EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

feat: frequency, min_new_tokens, presence, and repetition penalties (#973)

Browse Source
This commit is contained in:
Juwan Yoo
2024-08-08 04:21:08 -07:00
committed by GitHub
parent 228cf47547
commit ab7875941b
20 changed files with 1898 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/srt/openai_api/adapter.py
View File

@@ -392,10 +392,13 @@ def v1_generate_request(all_requests):
{
"temperature": request.temperature,
"max_new_tokens": request.max_tokens,
"min_new_tokens": request.min_tokens,
"stop": request.stop,
"stop_token_ids": request.stop_token_ids,
"top_p": request.top_p,
"presence_penalty": request.presence_penalty,
"frequency_penalty": request.frequency_penalty,
"repetition_penalty": request.repetition_penalty,
"regex": request.regex,
"n": request.n,
"ignore_eos": request.ignore_eos,
@@ -722,10 +725,13 @@ def v1_chat_generate_request(all_requests, tokenizer_manager):
{
"temperature": request.temperature,
"max_new_tokens": request.max_tokens,
"min_new_tokens": request.min_tokens,
"stop": stop,
"stop_token_ids": request.stop_token_ids,
"top_p": request.top_p,
"presence_penalty": request.presence_penalty,
"frequency_penalty": request.frequency_penalty,
"repetition_penalty": request.repetition_penalty,
"regex": request.regex,
"n": request.n,
}

6
python/sglang/srt/openai_api/protocol.py
View File

@@ -162,6 +162,9 @@ class CompletionRequest(BaseModel):
# Extra parameters for SRT backend only and will be ignored by OpenAI models.
regex: Optional[str] = None
ignore_eos: Optional[bool] = False
min_tokens: Optional[int] = 0
repetition_penalty: Optional[float] = 1.0
stop_token_ids: Optional[List[int]] = Field(default_factory=list)
class CompletionResponseChoice(BaseModel):
@@ -259,6 +262,9 @@ class ChatCompletionRequest(BaseModel):
# Extra parameters for SRT backend only and will be ignored by OpenAI models.
regex: Optional[str] = None
min_tokens: Optional[int] = 0
repetition_penalty: Optional[float] = 1.0
stop_token_ids: Optional[List[int]] = Field(default_factory=list)
class ChatMessage(BaseModel):