初始化项目,由ModelHub XC社区提供模型

Model: anime-sh/qwen3-8b-rmu-bm25-10b-rebuttal
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-09-26 10:33:16 +08:00
commit c98cee00ac
11 changed files with 336 additions and 0 deletions

6
train.log Normal file
View File

@@ -0,0 +1,6 @@
2026-07-24 02:13:14,694 INFO cleanslate.fsid.training.trainer: [RANK 0] FSID train: model=Qwen/Qwen3-8B objective=rmu run_dir=results/rebuttal/train/qwen3-8b-rmu-bm25-10b/rmu_Qwen3-8B_20260724_021302
2026-07-24 02:13:14,694 INFO cleanslate.fsid.training.trainer: [RANK 0] Creating frozen reference model for rmu (dtype=torch.bfloat16)
2026-07-24 02:13:20,360 INFO cleanslate.fsid.training.trainer: [RANK 0] Starting training: objective=rmu epochs=2
2026-07-24 02:13:20,775 INFO cleanslate.fsid.training.trainer: [RANK 0] RMU control_vec init: dim=4096 seed=0 rank=0 world_size=1 per_rank_norm=2.000000
2026-07-24 02:21:03,076 INFO cleanslate.fsid.training.trainer: [RANK 0] Saved checkpoint to results/rebuttal/train/qwen3-8b-rmu-bm25-10b/rmu_Qwen3-8B_20260724_021302
2026-07-24 02:21:03,080 INFO cleanslate.fsid.training.trainer: [RANK 0] Training complete (426.1s)