError:
 
 logging.basicConfig(level=logging.INFO)
 
+def convert_limit_mm_per_prompt(input_string: str):
+    key, value = input_string.split('=')
+    return {key: int(value)}
+
 def count_physical_cores():
     with open('/proc/cpuinfo') as f:
         content = f.readlines()
@@ -40,7 +44,11 @@ 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 

worker-config.json

@@ -1,5 +1,66 @@
 {
   "versions": {
+    "0.7.0": {
+      "imageName": "runpod/worker-v1-vllm:v1.9.0stable-cuda12.1.0",
+      "minimumCudaVersion": "12.1",
+      "categories": [
+        {
+          "title": "LLM Settings",
+          "settings": [
+            "TOKENIZER", "TOKENIZER_MODE", "SKIP_TOKENIZER_INIT", "TRUST_REMOTE_CODE",
+            "DOWNLOAD_DIR", "LOAD_FORMAT", "DTYPE", "KV_CACHE_DTYPE", "QUANTIZATION_PARAM_PATH",
+            "MAX_MODEL_LEN", "GUIDED_DECODING_BACKEND", "DISTRIBUTED_EXECUTOR_BACKEND",
+            "WORKER_USE_RAY", "RAY_WORKERS_USE_NSIGHT", "PIPELINE_PARALLEL_SIZE",
+            "TENSOR_PARALLEL_SIZE", "MAX_PARALLEL_LOADING_WORKERS", "ENABLE_PREFIX_CACHING",
+            "DISABLE_SLIDING_WINDOW", "NUM_LOOKAHEAD_SLOTS",
+            "SEED", "NUM_GPU_BLOCKS_OVERRIDE", "MAX_NUM_BATCHED_TOKENS", "MAX_NUM_SEQS",
+            "MAX_LOGPROBS", "DISABLE_LOG_STATS", "QUANTIZATION", "ROPE_SCALING", "ROPE_THETA",
+            "TOKENIZER_POOL_SIZE", "TOKENIZER_POOL_TYPE", "TOKENIZER_POOL_EXTRA_CONFIG",
+            "ENABLE_LORA", "MAX_LORAS", "MAX_LORA_RANK", "LORA_EXTRA_VOCAB_SIZE",
+            "LORA_DTYPE", "LONG_LORA_SCALING_FACTORS", "MAX_CPU_LORAS", "FULLY_SHARDED_LORAS",
+            "DEVICE", "SCHEDULER_DELAY_FACTOR", "ENABLE_CHUNKED_PREFILL", "SPECULATIVE_MODEL",
+            "NUM_SPECULATIVE_TOKENS", "SPECULATIVE_DRAFT_TENSOR_PARALLEL_SIZE",
+            "SPECULATIVE_MAX_MODEL_LEN", "SPECULATIVE_DISABLE_BY_BATCH_SIZE",
+            "NGRAM_PROMPT_LOOKUP_MAX", "NGRAM_PROMPT_LOOKUP_MIN", "SPEC_DECODING_ACCEPTANCE_METHOD",
+            "TYPICAL_ACCEPTANCE_SAMPLER_POSTERIOR_THRESHOLD", "TYPICAL_ACCEPTANCE_SAMPLER_POSTERIOR_ALPHA",
+            "MODEL_LOADER_EXTRA_CONFIG", "PREEMPTION_MODE", "PREEMPTION_CHECK_PERIOD",
+            "PREEMPTION_CPU_CAPACITY", "MAX_LOG_LEN", "DISABLE_LOGGING_REQUEST",
+            "ENABLE_AUTO_TOOL_CHOICE", "TOOL_CALL_PARSER"
+          ]
+        },
+        {
+          "title": "Tokenizer Settings",
+          "settings": [
+            "TOKENIZER_NAME", "TOKENIZER_REVISION", "CUSTOM_CHAT_TEMPLATE"
+          ]
+        },
+        {
+          "title": "System Settings",
+          "settings": [
+            "GPU_MEMORY_UTILIZATION", "MAX_PARALLEL_LOADING_WORKERS", "BLOCK_SIZE",
+            "SWAP_SPACE", "ENFORCE_EAGER", "MAX_SEQ_LEN_TO_CAPTURE", "DISABLE_CUSTOM_ALL_REDUCE"
+          ]
+        },
+        {
+          "title": "Streaming Settings",
+          "settings": [
+            "DEFAULT_BATCH_SIZE", "DEFAULT_MIN_BATCH_SIZE", "DEFAULT_BATCH_SIZE_GROWTH_FACTOR"
+          ]
+        },
+        {
+          "title": "OpenAI Settings",
+          "settings": [
+            "RAW_OPENAI_OUTPUT", "OPENAI_RESPONSE_ROLE", "OPENAI_SERVED_MODEL_NAME_OVERRIDE"
+          ]
+        },
+        {
+          "title": "Serverless Settings",
+          "settings": [
+            "MAX_CONCURRENCY", "DISABLE_LOG_STATS", "DISABLE_LOG_REQUESTS"
+          ]
+        }
+      ]
+    },
     "0.6.6": {
       "imageName": "runpod/worker-v1-vllm:v1.8.0stable-cuda12.1.0",
       "minimumCudaVersion": "12.1",