Remove requirement to use `--hf-overrides` for `DeepseekVLV2ForCausalLM` (#35203)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

docs/models/supported_models.md

@@ -682,7 +682,7 @@ These models primarily accept the [`LLM.generate`](./generative_models.md#llmgen
 | `Blip2ForConditionalGeneration` | BLIP-2 | T + I<sup>E</sup> | `Salesforce/blip2-opt-2.7b`, `Salesforce/blip2-opt-6.7b`, etc. | ✅︎ | ✅︎ |
 | `ChameleonForConditionalGeneration` | Chameleon | T + I | `facebook/chameleon-7b`, etc. | | ✅︎ |
 | `Cohere2VisionForConditionalGeneration` | Command A Vision | T + I<sup>+</sup> | `CohereLabs/command-a-vision-07-2025`, etc. | | ✅︎ |
-| `DeepseekVLV2ForCausalLM`<sup>^</sup> | DeepSeek-VL2 | T + I<sup>+</sup> | `deepseek-ai/deepseek-vl2-tiny`, `deepseek-ai/deepseek-vl2-small`, `deepseek-ai/deepseek-vl2`, etc. | | ✅︎ |
+| `DeepseekVLV2ForCausalLM` | DeepSeek-VL2 | T + I<sup>+</sup> | `deepseek-ai/deepseek-vl2-tiny`, `deepseek-ai/deepseek-vl2-small`, `deepseek-ai/deepseek-vl2`, etc. | | ✅︎ |
 | `DeepseekOCRForCausalLM` | DeepSeek-OCR | T + I<sup>+</sup> | `deepseek-ai/DeepSeek-OCR`, etc. | ✅︎ | ✅︎ |
 | `DeepseekOCR2ForCausalLM` | DeepSeek-OCR-2 | T + I<sup>+</sup> | `deepseek-ai/DeepSeek-OCR-2`, etc. | ✅︎ | ✅︎ |
 | `Eagle2_5_VLForConditionalGeneration` | Eagle2.5-VL | T + I<sup>E+</sup> | `nvidia/Eagle2.5-8B`, etc. | ✅︎ | ✅︎ |
@@ -762,10 +762,8 @@ Some models are supported only via the [Transformers modeling backend](#transfor
 |--------------|--------|--------|-------------------|-----------------------------|-----------------------------------------|
 | `Emu3ForConditionalGeneration` | Emu3 | T + I | `BAAI/Emu3-Chat-hf` | ✅︎ | ✅︎ |
 
-<sup>^</sup> You need to set the architecture name via `--hf-overrides` to match the one in vLLM.
-&nbsp;&nbsp;&nbsp;&nbsp;• For example, to use DeepSeek-VL2 series models:
-&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;`--hf-overrides '{"architectures": ["DeepseekVLV2ForCausalLM"]}'`
-<sup>E</sup> Pre-computed embeddings can be inputted for this modality.
+<sup>^</sup> You need to set the architecture name via `--hf-overrides` to match the one in vLLM.</br>
+<sup>E</sup> Pre-computed embeddings can be inputted for this modality.</br>
 <sup>+</sup> Multiple items can be inputted per text prompt for this modality.
 
 !!! note

tests/models/registry.py

@@ -715,7 +715,6 @@ _MULTIMODAL_EXAMPLE_MODELS = {
         extras={"fork": "Isotr0py/deepseek-vl2-tiny"},
         max_transformers_version="4.48",
         transformers_version_reason={"hf": "HF model is not compatible."},
-        hf_overrides={"architectures": ["DeepseekVLV2ForCausalLM"]},
     ),
     "DeepseekOCRForCausalLM": _HfExamplesInfo(
         "deepseek-ai/DeepSeek-OCR",

tests/v1/kv_connector/nixl_integration/run_accuracy_test.sh

@@ -95,18 +95,6 @@ cleanup_instances() {
   sleep 2
 }
 
-# Handle to get model-specific arguments for deepseek
-get_model_args() {
-  local model_name=$1
-  local extra_args=""
-
-  if [[ "$model_name" == "deepseek-ai/deepseek-vl2-tiny" ]]; then
-    extra_args="--hf_overrides '{\"architectures\": [\"DeepseekVLV2ForCausalLM\"]}' --trust-remote-code"
-  fi
-
-  echo "$extra_args"
-}
-
 get_num_gpus() {
   if [[ "$SMI_BIN" == *"nvidia"* ]]; then
     $SMI_BIN --query-gpu=name --format=csv,noheader | wc -l
@@ -127,9 +115,6 @@ run_tests_for_model() {
   echo "Testing model: $model_name"
   echo "================================"
 
-  # Get model-specific arguments
-  local model_args=$(get_model_args "$model_name")
-
   # Arrays to store all hosts and ports
   PREFILL_HOSTS=()
   PREFILL_PORTS=()
@@ -172,11 +157,7 @@ run_tests_for_model() {
       BASE_CMD="${BASE_CMD} --attention-backend=$ATTENTION_BACKEND"
     fi
 
-    if [ -n "$model_args" ]; then
-    FULL_CMD="$BASE_CMD $model_args"
-    else
     FULL_CMD="$BASE_CMD"
-    fi
 
     eval "$FULL_CMD &"
 
@@ -227,11 +208,7 @@ run_tests_for_model() {
     --tensor-parallel-size 1 --enable-expert-parallel"
   fi
 
-    if [ -n "$model_args" ]; then
-    FULL_CMD="$BASE_CMD $model_args"
-    else
     FULL_CMD="$BASE_CMD"
-    fi
 
     eval "$FULL_CMD &"
 

tests/v1/kv_connector/nixl_integration/run_edge_case_test.sh

@@ -55,19 +55,6 @@ cleanup_instances() {
   sleep 2
 }
 
-# Handle to get model-specific arguments for deepseek
-get_model_args() {
-  local model_name=$1
-  local extra_args=""
-
-  if [[ "$model_name" == "deepseek-ai/deepseek-vl2-tiny" ]]; then
-    extra_args="--hf_overrides '{\"architectures\": [\"DeepseekVLV2ForCausalLM\"]}' --trust-remote-code"
-  fi
-
-  echo "$extra_args"
-}
-
-
 # Function to run tests for a specific model
 run_tests_for_model() {
   local model_name=$1
@@ -75,9 +62,6 @@ run_tests_for_model() {
   echo "Testing model: $model_name"
   echo "================================"
 
-  # Get model-specific arguments
-  local model_args=$(get_model_args "$model_name")
-
   # Start prefill instance
   PREFILL_PORT=8001
 
@@ -87,11 +71,7 @@ run_tests_for_model() {
   --gpu-memory-utilization 0.2 \
   --kv-transfer-config '$KV_CONFIG'"
 
-  if [ -n "$model_args" ]; then
-  FULL_CMD="$BASE_CMD $model_args"
-  else
   FULL_CMD="$BASE_CMD"
-  fi
 
   eval "$FULL_CMD &"
 
@@ -105,11 +85,7 @@ run_tests_for_model() {
   --gpu-memory-utilization 0.2 \
   --kv-transfer-config '$KV_CONFIG'"
 
-  if [ -n "$model_args" ]; then
-  FULL_CMD="$BASE_CMD $model_args"
-  else
   FULL_CMD="$BASE_CMD"
-  fi
 
   eval "$FULL_CMD &"
 

vllm/transformers_utils/configs/deepseek_vl2.py

@@ -89,6 +89,7 @@ class MlpProjectorConfig(PretrainedConfig):
 
 class DeepseekVLV2Config(PretrainedConfig):
     model_type = "deepseek_vl_v2"
+    architectures: list[str] | None = None
     vision_config: VisionEncoderConfig
     projector_config: MlpProjectorConfig
 
@@ -105,6 +106,9 @@ class DeepseekVLV2Config(PretrainedConfig):
     ):
         super().__init__(**kwargs)
 
+        if self.architectures is None:
+            self.architectures = ["DeepseekVLV2ForCausalLM"]
+
         vision_config = kwargs.get("vision_config", {})
         self.vision_config = VisionEncoderConfig(**vision_config)
 
@@ -120,8 +124,7 @@ class DeepseekVLV2Config(PretrainedConfig):
         self.vocab_size = self.text_config.vocab_size
 
         # update model_type for OCR models
-        architectures = self.architectures or kwargs.get("architectures", [])
-        if "DeepseekOCRForCausalLM" in architectures:
+        if "DeepseekOCRForCausalLM" in self.architectures:
             self.model_type = "deepseek_ocr"
-        elif "DeepseekOCR2ForCausalLM" in architectures:
+        elif "DeepseekOCR2ForCausalLM" in self.architectures:
             self.model_type = "deepseek_ocr2"