[Model] Allow users to control skip reading cache per request. (#28194)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
Signed-off-by: wang.yuqi <noooop@126.com>

tests/models/language/pooling/test_extract_hidden_states.py

@@ -11,7 +11,7 @@ from vllm import TokensPrompt
     ["Qwen/Qwen3-0.6B"],
 )
 @torch.inference_mode
-def test_embed_models(hf_runner, vllm_runner, model: str):
+def test_extract_hidden_states(hf_runner, vllm_runner, model: str):
     n_prompt_tokens = [55, 56, 57]
     token_prompts = [[1024 + i for i in range(n)] for n in n_prompt_tokens]
 
@@ -21,7 +21,7 @@ def test_embed_models(hf_runner, vllm_runner, model: str):
         enforce_eager=True,
         runner="pooling",
         enable_chunked_prefill=False,
-        enable_prefix_caching=False,
+        enable_prefix_caching=True,
     ) as vllm_model:
         pooling_outputs = vllm_model.llm.encode(
             [TokensPrompt(prompt_token_ids=t) for t in token_prompts],
@@ -30,4 +30,29 @@ def test_embed_models(hf_runner, vllm_runner, model: str):
 
         for n, output in zip(n_prompt_tokens, pooling_outputs):
             assert len(output.prompt_token_ids) == n
+            assert len(output.outputs.data) == n
             assert output.num_cached_tokens == 0
+
+        # test enable_prefix_caching plus all pooling
+        # we need to skip reading cache at this request by
+        # request.skip_reading_prefix_cache
+        pooling_outputs = vllm_model.llm.encode(
+            [TokensPrompt(prompt_token_ids=t) for t in token_prompts],
+            pooling_task="token_embed",
+        )
+
+        for n, output in zip(n_prompt_tokens, pooling_outputs):
+            assert len(output.prompt_token_ids) == n
+            assert len(output.outputs.data) == n
+            assert output.num_cached_tokens == 0
+
+        # skip_reading_prefix_cache can still write to cache
+        # to accelerate following requests
+        pooling_outputs = vllm_model.llm.encode(
+            [TokensPrompt(prompt_token_ids=t) for t in token_prompts],
+            pooling_task="embed",
+        )
+
+        for n, output in zip(n_prompt_tokens, pooling_outputs):
+            assert len(output.prompt_token_ids) == n
+            assert output.num_cached_tokens > 0