yuanzhoulvpi
commited on
Commit
•
5573a11
1
Parent(s):
c8591a1
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,62 @@
|
|
1 |
---
|
2 |
-
license: bigscience-
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
license: bigscience-bloom-rail-1.0
|
3 |
+
language:
|
4 |
+
- zh
|
5 |
---
|
6 |
+
|
7 |
+
|
8 |
+
# 体验
|
9 |
+
🚀 点击链接,即可体验🔗 **[http://101.68.79.42:7861/](http://101.68.79.42:7861/)**
|
10 |
+
|
11 |
+
|
12 |
+
## 介绍
|
13 |
+
1. ✅ 对`bloom-7b`模型做了sft,本次版本为V2版本,相较于V1版本,效果更好!!!
|
14 |
+
2. 🚀 训练代码和推理代码全部分享,可以查看链接[https://github.com/yuanzhoulvpi2017/zero_nlp/tree/main/chinese_bloom](https://github.com/yuanzhoulvpi2017/zero_nlp/tree/main/chinese_bloom)
|
15 |
+
|
16 |
+
|
17 |
+
## 如何使用
|
18 |
+
|
19 |
+
```python
|
20 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
21 |
+
|
22 |
+
|
23 |
+
checkpoint = "yuanzhoulvpi/chinese_bloom_7b_chat_v2"#"bigscience/bloomz-3b" #"bigscience/bloom-7b1"# "output_dir/checkpoint-8260"#
|
24 |
+
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
|
25 |
+
model = AutoModelForCausalLM.from_pretrained(checkpoint).half().cuda()
|
26 |
+
|
27 |
+
PROMPT_DICT = {
|
28 |
+
"prompt_input": (
|
29 |
+
"Below is an instruction that describes a task, paired with an input that provides further context. "
|
30 |
+
"Write a response that appropriately completes the request.\n\n"
|
31 |
+
"### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:"
|
32 |
+
),
|
33 |
+
"prompt_no_input": (
|
34 |
+
"Below is an instruction that describes a task. "
|
35 |
+
"Write a response that appropriately completes the request.\n\n"
|
36 |
+
"### Instruction:\n{instruction}\n\n### Response:"
|
37 |
+
),
|
38 |
+
}
|
39 |
+
|
40 |
+
from typing import Optional
|
41 |
+
def generate_input(instruction:Optional[str]= None, input_str:Optional[str] = None) -> str:
|
42 |
+
if input_str is None:
|
43 |
+
return PROMPT_DICT['prompt_no_input'].format_map({'instruction':instruction})
|
44 |
+
else:
|
45 |
+
return PROMPT_DICT['prompt_input'].format_map({'instruction':instruction, 'input':input_str})
|
46 |
+
|
47 |
+
|
48 |
+
for i in range(5):
|
49 |
+
print("*"*80)
|
50 |
+
|
51 |
+
inputs = tokenizer.encode(generate_input(instruction="你是谁"), return_tensors="pt")
|
52 |
+
outputs = model.generate(inputs,num_beams=3,
|
53 |
+
max_new_tokens=512,
|
54 |
+
do_sample=False,
|
55 |
+
top_k=10,
|
56 |
+
penalty_alpha=0.6,
|
57 |
+
temperature=0.8,
|
58 |
+
repetition_penalty=1.2)
|
59 |
+
print(tokenizer.decode(outputs[0]))
|
60 |
+
```
|
61 |
+
|
62 |
+
|