acf15626vp
commited on
Commit
•
7eff671
1
Parent(s):
6291643
upload model
Browse files- README.md +58 -0
- config.json +26 -0
- generation_config.json +7 -0
- model.safetensors +3 -0
- pytorch_model.bin +3 -0
- special_tokens_map.json +6 -0
- tokenizer.json +0 -0
- tokenizer_config.json +10 -0
README.md
CHANGED
@@ -1,3 +1,61 @@
|
|
1 |
---
|
2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
+
language:
|
4 |
+
- ja
|
5 |
+
library_name: transformers
|
6 |
+
pipeline_tag: text-generation
|
7 |
+
tags:
|
8 |
+
- gpt_neox
|
9 |
+
- gpt-neox
|
10 |
+
- japanese
|
11 |
+
inference:
|
12 |
+
parameters:
|
13 |
+
max_new_tokens: 128
|
14 |
+
do_sample: false
|
15 |
+
repetition_penalty: 1.1
|
16 |
---
|
17 |
+
|
18 |
+
# stockmark/gpt-neox-japanese-1.4b
|
19 |
+
|
20 |
+
This repository provides a GPT-NeoX based model with 1.4B parameters pre-trained on Japanese corpus of about 20B tokens. This model is developed by [Stockmark Inc.](https://stockmark.co.jp/)
|
21 |
+
|
22 |
+
## How to use
|
23 |
+
|
24 |
+
```python
|
25 |
+
import torch
|
26 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
27 |
+
|
28 |
+
# Use torch.bfloat16 for A100 GPU and torch.flaot16 for the older generation GPUs
|
29 |
+
torch_dtype = torch.bfloat16 if torch.cuda.is_available() and hasattr(torch.cuda, "is_bf16_supported") and torch.cuda.is_bf16_supported() else torch.float16
|
30 |
+
|
31 |
+
model = AutoModelForCausalLM.from_pretrained("stockmark/gpt-neox-japanese-1.4b", device_map="auto", torch_dtype=torch_dtype)
|
32 |
+
tokenizer = AutoTokenizer.from_pretrained("stockmark/gpt-neox-japanese-1.4b")
|
33 |
+
|
34 |
+
inputs = tokenizer("自然言語処理とは", return_tensors="pt").to(model.device)
|
35 |
+
with torch.no_grad():
|
36 |
+
tokens = model.generate(
|
37 |
+
**inputs,
|
38 |
+
max_new_tokens=128,
|
39 |
+
repetition_penalty=1.1
|
40 |
+
)
|
41 |
+
|
42 |
+
output = tokenizer.decode(tokens[0], skip_special_tokens=True)
|
43 |
+
print(output)
|
44 |
+
```
|
45 |
+
|
46 |
+
## Training dataset
|
47 |
+
- Japanese Web Corpus (ja): 8.6B tokens (This dataset will not be released.)
|
48 |
+
- Wikipedia (ja): 0.88B tokens
|
49 |
+
- CC100 (ja): 10.5B tokens
|
50 |
+
|
51 |
+
## Training setting
|
52 |
+
- Trained using HuggingFace Trainer and DeepSpeed (ZeRO-2)
|
53 |
+
- 8 A100 GPUs (40GB) at ABCI
|
54 |
+
- Mixed Precision (BF16)
|
55 |
+
|
56 |
+
## License
|
57 |
+
[The MIT license](https://opensource.org/licenses/MIT)
|
58 |
+
|
59 |
+
## Developed by
|
60 |
+
Stockmark Inc.
|
61 |
+
https://stockmark.co.jp/
|
config.json
ADDED
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"GPTNeoXForCausalLM"
|
4 |
+
],
|
5 |
+
"bos_token_id": 0,
|
6 |
+
"classifier_dropout": 0.1,
|
7 |
+
"eos_token_id": 0,
|
8 |
+
"hidden_act": "gelu",
|
9 |
+
"hidden_size": 2048,
|
10 |
+
"initializer_range": 0.02,
|
11 |
+
"intermediate_size": 8192,
|
12 |
+
"layer_norm_eps": 1e-05,
|
13 |
+
"max_position_embeddings": 1024,
|
14 |
+
"model_type": "gpt_neox",
|
15 |
+
"num_attention_heads": 16,
|
16 |
+
"num_hidden_layers": 24,
|
17 |
+
"pad_token_id": 1,
|
18 |
+
"rotary_emb_base": 10000,
|
19 |
+
"rotary_pct": 0.25,
|
20 |
+
"tie_word_embeddings": false,
|
21 |
+
"torch_dtype": "bfloat16",
|
22 |
+
"transformers_version": "4.30.2",
|
23 |
+
"use_cache": true,
|
24 |
+
"use_parallel_residual": false,
|
25 |
+
"vocab_size": 50000
|
26 |
+
}
|
generation_config.json
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_from_model_config": true,
|
3 |
+
"bos_token_id": 0,
|
4 |
+
"eos_token_id": 0,
|
5 |
+
"pad_token_id": 1,
|
6 |
+
"transformers_version": "4.30.2"
|
7 |
+
}
|
model.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2ce45432f8538cee7b698f847dcb7f31cea2e27d0322bcf2fa7bf12796246baf
|
3 |
+
size 2852015168
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2039be3d49e59904559ab49df85fa8ed58a35967e535d19f174b296a8695b300
|
3 |
+
size 2852090557
|
special_tokens_map.json
ADDED
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": "<|endoftext|>",
|
3 |
+
"eos_token": "<|endoftext|>",
|
4 |
+
"pad_token": "<|padding|>",
|
5 |
+
"unk_token": "<|endoftext|>"
|
6 |
+
}
|
tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
tokenizer_config.json
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"add_prefix_space": false,
|
3 |
+
"bos_token": "<|endoftext|>",
|
4 |
+
"clean_up_tokenization_spaces": true,
|
5 |
+
"eos_token": "<|endoftext|>",
|
6 |
+
"model_max_length": 1000000000000000019884624838656,
|
7 |
+
"pad_token": "<|padding|>",
|
8 |
+
"tokenizer_class": "GPTNeoXTokenizer",
|
9 |
+
"unk_token": "<|endoftext|>"
|
10 |
+
}
|