Holy-fox commited on
Commit
11e00de
1 Parent(s): d5e7935

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +61 -3
README.md CHANGED
@@ -1,14 +1,71 @@
1
  ---
2
  base_model:
 
3
  - elyza/Llama-3-ELYZA-JP-8B
4
  - nvidia/Llama3-ChatQA-1.5-8B
5
  library_name: transformers
6
  tags:
7
  - mergekit
8
  - merge
9
-
 
10
  ---
11
- # merge
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
12
 
13
  This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
14
 
@@ -20,6 +77,7 @@ This model was merged using the [TIES](https://arxiv.org/abs/2306.01708) merge m
20
  ### Models Merged
21
 
22
  The following models were included in the merge:
 
23
  * [elyza/Llama-3-ELYZA-JP-8B](https://huggingface.co/elyza/Llama-3-ELYZA-JP-8B)
24
  * [nvidia/Llama3-ChatQA-1.5-8B](https://huggingface.co/nvidia/Llama3-ChatQA-1.5-8B)
25
 
@@ -43,4 +101,4 @@ base_model: ./llama3.1-8B
43
  parameters:
44
  normalize: false
45
  dtype: bfloat16
46
- ```
 
1
  ---
2
  base_model:
3
+ - meta-llama/Meta-Llama-3.1-8B-Instruct
4
  - elyza/Llama-3-ELYZA-JP-8B
5
  - nvidia/Llama3-ChatQA-1.5-8B
6
  library_name: transformers
7
  tags:
8
  - mergekit
9
  - merge
10
+ language:
11
+ - ja
12
  ---
13
+
14
+ ## 概要 
15
+
16
+ このモデルはllama3.1-8B-instructをもとに日本語性能を高めることを目的にMergekit&ファインチューニングを用いて作成されました。
17
+
18
+ meta,ELYZA,nvidiaの皆様に感謝します。
19
+
20
+ ## how to use
21
+
22
+
23
+ ```python
24
+ import torch
25
+ from transformers import AutoModelForCausalLM, AutoTokenizer
26
+
27
+ DEFAULT_SYSTEM_PROMPT = "あなたは誠実で優秀な日本人のアシスタントです。特に指示が無い場合は、常に日本語で回答してください。"
28
+ text = "Youtuberとして成功するために大切な5つのことを小学生にでもわかるように教えてください。"
29
+
30
+ model_name = "DataPilot/Llama3.1-ArrowSE-v0.4"
31
+
32
+ tokenizer = AutoTokenizer.from_pretrained(model_name)
33
+ model = AutoModelForCausalLM.from_pretrained(
34
+ model_name,
35
+ torch_dtype="auto",
36
+ device_map="auto",
37
+ )
38
+ model.eval()
39
+
40
+ messages = [
41
+ {"role": "system", "content": DEFAULT_SYSTEM_PROMPT},
42
+ {"role": "user", "content": text},
43
+ ]
44
+ prompt = tokenizer.apply_chat_template(
45
+ messages,
46
+ tokenize=False,
47
+ add_generation_prompt=True
48
+ )
49
+ token_ids = tokenizer.encode(
50
+ prompt, add_special_tokens=False, return_tensors="pt"
51
+ )
52
+
53
+ with torch.no_grad():
54
+ output_ids = model.generate(
55
+ token_ids.to(model.device),
56
+ max_new_tokens=1200,
57
+ do_sample=True,
58
+ temperature=0.6,
59
+ top_p=0.9,
60
+ )
61
+ output = tokenizer.decode(
62
+ output_ids.tolist()[0][token_ids.size(1):], skip_special_tokens=True
63
+ )
64
+ print(output)
65
+ ```
66
+
67
+
68
+ ## merge
69
 
70
  This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
71
 
 
77
  ### Models Merged
78
 
79
  The following models were included in the merge:
80
+ * [meta-llama/Meta-Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct)
81
  * [elyza/Llama-3-ELYZA-JP-8B](https://huggingface.co/elyza/Llama-3-ELYZA-JP-8B)
82
  * [nvidia/Llama3-ChatQA-1.5-8B](https://huggingface.co/nvidia/Llama3-ChatQA-1.5-8B)
83
 
 
101
  parameters:
102
  normalize: false
103
  dtype: bfloat16
104
+ ```