Files

T

张宗平 357e2e60bd feat(m4): stage 2 SFT training script (LoRA r=16) + smoke (T4.1)

- src/tsmm/train/stage2.py: loads stage1 ckpt (optional), freeze_llm +
  enable_lora(r=16) on q/v_proj, masked LM loss on sft.jsonl, bs=4/grad_accum=8/
  ctx=1024/BF16/gradient checkpointing, ckpt every 2000 steps (encoder+projector
  state + peft LoRA adapter via save_pretrained).
- scripts/train_stage2.sh: wrapper with design defaults.
- Smoke verified: --max_steps 4, LoRA injected, finite loss, peak 3.29 GB
  (≤11GB budget), ckpt+adapter saved.

2026-06-30 03:11:00 +00:00

gen_synthetic.py

feat(m1): offline generation pipeline + CLI (T1.5)

2026-06-29 23:29:04 +08:00

README.md

feat(m1): scaffold tsmm package (T1.1)

2026-06-29 23:21:47 +08:00

train_stage1.sh

feat(m3): stage 1 alignment training script + smoke (T3.2)

2026-06-30 02:22:32 +00:00

train_stage2.sh

feat(m4): stage 2 SFT training script (LoRA r=16) + smoke (T4.1)

2026-06-30 03:11:00 +00:00

README.md

Scripts directory

Offline entrypoints: gen_synthetic.py, train_stage1.sh, train_stage2.sh, run_eval.sh.