README.md

Conversation SFT Dataset

Dataset processed from local .txt file with special tokens for user/assistant conversation.

Special Tokens

<|user_start|>: User input start <|user_end|>: User input end <|assistant_start|>: Assistant response start <|assistant_end|>: Assistant response end

Format

<|user_start|>User message<|user_end|><|assistant_start|>Assistant response<|assistant_end|>

Splits

• full: 2,204,908 samples
• train: 2,193,908 samples
• eval: 1,000 samples
• test: 10,000 samples