Update README
Browse files
README.md
CHANGED
@@ -1,3 +1,57 @@
|
|
1 |
---
|
2 |
license: bigscience-bloom-rail-1.0
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: bigscience-bloom-rail-1.0
|
3 |
---
|
4 |
+
|
5 |
+
# Bloom CTranslate2's model
|
6 |
+
|
7 |
+
This is a collection of some of the [Bigscience Bloom](https://huggingface.co/bigscience/bloom) exported to
|
8 |
+
[CTranslate2](https://github.com/OpenNMT/CTranslate2) model format. This allows to load and usage these models
|
9 |
+
efficently on CPU or GPU.
|
10 |
+
|
11 |
+
## Models
|
12 |
+
|
13 |
+
The models have been converted to *float16* and can be load in with any other quantification method (e.g. *int 8*).
|
14 |
+
|
15 |
+
|
16 |
+
| Model name | Description |
|
17 |
+
| --- | --- |
|
18 |
+
| [bloom-560m](https://huggingface.co/bigscience/bloom-560m) | 560M parameter model pretrained on ROOTS|
|
19 |
+
| [bloom-3b](https://huggingface.co/bigscience/bloom-3b) | 3B parameter model pretrained on ROOTS
|
20 |
+
| [bloomz-7b1](https://huggingface.co/bigscience/bloomz-7b1) | 7.1B parameter model finetuned on xP3|
|
21 |
+
| [bloomz-7b1-mt](https://huggingface.co/bigscience/bloomz-7b1-mt) | 7.1B parameter model finetuned on xP3mt |
|
22 |
+
| [mt0-xxl-mt](https://huggingface.co/bigscience/mt0-xxl-mt) | 13B parameter model finetuned on xP3|
|
23 |
+
|
24 |
+
## Simple code to use them
|
25 |
+
|
26 |
+
Install dependencies:
|
27 |
+
|
28 |
+
```shell
|
29 |
+
pip install huggingface_hub ctranslate2 transformers torch
|
30 |
+
```
|
31 |
+
|
32 |
+
Usage:
|
33 |
+
|
34 |
+
```python
|
35 |
+
model_name = "bloomz-7b1"
|
36 |
+
prompt = "Hello, I am Joan and I am from Barcelona and"
|
37 |
+
|
38 |
+
repo_id = "jordimas/bloom-ctranslate2"
|
39 |
+
output_dir = "output/"
|
40 |
+
|
41 |
+
kwargs = {
|
42 |
+
"local_dir" : output_dir,
|
43 |
+
"local_dir_use_symlinks" : False,
|
44 |
+
}
|
45 |
+
huggingface_hub.snapshot_download(repo_id = repo_id, allow_patterns=f"*{model_name}*", **kwargs)
|
46 |
+
|
47 |
+
model = f"{output_dir}{model_name}"
|
48 |
+
print(f"model: {model}")
|
49 |
+
generator = ctranslate2.Generator(model, compute_type="int8")
|
50 |
+
tokenizer = transformers.AutoTokenizer.from_pretrained(model)
|
51 |
+
|
52 |
+
start_tokens = tokenizer.convert_ids_to_tokens(tokenizer.encode(prompt))
|
53 |
+
results = generator.generate_batch([start_tokens], max_length=90)
|
54 |
+
result = tokenizer.decode(results[0].sequences_ids[0])
|
55 |
+
print(f"Result: {result}")
|
56 |
+
```
|
57 |
+
|