File size: 693 Bytes
8b9ceac
87f166d
 
771a690
8b9ceac
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
---
base_model:
- meta-llama/Llama-3.2-1B-Instruct
base_model_relation: quantized
---
This is a version of the Llama-3.2-1B-Instruct model quantized to 4-bit via Half-Quadratic Quantization (HQQ): https://mobiusml.github.io/hqq_blog/

To run the model, install the HQQ library from https://github.com/mobiusml/hqq and use it as follows:
``` Python
model_id = 'brunopio/Llama-3.2-1B-Instruct-nbits4-GS64-Axis1-HQQ'

from hqq.engine.hf import HQQModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id)
model     = HQQModelForCausalLM.from_quantized(model_id)
```

*Limitations*: <br>
-Only supports single GPU runtime.<br>
-Not compatible with HuggingFace's PEFT.<br>