[Misc] unify variable for LLM instance (#20996)
Signed-off-by: Andy Xie <andy.xning@gmail.com>
This commit is contained in:
@@ -39,7 +39,7 @@ def test_gptq_with_dynamic(vllm_runner, model_id: str, use_marlin_kernel: bool,
|
||||
linear_method_cls = GPTQMarlinLinearMethod if use_marlin_kernel else (
|
||||
GPTQLinearMethod)
|
||||
|
||||
for name, submodule in (vllm_model.model.llm_engine.model_executor.
|
||||
for name, submodule in (vllm_model.llm.llm_engine.model_executor.
|
||||
driver_worker.model_runner.model.named_modules()):
|
||||
if name == "lm_head":
|
||||
assert isinstance(submodule.quant_method, linear_method_cls)
|
||||
|
||||
@@ -107,11 +107,11 @@ def test_quark_fp8_parity(vllm_runner):
|
||||
}
|
||||
with (vllm_runner(quark_model_id, **llm_kwargs) as
|
||||
quark_handle, vllm_runner(fp8_model_id, **llm_kwargs) as fp8_handle):
|
||||
quark_model = (quark_handle.model.llm_engine.model_executor.
|
||||
quark_model = (quark_handle.llm.llm_engine.model_executor.
|
||||
driver_worker.model_runner.model)
|
||||
quark_state_dict = quark_model.state_dict()
|
||||
|
||||
fp8_model = (fp8_handle.model.llm_engine.model_executor.driver_worker.
|
||||
fp8_model = (fp8_handle.llm.llm_engine.model_executor.driver_worker.
|
||||
model_runner.model)
|
||||
fp8_state_dict = fp8_model.state_dict()
|
||||
|
||||
|
||||
@@ -111,7 +111,7 @@ def test_custom_quant(vllm_runner, model, monkeypatch):
|
||||
quantization="custom_quant",
|
||||
enforce_eager=True) as llm:
|
||||
|
||||
model = llm.model.llm_engine.model_executor.driver_worker.model_runner.model # noqa: E501
|
||||
model = llm.llm.llm_engine.model_executor.driver_worker.model_runner.model # noqa: E501
|
||||
layer = model.model.layers[0]
|
||||
qkv_proj = layer.self_attn.qkv_proj
|
||||
|
||||
|
||||
Reference in New Issue
Block a user