You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In RAFT, the jsonl file format after conversion using "format.py" is as follows:
{"messages":[{"content":"When was the University of California, Berkeley established?","role":"user"},{"content":"To answer the question, we need to find the information related to the establishment date of the University of California, Berkeley in the provided context. \n\nThe relevant information is found in the line ##begin_quote##"Established March 23, 1868[1]"##end_quote##. This line directly states the date when the University of California, Berkeley was established.\n\n: The University of California, Berkeley was established on March 23, 1868.","role":"assistant"}]}
But the file format used for fine-tuning in the paper is as follows, which centrally contains question, context, instruction and CoT Answer:
What format was used for the final fine-tuning of LLaMA2-7B in RAFT? Does it include context or other content?
The text was updated successfully, but these errors were encountered:
In RAFT, the jsonl file format after conversion using "format.py" is as follows:
{"messages":[{"content":"When was the University of California, Berkeley established?","role":"user"},{"content":"To answer the question, we need to find the information related to the establishment date of the University of California, Berkeley in the provided context. \n\nThe relevant information is found in the line ##begin_quote##"Established March 23, 1868[1]"##end_quote##. This line directly states the date when the University of California, Berkeley was established.\n\n: The University of California, Berkeley was established on March 23, 1868.","role":"assistant"}]}
But the file format used for fine-tuning in the paper is as follows, which centrally contains question, context, instruction and CoT Answer:
![image](https://private-user-images.githubusercontent.com/22972165/333532086-fe096015-b4ae-40a1-a900-7aab33f112c7.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.sfT2EKyqGWMwpTmF9jIubytgVD5m4FM7houmUw731u4)
What format was used for the final fine-tuning of LLaMA2-7B in RAFT? Does it include context or other content?
The text was updated successfully, but these errors were encountered: