File size: 276 Bytes
54ee5f5 53b4f6e 54ee5f5 53b4f6e |
1 2 3 4 5 6 7 8 9 10 11 |
## Mixtral-8x7B-Instruct-v0.1-FP8-v3
* Weights and activations are per-tensor quantized to float8_e4m3.
* Quantization with AutoFP8 with updated activation scaling factor names.
* Calibration dataset: Ultrachat-200k
* Samples: 4096
* Sequence length: 8192
## Evaluation
TBA |