haohuiliang/RobustOps-RL-4B-BF16

Text Generation • 4B • Updated 12 days ago • 1.14k • 5

Note Recommended BF16 weights for inference and evaluation; RL step 120. [GitHub: code, setup and evaluation](https://github.com/haohuiliang/RobustOps#user-content-evaluation).

haohuiliang/RobustOps-SFT-4B

Text Generation • 4B • Updated 12 days ago • 891 • 4

Note SFT epoch 2 / step 174, used to initialize RL. [GitHub: training data, setup and recipes](https://github.com/haohuiliang/RobustOps#user-content-training).

haohuiliang/RobustOps-RL-4B

Text Generation • 4B • Updated 12 days ago • 921 • 4

Note RL step-120 weights in original FP32 export precision for further training. [GitHub: training and reproduction](https://github.com/haohuiliang/RobustOps#user-content-training).

Note Execution traces for Base, guided and unguided teacher runs, SFT and RL. [GitHub: training data and reproduction](https://github.com/haohuiliang/RobustOps/tree/main/data).