[Model] Add BNB support to Llava and Pixtral-HF (#10795)

Signed-off-by: Isotr0py <2037008807@qq.com>

vllm/model_executor/models/llava.py

@@ -287,6 +287,15 @@ def init_vision_tower_for_llava(
 @INPUT_REGISTRY.register_dummy_data(dummy_data_for_llava)
 @INPUT_REGISTRY.register_input_processor(input_processor_for_llava)
 class LlavaForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP):
+    # BitandBytes specific attributes
+    bitsandbytes_stacked_params_mapping = {
+        # shard_name, weight_name, index
+        "q_proj": ("qkv_proj", 0),
+        "k_proj": ("qkv_proj", 1),
+        "v_proj": ("qkv_proj", 2),
+        "gate_proj": ("gate_up_proj", 0),
+        "up_proj": ("gate_up_proj", 1),
+    }
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
         super().__init__()