Downloads · 30 days
0
scidecc-qwen35-9b-train
Chenhangcui/scidecc-qwen35-9b-train
scidecc-qwen35-9b-train is a machine learning model from Chenhangcui. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as other.
Access
Public
Updated Oct 12, 2026
Repo size
54.4 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors36.3 GB · 99%
At a glance
- License
- other
- Access
- Public
- Created
- Sep 23, 2026
- Updated
- Oct 12, 2026
- SHA
- fd43912f
Base models
Model card
From the Hugging Face model README
SciDeCC Protocol-A — Qwen3.5-9B final training checkpoints
z`。
目录
| 路径 | 是什么 | 训练步 | 服务器原路径 |
|---|---|---|---|
rft_final/ | 最终 RFT：Qwen3.5-9B 全参 SFT，语料 = 3,368 条通过确定性评分器的专家 Protocol-A 轨迹 / 1,032 题（_rft_traces_v2/rft_train_v2.jsonl） | 46（2 epoch） | ckpt/qwen35_9b_sft_v2corpus |
dapo_final/ | 最终 DAPO：在 rft_final 上跑 DAPO，取 checkpoint-30 | 30 / 48 | ckpt/qwen35_9b_dapo_v2corpus/checkpoint-30 |
dapo_final/rollouts/ | DAPO 训练时每步的 completions（prompt、生成、reward），parquet，48 个文件 = 全部 48 步 | 1–48 | ckpt/qwen35_9b_dapo_v2corpus/completions/ |
*/train_state/checkpoint-*/ | 每个保存点的 trainer_state.json（loss / reward / clipped_ratio 曲线）、training_args.bin、scheduler.pt、config | — | 同上 |
*/vllm_wrapper/ | vLLM 可加载所需的外层 Qwen3_5Config 与预处理配置 | — | ckpt/*_vllm/ |
lora_gen1/rft_final/、lora_gen1/dapo_final/ | 第一代（LoRA r=32，底座上直接训）的 RFT 与 DAPO adapter，EXP_LOG 里 rft134 / dapo81920 行引用的就是它们 | RFT 134 / DAPO 33 | ckpt/qwen35_9b_rft_full、ckpt/qwen35_9b_dapo_full |
- License
- other
- Created
- Sep 23, 2026
- Updated
- Oct 12, 2026