24 days ago
-
README: SFT 16k adapters and zero-shot vs SFT metrics (510 test cases); mark old onboarding guide as stale
-
evaluations: SFT 16k vs zero-shot ROUGE on 510 test cases (metrics only, no case text)
-
rosy-thesis-sft-16k-gemma4_12b: adapter + epoch checkpoints 575, 1150, 1725, 2300, 2875
-
rosy-thesis-sft-16k-sahabatai_9b: adapter + epoch checkpoints 575, 1150, 1725, 2300, 2875
25 days ago
-
rosy-thesis-sft-16k-qwen35_9b: adapter + epoch checkpoints 575, 1150, 1725, 2300, 2875
-
rosy-thesis-sft-16k-qwen25_7b: adapter + epoch checkpoints 575, 1150, 1725, 2300, 2875
27 days ago
-
qwen25_7b: status
-
qwen25_7b: adapter + trainer state
-
qwen25_7b: checkpoint-134
-
qwen25_7b: checkpoint-75
-
qwen25_7b: checkpoint-50
-
qwen25_7b: status
-
qwen25_7b: checkpoint-25
-
remove sync test
-
_synctest: status
-
_synctest: adapter + trainer state
-
_synctest: checkpoint-1
-
feat(eval): upload benchmark summary comparison CSV table
-
feat(model): upload Gemma 4 12B BF16 LoRA adapter weights, checkpoints & training artifacts
-
feat(eval): add eval1 completions (LoRA & baseline) for Gemma 4 12B-it (BF16 LoRA)
-
feat(eval): update Rosy Thesis 5-epoch SFT benchmark and completions (including Gemma 4 12B BF16)
-
feat(eval): add Rosy Thesis 5-epoch SFT vs baseline zero-shot evaluation benchmark & completions
-
feat(eval): add eval1 completions (LoRA & baseline) for Sahabat-AI Gemma2 9B
-
feat(eval): add eval1 completions (LoRA & baseline) for Gemma 4 12B-it
-
feat(eval): add eval1 completions (LoRA & baseline) for Qwen3.5-9B
-
feat(eval): add eval1 completions (LoRA & baseline) for Qwen2.5-7B-Instruct
-
feat(eval): add Rosy Thesis 5-epoch SFT vs baseline zero-shot evaluation benchmark & completions
-
feat: upload SFT adapter weights & artifacts for qwen25_7b_test (rosy thesis 5-epoch)
-
feat: upload Sahabat-AI Gemma2 9B LoRA SFT final weights + all epoch checkpoints
-
feat: upload Gemma 4 12B-it QLoRA SFT final weights + all epoch checkpoints
-
feat: upload Qwen3.5-9B LoRA SFT final weights + all epoch checkpoints
-
feat: upload Qwen2.5-7B-Instruct LoRA SFT final weights + all 5 epoch checkpoints
-
docs: add Sahabat-AI Gemma 2 9B and Qwen 2.5 7B evaluation results & samples to README
-
eval: update master ROUGE benchmark comparison table
-
eval: upload 52 test cases samples for Sahabat-AI Gemma 2 9B (5-ep)
-
eval: update master ROUGE benchmark comparison table
-
feat: upload SFT v2 5-epoch LoRA adapter for Sahabat-AI Gemma 2 9B (5-ep)
-
Upload pre-built vLLM wheel for Python 3.13 and CUDA 12.8 (A100)
-
eval: update master ROUGE benchmark comparison table
-
eval: upload 52 test cases samples for Qwen 3.5 9B (5-ep)
-
feat: upload SFT v2 5-epoch LoRA adapter for Qwen 3.5 9B (5-ep)
-
feat: upload SFT v2 5-epoch LoRA adapter for Qwen 3.5 9B (5-ep)
-
eval: update master ROUGE benchmark comparison table
-
eval: upload 52 test cases samples for Qwen 2.5 7B-Instruct (5-ep)
-
feat: upload SFT v2 5-epoch LoRA adapter for Qwen 2.5 7B-Instruct (5-ep)
-
docs: sync onboarding guide with substantive evaluation findings and LLM-as-a-judge roadmap
-
eval: update comparative ROUGE benchmark table with completed Qwen 3.8 27B (52/52 cases)
-
eval(qwen38-27b): complete all 52/52 test cases evaluation
-
eval(qwen38-27b): sync progress 1314/52 test cases
-
eval(qwen38-27b): sync progress 1204/52 test cases
