Skip to content

Commit d161b64

Browse files
authored
P2L Extra body for cost param (#3666)
1 parent 1ffd4a6 commit d161b64

File tree

2 files changed

+9
-0
lines changed

2 files changed

+9
-0
lines changed

fastchat/serve/api_provider.py

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -23,6 +23,7 @@ def get_api_provider_stream_iter(
2323
top_p,
2424
max_new_tokens,
2525
state,
26+
extra_body=None,
2627
):
2728
if model_api_dict["api_type"] == "openai":
2829
if model_api_dict.get("vision-arena", False):
@@ -256,6 +257,7 @@ def get_api_provider_stream_iter(
256257
max_new_tokens,
257258
api_base=model_api_dict["api_base"],
258259
api_key=model_api_dict["api_key"],
260+
extra_body=extra_body,
259261
)
260262
else:
261263
raise NotImplementedError()
@@ -431,6 +433,7 @@ def p2l_api_stream_iter(
431433
max_new_tokens,
432434
api_base=None,
433435
api_key=None,
436+
extra_body=None,
434437
):
435438
import openai
436439

@@ -459,6 +462,7 @@ def p2l_api_stream_iter(
459462
"temperature": None,
460463
"top_p": None,
461464
"max_new_tokens": max_new_tokens,
465+
"extra_body": extra_body,
462466
}
463467
logger.info(f"==== request ====\n{gen_params}")
464468

@@ -467,6 +471,7 @@ def p2l_api_stream_iter(
467471
messages=messages,
468472
max_tokens=max_new_tokens,
469473
stream=True,
474+
extra_body=extra_body,
470475
)
471476
text = ""
472477
for chunk_idx, chunk in enumerate(res):

fastchat/serve/gradio_web_server.py

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -526,6 +526,8 @@ def bot_response(
526526
if not custom_system_prompt:
527527
conv.set_system_message("")
528528

529+
extra_body = None
530+
529531
if use_recommended_config:
530532
recommended_config = model_api_dict.get("recommended_config", None)
531533
if recommended_config is not None:
@@ -534,6 +536,7 @@ def bot_response(
534536
max_new_tokens = recommended_config.get(
535537
"max_new_tokens", max_new_tokens
536538
)
539+
extra_body = recommended_config.get("extra_body", None)
537540

538541
stream_iter = get_api_provider_stream_iter(
539542
conv,
@@ -543,6 +546,7 @@ def bot_response(
543546
top_p,
544547
max_new_tokens,
545548
state,
549+
extra_body=extra_body,
546550
)
547551

548552
html_code = ' <span class="cursor"></span> '

0 commit comments

Comments
 (0)