Update README.md
Browse files
README.md
CHANGED
@@ -13,3 +13,41 @@ language:
|
|
13 |
# huihui-ai/dots.llm1.inst
|
14 |
|
15 |
This version only allows local loading of dots.llm1.inst using transformers, with only the local import issue modified and no other changes.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
# huihui-ai/dots.llm1.inst
|
14 |
|
15 |
This version only allows local loading of dots.llm1.inst using transformers, with only the local import issue modified and no other changes.
|
16 |
+
|
17 |
+
## Usage
|
18 |
+
Copy the four files to the model directory, and then you can use the following program.
|
19 |
+
|
20 |
+
```
|
21 |
+
import sys
|
22 |
+
import os
|
23 |
+
import torch
|
24 |
+
from transformers import AutoTokenizer, AutoConfig, AutoModel, BitsAndBytesConfig
|
25 |
+
|
26 |
+
AutoConfig.register("dots1", Dots1Config)
|
27 |
+
AutoModel.register(Dots1Config, Dots1ForCausalLM)
|
28 |
+
|
29 |
+
config = AutoConfig.from_pretrained(MODEL_ID)
|
30 |
+
print(config)
|
31 |
+
|
32 |
+
quant_config_4 = BitsAndBytesConfig(
|
33 |
+
load_in_4bit=True,
|
34 |
+
bnb_4bit_compute_dtype=torch.bfloat16,
|
35 |
+
bnb_4bit_use_double_quant=True,
|
36 |
+
llm_int8_enable_fp32_cpu_offload=True,
|
37 |
+
)
|
38 |
+
|
39 |
+
model = Dots1ForCausalLM.from_pretrained(
|
40 |
+
MODEL_ID,
|
41 |
+
device_map="auto",
|
42 |
+
trust_remote_code=True,
|
43 |
+
quantization_config=quant_config_4,
|
44 |
+
torch_dtype=torch.bfloat16,
|
45 |
+
low_cpu_mem_usage=True,
|
46 |
+
)
|
47 |
+
|
48 |
+
print(model)
|
49 |
+
print(model.config)
|
50 |
+
|
51 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
|
52 |
+
|
53 |
+
```
|