Add --max-model-len auto to auto-fit context to available memory (#29431)

Signed-off-by: mgoin <mgoin64@gmail.com>

vllm/v1/worker/gpu_worker.py

@@ -387,6 +387,19 @@ class Worker(WorkerBase):
     def get_kv_cache_spec(self) -> dict[str, KVCacheSpec]:
         return self.model_runner.get_kv_cache_spec()
 
+    def update_max_model_len(self, max_model_len: int) -> None:
+        """Update max_model_len after auto-fit to GPU memory.
+
+        This is called when max_model_len=-1 is used and the engine
+        automatically determines the maximum context length that fits
+        in GPU memory. Workers need to update their cached max_model_len
+        to match the engine's decision.
+        """
+        self.model_config.max_model_len = max_model_len
+        if self.model_runner is not None:
+            self.model_runner.max_model_len = max_model_len
+        logger.debug("Updated max_model_len to %d", max_model_len)
+
     def initialize_from_config(self, kv_cache_config: KVCacheConfig) -> None:
         """Allocate GPU KV cache with the specified kv_cache_config."""