[Refactor] Pass Renderer to Input Processor (#34329)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-02-11 19:38:11 -08:00
committed by GitHub
parent ced2a92f40
commit b96f7314b4
20 changed files with 107 additions and 106 deletions
@@ -125,6 +125,7 @@ class TestInitializeToolSessions:
engine_client = MagicMock()
model_config = MagicMock()
model_config.max_model_len = 100
model_config.hf_config.model_type = "test"
model_config.get_diff_sampling_param.return_value = {}
engine_client.model_config = model_config
@@ -212,6 +213,7 @@ class TestValidateGeneratorInput:
engine_client = MagicMock()
model_config = MagicMock()
model_config.max_model_len = 100
model_config.hf_config.model_type = "test"
model_config.get_diff_sampling_param.return_value = {}
engine_client.model_config = model_config
@@ -231,9 +233,6 @@ class TestValidateGeneratorInput:
chat_template_content_format="auto",
)
# Set max_model_len for testing
instance.max_model_len = 100
return instance
def test_validate_generator_input(self, serving_responses_instance):
@@ -507,7 +507,8 @@ def test_apc_single_prompt_block_align_alignment(
vllm_runner_kwargs["enable_prefix_caching"] = True
with vllm_runner(**vllm_runner_kwargs) as vllm_model:
# Retrieve the default mamba state block size
mamba_block_size = vllm_model.llm.llm_engine.cache_config.mamba_block_size
vllm_config = vllm_model.llm.llm_engine.vllm_config
mamba_block_size = vllm_config.cache_config.mamba_block_size
# In case the hybrid model does not have the
# "mamba_block_size" assume a fixed constant
@@ -660,7 +661,8 @@ def test_apc_multiple_prompts_block_align_alignment(
vllm_runner_kwargs["enable_prefix_caching"] = True
with vllm_runner(**vllm_runner_kwargs) as vllm_model:
# Retrieve the default mamba state block size
mamba_block_size = vllm_model.llm.llm_engine.cache_config.mamba_block_size
vllm_config = vllm_model.llm.llm_engine.vllm_config
mamba_block_size = vllm_config.cache_config.mamba_block_size
# In case the hybrid model does not have the
# "mamba_block_size" assume a fixed constant
@@ -25,7 +25,8 @@ def test_classify_models(
with vllm_runner(
model, max_model_len=512, dtype=dtype, enable_prefix_caching=True
) as vllm_model:
cache_config = vllm_model.llm.llm_engine.cache_config
vllm_config = vllm_model.llm.llm_engine.vllm_config
cache_config = vllm_config.cache_config
assert cache_config.enable_prefix_caching
# First Run
@@ -74,7 +75,8 @@ def test_embed_models(
max_model_len=None,
enable_prefix_caching=True,
) as vllm_model:
cache_config = vllm_model.llm.llm_engine.cache_config
vllm_config = vllm_model.llm.llm_engine.vllm_config
cache_config = vllm_config.cache_config
assert cache_config.enable_prefix_caching
# First Run
@@ -106,5 +108,6 @@ def test_non_causal_models(
hf_runner, vllm_runner, example_prompts, model: str, dtype: str
) -> None:
with vllm_runner(model, max_model_len=512, dtype=dtype) as vllm_model:
cache_config = vllm_model.llm.llm_engine.cache_config
vllm_config = vllm_model.llm.llm_engine.vllm_config
cache_config = vllm_config.cache_config
assert not cache_config.enable_prefix_caching
+2 -1
View File
@@ -161,7 +161,8 @@ def test_pooling_prefix_cache(vllm_runner, monkeypatch):
assert chunks[0] <= prompt1_len
assert chunks[0] < prompt2_len
cache_config = llm.get_llm().llm_engine.cache_config
vllm_config = llm.get_llm().llm_engine.vllm_config
cache_config = vllm_config.cache_config
print(f"{cache_config=}")
# Prefixes are cached in blocks
assert (prompt2_len - chunks[0]) % cache_config.block_size == 0
+2 -1
View File
@@ -311,7 +311,8 @@ def test_get_logprobs_and_prompt_logprobs(
temperature: "temperature" sampling parameter
example_prompts: example prompt fixture
"""
do_apc = vllm_model.llm.llm_engine.cache_config.enable_prefix_caching
vllm_config = vllm_model.llm.llm_engine.vllm_config
do_apc = vllm_config.cache_config.enable_prefix_caching
if do_apc and (temperature < 2.0 or batch_logprobs_composition != SAMPLE_PROMPT):
# Skip some test-cases to save time.
pytest.skip()