GPT-1900 Drafts
Collection
Experimental and intermediate GPT-1900 checkpoints. Working artifacts, not for general use. • 49 items • Updated
GPT-1900 D34 Math RL
3.29B parameter GPT-1900 with math RL. GSM8K (30%) + MATH (70%), SymPy verification, format reward. Chain: base -> physics CLM expanded -> v3 SFT safe -> R1 SFT -> math RL.
Training
- Training: Math RL (step 60), from r1-reasoning-sft
- Parameters: 3.29B
- Architecture: Custom GPT with RoPE, QK-norm, ReLU², value embeddings
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support