File size: 1,702 Bytes
6d93905
 
 
9b2de9d
 
 
274d466
9b2de9d
274d466
9b2de9d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
40
41
42
43
44
45
46
47
48
49
50
51
52
53
---
license: apache-2.0
---
language:
- zh
- en

library_name: transformers

tags:
- baichuan
---

This is an SFT model trained using https://github.com/hiyouga/LLaMA-Efficient-Tuning.

Thanks to the original author for their hard work.

All work is based on https://huggingface.co./baichuan-inc/baichuan-7B.

You can find the matching data set on the github of the fine-tuning framework.

We carried out 4 epoch of distributed training on the 8-card H100 machine, which took a short time. However, there is not much change in the loss.
In the future, we will update the data set to see how it will perform in a vertical field.

Of course, this is the inference code of the original author. You can use it directly.

Usage:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
from peft import PeftModel


tokenizer = AutoTokenizer.from_pretrained("baichuan-inc/baichuan-7B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("baichuan-inc/baichuan-7B", device_map="auto", trust_remote_code=True)
model = PeftModel.from_pretrained(model, "/data/baichuan-7b-sft") #change to your own path.
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

query = "晚上睡不着怎么办"

inputs = tokenizer(["<human>:{}\n<bot>:".format(query)], return_tensors="pt")
inputs = inputs.to("cuda")
generate_ids = model.generate(**inputs, max_new_tokens=256, streamer=streamer)
```

You could also alternatively launch a CLI demo by using the script in https://github.com/hiyouga/LLaMA-Efficient-Tuning
```bash
python src/cli_demo.py \
    --model_name_or_path baichuan-inc/baichuan-7B \
    --checkpoint_dir hiyouga/baichuan-7b-sft \

```