fix: chat API logprobs format #1788
Open
+59
−18
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Demo Request{
"model": "gpt-3.5-turbo",
"messages": [
{
"role": "user",
"content": "What is the capital of France?"
}
],
"logprobs": true,
"top_logprobs": 10,
"max_tokens": 5
}Response{
"id": "chatcmpl-1898ccce-2bf6-431c-b9e0-2a82e90a9604",
"object": "chat.completion",
"created": 1728184671,
"model": "gpt-3.5-turbo",
"choices": [
{
"index": 0,
"message": {
"content": "The capital of France is",
"role": "assistant"
},
"logprobs": {
"content": [
{
"token": "The",
"logprob": -0.008244173601269722,
"bytes": null,
"top_logprobs": [
{
"token": "The",
"logprob": -0.008244173601269722,
"bytes": null
},
{
"token": "Paris",
"logprob": -5.3227219581604,
"bytes": null
},
{
"token": "Sure",
"logprob": -5.770838260650635,
"bytes": null
},
{
"token": "Answer",
"logprob": -9.54023265838623,
"bytes": null
},
{
"token": "Yes",
"logprob": -9.896768569946289,
"bytes": null
},
{
"token": "France",
"logprob": -10.62641429901123,
"bytes": null
},
{
"token": " The",
"logprob": -11.367059707641602,
"bytes": null
},
{
"token": "According",
"logprob": -11.45943546295166,
"bytes": null
},
{
"token": "**",
"logprob": -11.586193084716797,
"bytes": null
},
{
"token": " Paris",
"logprob": -11.59852409362793,
"bytes": null
}
]
},
{
"token": " capital",
"logprob": -0.0005453529884107411,
"bytes": null,
"top_logprobs": [
{
"token": " capital",
"logprob": -0.0005453529884107411,
"bytes": null
},
{
"token": " Capital",
"logprob": -7.571288108825684,
"bytes": null
},
{
"token": " city",
"logprob": -11.57780647277832,
"bytes": null
},
{
"token": " current",
"logprob": -12.473557472229004,
"bytes": null
},
{
"token": " correct",
"logprob": -12.674555778503418,
"bytes": null
},
{
"token": " ",
"logprob": -12.77519416809082,
"bytes": null
},
{
"token": " answer",
"logprob": -12.833593368530273,
"bytes": null
},
{
"token": " French",
"logprob": -13.656529426574707,
"bytes": null
},
{
"token": " Paris",
"logprob": -13.73013687133789,
"bytes": null
},
{
"token": " **",
"logprob": -13.916248321533203,
"bytes": null
}
]
},
{
"token": " of",
"logprob": -0.019254328683018684,
"bytes": null,
"top_logprobs": [
{
"token": " of",
"logprob": -0.019254328683018684,
"bytes": null
},
{
"token": " city",
"logprob": -3.9625728130340576,
"bytes": null
},
{
"token": " and",
"logprob": -10.33055305480957,
"bytes": null
},
{
"token": " ",
"logprob": -12.015106201171875,
"bytes": null
},
{
"token": " is",
"logprob": -12.049043655395508,
"bytes": null
},
{
"token": " City",
"logprob": -12.161520957946777,
"bytes": null
},
{
"token": " o",
"logprob": -12.770393371582031,
"bytes": null
},
{
"token": " cities",
"logprob": -14.372736930847168,
"bytes": null
},
{
"token": " của",
"logprob": -14.63923454284668,
"bytes": null
},
{
"token": " ",
"logprob": -14.65132999420166,
"bytes": null
}
]
},
{
"token": " France",
"logprob": -0.0000252720492426306,
"bytes": null,
"top_logprobs": [
{
"token": " France",
"logprob": -0.0000252720492426306,
"bytes": null
},
{
"token": " the",
"logprob": -11.084362030029297,
"bytes": null
},
{
"token": " ",
"logprob": -12.06197738647461,
"bytes": null
},
{
"token": "France",
"logprob": -12.9952974319458,
"bytes": null
},
{
"token": " French",
"logprob": -13.759483337402344,
"bytes": null
},
{
"token": " is",
"logprob": -15.239158630371094,
"bytes": null
},
{
"token": " **",
"logprob": -15.40572452545166,
"bytes": null
},
{
"token": " france",
"logprob": -15.767807960510254,
"bytes": null
},
{
"token": " ",
"logprob": -16.346908569335938,
"bytes": null
},
{
"token": " Frankreich",
"logprob": -17.035612106323242,
"bytes": null
}
]
},
{
"token": " is",
"logprob": -0.000060437283536884934,
"bytes": null,
"top_logprobs": [
{
"token": " is",
"logprob": -0.000060437283536884934,
"bytes": null
},
{
"token": " ",
"logprob": -9.920828819274902,
"bytes": null
},
{
"token": ",",
"logprob": -12.151354789733887,
"bytes": null
},
{
"token": " was",
"logprob": -13.53709602355957,
"bytes": null
},
{
"token": " ",
"logprob": -14.004632949829102,
"bytes": null
},
{
"token": " in",
"logprob": -14.70918083190918,
"bytes": null
},
{
"token": " **",
"logprob": -14.768845558166504,
"bytes": null
},
{
"token": " the",
"logprob": -14.776985168457031,
"bytes": null
},
{
"token": " ",
"logprob": -14.940979957580566,
"bytes": null
},
{
"token": " Is",
"logprob": -14.942352294921875,
"bytes": null
}
]
}
],
"refusal": null
},
"finish_reason": "length"
}
],
"usage": {
"prompt_tokens": 29,
"completion_tokens": 5,
"total_tokens": 34
}
} |
This was referenced Oct 6, 2024
lukestanley
reviewed
Oct 14, 2024
| uvicorn --factory llama.server:app --host ${HOST} --port ${PORT} | ||
| python llama_cpp/server --model ${MODEL} |
There was a problem hiding this comment.
Why this change? It is unrelated to logprobs? @domdomegg
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
The OpenAI compatible server should match the response structure of the OpenAI API for chat completions. Unfortunately there is a discrepancy with the format of logprobs: we return the logprobs format for the completions API, rather than the chat completions API.
This PR:
_convert_text_completion_logprobs_to_chatwhich is used in the chat completion responses to convert the logprobs to the new API formatIssues fixed
Fixes #1787