| File | Description |
|---|---|
dfssm_dfw.pt | Frozen 1-bit scaffold (155 MB) |
dfssm_dfw_lora.pt | Int8 LoRA correction (12 MB) |
1# See https://github.com/cs-cmyk/df-ssm for full inference code
2python inference/gpu/df_ssm_inference.py \
3 --scaffold dfssm_dfw.pt \
4 --lora dfssm_dfw_lora.pt| Task | DF-SSM (32M tok, 1 GPU, 6h) | FP16 Teacher (300B+ tok) | Retention | BitMamba-2 (150B tok, multi-GPU) |
|---|---|---|---|---|
| BoolQ | 60.8% | 64.2% | 94.7% | 62.4% |
| PIQA | 67.1% | 73.2% | 91.7% | 68.8% |
| HellaSwag | 41.4% | 59.9% | 69.1% | 45.6% |
| WinoGrande | 54.7% | 60.9% | 89.8% | 52.8% |
| ARC-easy | 50.2% | 64.1% | 78.3% | — |