Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-27 06:39:25 +01:00
Code Issues Packages Projects Releases Wiki Activity

server : fix tests expecting old repeat penalty

Browse Source
This commit is contained in:
Francis Couture-Harpin 2024-03-19 17:12:28 -04:00
parent d8b009a945
commit 9a424a3872
1 changed files with 5 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
examples/server/tests/features/steps/steps.py
View File

@ -718,7 +718,9 @@ async def request_completion(prompt,
"prompt": prompt, "prompt": prompt,
"input_suffix": prompt_suffix, "input_suffix": prompt_suffix,
"n_predict": n_predict if n_predict is not None else -1, "n_predict": n_predict if n_predict is not None else -1,
"seed": seed if seed is not None else 42 "seed": seed if seed is not None else 42,
# TODO: change the expected outputs to match no repeat penalty
"repeat_penalty": 1.1, # use old defaults to match expected outputs
}, },
headers=headers, headers=headers,
timeout=3600) as response: timeout=3600) as response:
@ -763,6 +765,8 @@ async def oai_chat_completions(user_prompt,
"max_tokens": n_predict, "max_tokens": n_predict,
"stream": enable_streaming, "stream": enable_streaming,
"seed": seed "seed": seed
# TODO: change the expected outputs to match no repeat penalty
"repeat_penalty": 1.1, # use old defaults to match expected outputs
} }
completion_response = { completion_response = {
'content': '', 'content': '',