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

Rename prefill_token_logprobs -> input_token_logprobs; decode_token_logprobs -> output_token_logprobs (#776)

Browse Source
This commit is contained in:
Lianmin Zheng
2024-07-27 19:50:34 -07:00
committed by GitHub
parent 0a409bd438
commit 30db99b3d9
16 changed files with 188 additions and 184 deletions
Download Patch File Download Diff File Expand all files Collapse all files

66
python/sglang/srt/openai_api/adapter.py
View File

@@ -140,29 +140,29 @@ async def v1_completions(tokenizer_manager, raw_request: Request):
if request.logprobs:
# The first chunk and echo is enabled.
if not stream_buffer and request.echo:
prefill_token_logprobs = content["meta_info"][
"prefill_token_logprobs"
input_token_logprobs = content["meta_info"][
"input_token_logprobs"
]
prefill_top_logprobs = content["meta_info"][
"prefill_top_logprobs"
input_top_logprobs = content["meta_info"][
"input_top_logprobs"
]
else:
prefill_token_logprobs = None
prefill_top_logprobs = None
input_token_logprobs = None
input_top_logprobs = None
logprobs = to_openai_style_logprobs(
prefill_token_logprobs=prefill_token_logprobs,
prefill_top_logprobs=prefill_top_logprobs,
decode_token_logprobs=content["meta_info"][
"decode_token_logprobs"
input_token_logprobs=input_token_logprobs,
input_top_logprobs=input_top_logprobs,
output_token_logprobs=content["meta_info"][
"output_token_logprobs"
][n_prev_token:],
decode_top_logprobs=content["meta_info"][
"decode_top_logprobs"
output_top_logprobs=content["meta_info"][
"output_top_logprobs"
][n_prev_token:],
)
n_prev_token = len(
content["meta_info"]["decode_token_logprobs"]
content["meta_info"]["output_token_logprobs"]
)
else:
logprobs = None
@@ -218,17 +218,17 @@ async def v1_completions(tokenizer_manager, raw_request: Request):
if request.logprobs:
if request.echo:
prefill_token_logprobs = ret_item["meta_info"]["prefill_token_logprobs"]
prefill_top_logprobs = ret_item["meta_info"]["prefill_top_logprobs"]
input_token_logprobs = ret_item["meta_info"]["input_token_logprobs"]
input_top_logprobs = ret_item["meta_info"]["input_top_logprobs"]
else:
prefill_token_logprobs = None
prefill_top_logprobs = None
input_token_logprobs = None
input_top_logprobs = None
logprobs = to_openai_style_logprobs(
prefill_token_logprobs=prefill_token_logprobs,
prefill_top_logprobs=prefill_top_logprobs,
decode_token_logprobs=ret_item["meta_info"]["decode_token_logprobs"],
decode_top_logprobs=ret_item["meta_info"]["decode_top_logprobs"],
input_token_logprobs=input_token_logprobs,
input_top_logprobs=input_top_logprobs,
output_token_logprobs=ret_item["meta_info"]["output_token_logprobs"],
output_top_logprobs=ret_item["meta_info"]["output_top_logprobs"],
)
else:
logprobs = None
@@ -401,10 +401,10 @@ async def v1_chat_completions(tokenizer_manager, raw_request: Request):
def to_openai_style_logprobs(
prefill_token_logprobs=None,
decode_token_logprobs=None,
prefill_top_logprobs=None,
decode_top_logprobs=None,
input_token_logprobs=None,
output_token_logprobs=None,
input_top_logprobs=None,
output_top_logprobs=None,
):
ret_logprobs = LogProbs()
@@ -425,13 +425,13 @@ def to_openai_style_logprobs(
else:
ret_logprobs.top_logprobs.append(None)
if prefill_token_logprobs is not None:
append_token_logprobs(prefill_token_logprobs)
if decode_token_logprobs is not None:
append_token_logprobs(decode_token_logprobs)
if prefill_top_logprobs is not None:
append_top_logprobs(prefill_top_logprobs)
if decode_top_logprobs is not None:
append_top_logprobs(decode_top_logprobs)
if input_token_logprobs is not None:
append_token_logprobs(input_token_logprobs)
if output_token_logprobs is not None:
append_token_logprobs(output_token_logprobs)
if input_top_logprobs is not None:
append_top_logprobs(input_top_logprobs)
if output_top_logprobs is not None:
append_top_logprobs(output_top_logprobs)
return ret_logprobs