File size: 752 Bytes
01dc3fb
 
8509d32
 
 
01dc3fb
8509d32
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
---
library_name: transformers
base_model:
- HuggingFaceM4/Idefics3-8B-Llama3
pipeline_tag: image-text-to-text
---
# Idefics3-8B-Llama3-bnb_nf4
BitsAndBytes NF4 quantization.

### Quantization
Quantization created with:
``` python
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

model_id = "HuggingFaceM4/Idefics3-8B-Llama3"

nf4_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
    llm_int8_enable_fp32_cpu_offload=True,
    llm_int8_skip_modules=["lm_head", "model.vision_model", "model.connector"],
    )

model_nf4 = AutoModelForVision2Seq.from_pretrained(model_id, quantization_config=nf4_config)
```