tokenizer: fix bos_token to <|im_start|> (was duplicate-keyed null) 79d0501 verified AntonXue commited on about 23 hours ago
tokenizer: set bos_token=<|im_start|> (matches model config bos_token_id=151644) 781c80b verified AntonXue commited on about 23 hours ago
tokenizer: set bos_token=<|im_start|> (matches model config bos_token_id=151644) f82c77f verified AntonXue commited on about 23 hours ago
README: report 50k training steps (matches truncated log + SWA window) 0cdeb73 verified AntonXue commited on about 24 hours ago
Initial release: SWA-averaged Stage-1 endpoint (steps 46k-50k, 1k stride) 4a4735e verified AntonXue commited on about 24 hours ago