File size: 1,443 Bytes
931f35a
 
3ea080c
 
 
 
 
 
 
 
 
 
931f35a
3ea080c
 
 
 
 
 
 
 
 
 
 
 
 
e911ee1
 
 
 
 
3ea080c
e911ee1
3ea080c
 
 
e911ee1
 
3ea080c
 
 
 
 
 
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
---
license: bigscience-openrail-m
datasets:
- laion/Anh
library_name: transformers
pipeline_tag: text-generation
tags:
- pytorch
- casual-lm
- multilingual
- instruct
- bloomz
---

### Model description

This model is [`bloomz-7b1-mt`](https://huggingface.co/bigscience/bloomz-7b1-mt) model finetuned on instruct dataset `cross_lingual.jsonl` from [`laion/Anh`](https://huggingface.co/datasets/laion/Anh).


### How to use

anh-bloomz-7b1-mt-cross-lingual model can be loaded and used via the following code:

```python
import re
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "laion/anh-bloomz-7b1-mt-cross-lingual"
model = AutoModelForCausalLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

whitespace_tokens_map = {'\n': '<n>', '  ': '<w>'}
text = "User: Apakah kita akan bisa menyembuhkan penyakit kanker? Jawab dalam bahasa China.\n"
for k, v in whitespace_tokens_map.items():
    text = text.replace(k, v)
inputs = tokenizer(text, return_tensors="pt")
tokens = model.generate(**inputs, max_new_tokens=200, do_sample=True, top_k=40, top_p=0.9, temperature=0.2, 
                        repetition_penalty=1.2,num_return_sequences=1)
output = tokenizer.decode(tokens[0], skip_special_tokens=True)
for v in whitespace_tokens_map.values():
    output = re.sub(rf"{v}\s+(\S+)", rf"{v}\1", output)
for k, v in whitespace_tokens_map.items():
    output = output.replace(v, k)
```