[Core] Interface for accessing model from VllmRunner (#10353)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/v1/worker/gpu_model_runner.py

@@ -689,6 +689,9 @@ class GPUModelRunner:
                 encoder_outputs.append(encoder_output[start_idx:end_idx])
         return encoder_outputs
 
+    def get_model(self) -> nn.Module:
+        return self.model
+
     @torch.inference_mode()
     def execute_model(
         self,

vllm/v1/worker/gpu_worker.py

@@ -5,6 +5,7 @@ from typing import TYPE_CHECKING, Optional
 
 import torch
 import torch.distributed
+import torch.nn as nn
 
 import vllm.envs as envs
 from vllm.config import CacheConfig, ModelConfig, ParallelConfig, VllmConfig
@@ -176,6 +177,9 @@ class Worker:
         # the model initialization and profiling.
         set_random_seed(self.model_config.seed)
 
+    def get_model(self) -> nn.Module:
+        return self.model_runner.get_model()
+
     @torch.inference_mode()
     def execute_model(
         self,