[Test] Add parameters to SRTRunner (#12227)
This commit is contained in:
@@ -515,6 +515,9 @@ class SRTRunner:
|
|||||||
disable_cuda_graph: bool = False,
|
disable_cuda_graph: bool = False,
|
||||||
disable_radix_cache: bool = False,
|
disable_radix_cache: bool = False,
|
||||||
chunked_prefill_size: Optional[int] = None,
|
chunked_prefill_size: Optional[int] = None,
|
||||||
|
context_length: Optional[int] = None,
|
||||||
|
max_total_tokens: Optional[int] = None,
|
||||||
|
page_size: Optional[int] = None,
|
||||||
dp_size: int = 1,
|
dp_size: int = 1,
|
||||||
tokenizer_path: Optional[str] = None,
|
tokenizer_path: Optional[str] = None,
|
||||||
mem_fraction_static: float = 0.65,
|
mem_fraction_static: float = 0.65,
|
||||||
@@ -571,6 +574,9 @@ class SRTRunner:
|
|||||||
disable_cuda_graph=disable_cuda_graph,
|
disable_cuda_graph=disable_cuda_graph,
|
||||||
disable_radix_cache=disable_radix_cache,
|
disable_radix_cache=disable_radix_cache,
|
||||||
chunked_prefill_size=chunked_prefill_size,
|
chunked_prefill_size=chunked_prefill_size,
|
||||||
|
context_length=context_length,
|
||||||
|
max_total_tokens=max_total_tokens,
|
||||||
|
page_size=page_size,
|
||||||
enable_dp_attention=enable_dp_attention,
|
enable_dp_attention=enable_dp_attention,
|
||||||
dp_size=dp_size,
|
dp_size=dp_size,
|
||||||
tokenizer_path=tokenizer_path,
|
tokenizer_path=tokenizer_path,
|
||||||
|
|||||||
Reference in New Issue
Block a user