4 Commits

Author	SHA1	Message	Date
kibitzing	42b05eea7e	Add guard against division by zero in chat_sft when num_tokens is 0	2025-10-15 13:24:00 +00:00
kibitzing	f5001141ec	Revert model source to mid	2025-10-15 10:29:49 +00:00
kibitzing	b48d210795	Fix gradient accumulation for variable length sequences	2025-10-15 08:56:58 +00:00
karpathy	3a5e0bc50b	initial commit	2025-10-13 06:49:24 -07:00