File size: 763 Bytes
7ce056f ac9c768 7d65ce3 7ce056f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 |
---
license: apache-2.0
train: false
inference: false
pipeline_tag: text-generation
---
## Mixtral-8x7B-v0.1-hf-4bit_g64-HQQ
This is a version of the Mixtral-8x7B-v0.1 model (https://huggingface.co/mistralai/Mixtral-8x7B-v0.1) quantized to 4-bit via Half-Quadratic Quantization (HQQ).
### Basic Usage
To run the model, install the HQQ library from https://github.com/mobiusml/hqq and use it as follows:
``` Python
model_id = 'mobiuslabsgmbh/Mixtral-8x7B-v0.1-hf-4bit_g64-HQQ/'
#Load the model
from hqq.engine.hf import HQQModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = HQQModelForCausalLM.from_quantized(model_id)
#Optional
from hqq.core.quantize import *
HQQLinear.set_backend(HQQBackend.PYTORCH_COMPILE)
```
|