comaniac
/

Mixtral-8x7B-Instruct-v0.1-FP8-v1

Text Generation

text-generation-inference

Inference Endpoints

Model card Files Files and versions Community

Mixtral-8x7B-Instruct-v0.1-FP8-v1 / README.md

comaniac's picture

Create README.md

2b2872a verified 4 months ago

|

history blame contribute delete

No virus

247 Bytes

Mixtral-8x7B-Instruct-v0.1-FP8-v1

Weights and activations are per-tensor quantized to float8_e4m3.
Quantization with AutoFP8.
Calibration dataset: Ultrachat (mgoin/ultrachat_2k)
Samples: 1024
Sequence length: 8192

Evaluation

TBA