feat: update auto_choose_speculative_params (#9470)
This commit is contained in:
@@ -2411,8 +2411,12 @@ def auto_choose_speculative_params(self: ServerArgs):
|
|||||||
if arch in ["LlamaForCausalLM"]:
|
if arch in ["LlamaForCausalLM"]:
|
||||||
# The default value for llama
|
# The default value for llama
|
||||||
return (5, 4, 8)
|
return (5, 4, 8)
|
||||||
elif arch in ["DeepseekV3ForCausalLM", "DeepseekV2ForCausalLM"]:
|
elif arch in [
|
||||||
# The default value for deepseek
|
"DeepseekV3ForCausalLM",
|
||||||
|
"DeepseekV2ForCausalLM",
|
||||||
|
"GptOssForCausalLM",
|
||||||
|
]:
|
||||||
|
# The default value for deepseek and gpt-oss
|
||||||
return (3, 1, 4)
|
return (3, 1, 4)
|
||||||
elif arch in ["Grok1ForCausalLM", "Grok1VForCausalLM"]:
|
elif arch in ["Grok1ForCausalLM", "Grok1VForCausalLM"]:
|
||||||
return (5, 4, 8)
|
return (5, 4, 8)
|
||||||
|
|||||||
Reference in New Issue
Block a user