ArabicFP4 — FP8 with Arabic calibration
Collection
21 items • Updated • 1
gemma2-9b · FP8 · calibrated on Modern Standard Arabic
Post-training W8A8 quantization with NVIDIA ModelOpt (FP8_DEFAULT_CFG).
| base model | google/gemma-2-9b-it |
| base dtype | float16 |
| precision | FP8 (W8A8) |
| calibration variety | Modern Standard Arabic |
| calibration file | calib1_msa.txt |
| calibration samples | 128 dialogues, truncated at 128 tokens |
| calibration source | Almheiri/ArabCulture-Dialogue rev 9acd60cbbb4f, seed 1448 |
| weight MSE | 3.164e-08 |
| activation quantizers calibrated | 294/294 |
| environment | torch 2.8.0+cu128 · transformers 4.57.6 · modelopt 0.45.0 |
FP8 weight scales are data-free. What the calibration variety changes is the static
per-tensor activation scales, exported as input_scale. Sibling checkpoints
-fp8-msa, -fp8-gulf and -fp8-mixed differ in exactly that and nothing else.
Not loadable by plain transformers — config.json declares quantization type modelopt.
vllm serve NouraAlqasim/gemma2-9b-fp8-msa --quantization modelopt
- Downloads last month
- 14
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support