4 Commits

Author	SHA1	Message	Date
bingshuo	5a9c2db952	distill_mother.py v7: fix OOM - expandable_segments:True + empty_cache between epochs and model loading	2026-05-18 22:59:27 +08:00
bingshuo	7d53003fb1	distill_mother.py v6: fix vocab_size mismatch (7B=152064, 1.5B=151936)	2026-05-18 21:16:01 +08:00
bingshuo	9e7bc2000a	fix: 蒸馏保存段增加 config.json + generation_config.json + tokenizer 三重修复	2026-05-18 18:06:27 +08:00
bingshuo	8f31ba4c3a	D101 新增：母模型→1.5B霜砚蒸馏脚本	2026-05-18 13:32:14 +08:00