File size: 820 Bytes
887486f
 
 
 
8707caf
e3464c0
 
 
 
 
 
 
 
 
7fd60cb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e3464c0
887486f
 
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
28
29
30
31
32
33
34
35
36
37
38
39
---
license: apache-2.0
---

#### Quantization config
```

    "zero_point": true,
    "q_group_size": 128,
    "w_bit": 4,
    "version": "GEMM"

```

#### Script to AWQ quantization
```
from awq import AutoAWQForCausalLM
from transformers import AutoTokenizer

model_path = 'PATH_TO Poro-34B'
quant_path = 'Poro-34B-AWQ'
quant_config = { "zero_point": True, "q_group_size": 128, "w_bit": 4, "version": "GEMM" }

# Load model
model = AutoAWQForCausalLM.from_pretrained(model_path, safetensors=True)
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)

# Quantize
model.quantize(tokenizer, quant_config=quant_config)

# Save quantized model
model.save_quantized(quant_path)
tokenizer.save_pretrained(quant_path)
```



#### Work supported by https://datacrunch.io/
##### Quantized by: gradjitta