5 days ago
-
Replace release weights and refresh evaluation scores
16 days ago
-
Add GGUF (Q8_0) release link near the top
-
Add GGUF (Q8_0) release link near the top
17 days ago
-
Make the live demo prominent
23 days ago
-
Update README_zh.md: simplify intro wording
-
Update README.md: simplify intro wording
-
Remove superseded iter167 predictions
-
Remove superseded iter167 predictions
-
Remove superseded iter167 predictions
-
Remove superseded iter167 predictions
-
Remove superseded iter167 predictions
-
Remove superseded iter167 predictions
-
Remove superseded iter167 predictions
-
Remove superseded iter167 predictions
-
Add greedy/beam dev+devtest predictions for the distillation release
-
Update README_zh.md for the distillation release
-
Update README.md for the distillation release
-
Update weights: three-stage (SFT + quality-remix anneal + 27B-teacher KD) final checkpoint
24 days ago
-
V3.2: Stage2R iter167 weights + rp=1.005 release generation_config + eval preds (dev 28.56/17.86, devtest 29.21/17.51)
-
V3.2 card: Stage2R iter167 scores (dev 28.56/17.86, devtest 29.21/17.51, rp=1.005 release config)
27 days ago
-
Remove repetition_penalty: pure greedy decoding (benchmark configuration)
30 days ago
-
Remove superseded 5ep predictions
-
Remove superseded 5ep predictions
-
Add 4ep devtest predictions
-
Add 4ep dev predictions
-
Update zh model card for 4ep release
-
Switch to epoch-4 checkpoint (sweet spot): dev en-zh 27.56/zh-en 17.17, devtest 28.48/17.01
-
Remove superseded 2ep predictions
-
Remove superseded 2ep predictions
-
Add 5ep devtest predictions
-
Add 5ep devtest predictions
-
Add 5ep dev predictions
-
Update zh model card for 5ep
-
8M-trans 5-epoch: dev en-zh 27.35/zh-en 17.08, devtest 26.97/16.80
-
Haidass-Translate-143M: weights, bilingual card, widget, demo link, eval predictions