Merge pull request #200 from runpod-workers/release/0.10.0

chore(release): 0.10.0

.runpod/tests.json

@@ -15,14 +15,6 @@
             {
                 "key": "MODEL_NAME",
                 "value": "facebook/opt-350m"
-            },
-            {
-                "key": "MAX_MODEL_LEN",
-                "value": "8192"
-            },
-            {
-                "key": "GPU_MEMORY_UTILIZATION",
-                "value": "0.95"
             }
         ],
         "allowedCudaVersions": [

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==0.8.4 && \
+RUN python3 -m pip install vllm==0.10.0 && \
     python3 -m pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.3
 
 # Setup for Option 2: Building the Image with the Model included
@@ -47,4 +47,4 @@ RUN --mount=type=secret,id=HF_TOKEN,required=false \
     fi
 
 # Start the handler
-CMD ["python3", "/src/handler.py"]
+CMD ["python3", "/src/handler.py"]

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 `v2.5.0` with vLLM `0.8.5` now available under `stable` tags 
+### 2. Worker vLLM `v2.7.0` with vLLM `0.9.1` now available under `stable` tags 
 
-Update v2.5.0 is now available, use the image tag `runpod/worker-v1-vllm:v2.5.0stable-cuda12.1.0`.
+Update v2.7.0 is now available, use the image tag `runpod/worker-v1-vllm:v2.7.0stable-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! 
@@ -81,7 +81,7 @@ 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:v2.5.0stable-cuda12.1.0` | `runpod/worker-v1-vllm:v2.5.0dev-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:v2.7.0stable-cuda12.1.0` | `runpod/worker-v1-vllm:v2.7.0dev-cuda12.1.0` | When creating an Endpoint, select CUDA Version 12.3, 12.2 and 12.1 in the filter. |
 
 
 
@@ -139,6 +139,7 @@ Below is a summary of the available RunPod Worker images, categorized by image s
 | `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"}`|
 | `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. |

builder/requirements.txt

@@ -8,5 +8,7 @@ typing-extensions>=4.8.0
 pydantic
 pydantic-settings
 hf-transfer
-transformers
+transformers>=4.55.0
 bitsandbytes>=0.45.0
+kernels
+torch==2.6.0

docker-bake.hcl

@@ -7,7 +7,7 @@ variable "REPOSITORY" {
 }
 
 variable "BASE_IMAGE_VERSION" {
-  default = "v2.4.0stable"
+  default = "v2.8.0stable"
 }
 
 group "all" {
@@ -29,4 +29,4 @@ target "worker-1210" {
     WORKER_CUDA_VERSION = "12.1.0"
   }
   output = ["type=docker,push=${PUSH}"]
-}
+}

src/engine.py

@@ -25,15 +25,68 @@ class vLLMEngine:
         load_dotenv() # For local development
         self.engine_args = get_engine_args()
         logging.info(f"Engine args: {self.engine_args}")
-        self.tokenizer = TokenizerWrapper(self.engine_args.tokenizer or self.engine_args.model, 
-                                          self.engine_args.tokenizer_revision, 
-                                          self.engine_args.trust_remote_code)
+        
+        # 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)
                            
@@ -55,7 +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}
@@ -122,41 +176,55 @@ class OpenAIvLLMEngine(vLLMEngine):
         super().__init__(vllm_engine)
         self.served_model_name = os.getenv("OPENAI_SERVED_MODEL_NAME_OVERRIDE") or self.engine_args.model
         self.response_role = os.getenv("OPENAI_RESPONSE_ROLE") or "assistant"
+        self.lora_adapters = self._load_lora_adapters()
         asyncio.run(self._initialize_engines())
         self.raw_openai_output = bool(int(os.getenv("RAW_OPENAI_OUTPUT", 1)))
-        
+
+    def _load_lora_adapters(self):
+        adapters = []
+        try:
+            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)
         ]
 
-        lora_modules = os.getenv('LORA_MODULES', None)
-        if lora_modules is not None:
-            try:
-                lora_modules = json.loads(lora_modules)
-                lora_modules = [LoRAModulePath(**lora_modules)]
-            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,
+            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=self.tokenizer.tokenizer.chat_template,
+            chat_template=chat_template,
             chat_template_content_format="auto",
             # enable_reasoning=os.getenv('ENABLE_REASONING', 'false').lower() == 'true',
-            # reasoning_parser=None,
+            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,

src/engine_args.py

@@ -26,6 +26,7 @@ DEFAULT_ARGS = {
     "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),
@@ -92,6 +93,9 @@ DEFAULT_ARGS = {
     "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:
@@ -122,7 +126,7 @@ def get_local_args():
         local_args = json.load(f)
 
     if local_args.get("MODEL_NAME") is None:
-        raise ValueError("Model name not found in /local_model_args.json. There was a problem when baking the model in.")
+        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"

src/utils.py

@@ -15,9 +15,14 @@ except ImportError:
 
 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):
-    key, value = input_string.split('=')
-    return {key: int(value)}
+    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:

worker-config.json

@@ -1,5 +1,249 @@
 {
   "versions": {
+    "0.10.0": {
+      "imageName": "runpod/worker-v1-vllm:v2.8.0stable-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"
+          ]
+        }
+      ]
+    },
+    "0.9.0": {
+      "imageName": "runpod/worker-v1-vllm:v2.6.0stable-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"
+          ]
+        }
+      ]
+    },
+    "0.9.1": {
+      "imageName": "runpod/worker-v1-vllm:v2.7.0stable-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"
+          ]
+        }
+      ]
+    },
+    "0.9.0": {
+      "imageName": "runpod/worker-v1-vllm:v2.6.0stable-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"
+          ]
+        }
+      ]
+    },
     "0.8.5": {
       "imageName": "runpod/worker-v1-vllm:v2.5.0stable-cuda12.1.0",
       "minimumCudaVersion": "12.1",
@@ -548,126 +792,6 @@
           ]
         }
       ]
-    },
-    "0.6.2": {
-      "imageName": "runpod/worker-v1-vllm:v1.5.0stable-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", "USE_V2_BLOCK_MANAGER", "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"
-          ]
-        },
-        {
-          "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"
-          ]
-        }
-      ]
-    },
-    "0.6.1": {
-      "imageName": "runpod/worker-v1-vllm:v1.4.0stable-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", "USE_V2_BLOCK_MANAGER", "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"
-          ]
-        },
-        {
-          "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"
-          ]
-        }
-      ]
     }
   },
   "schema": {