umiyuki commited on
Commit
ecab5cf
1 Parent(s): fcee75f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +64 -2
README.md CHANGED
@@ -8,9 +8,67 @@ library_name: transformers
8
  tags:
9
  - mergekit
10
  - merge
11
-
 
 
12
  ---
13
- # final_model
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14
 
15
  This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
16
 
@@ -175,3 +233,7 @@ slices:
175
  parameters:
176
  weight: 0.342193342214983
177
  ```
 
 
 
 
 
8
  tags:
9
  - mergekit
10
  - merge
11
+ license: llama3
12
+ language:
13
+ - ja
14
  ---
15
+ # Llama-3-Umievo-itr014-Shizuko-8b
16
+
17
+ このモデルは日本語に対応しているLlama-3ベースの4つのモデルを進化的アルゴリズムで進化的マージしたものです。Meta-Llama-3-8B-Instruct、Llama-3-youko-8b-instruct-chatvector、suzume-llama-3-8B-multilingual、sa-v1-llama3-8bの4つのモデルを使用させていただきました。
18
+ マージに使用させていただいたモデル制作者のMeta、aixsatoshiさん、LightBlue、Shisa-AIのみなさまに感謝します。
19
+
20
+ This model is an evolutionary merge of four Llama-3-based models for Japanese using an evolutionary algorithm: Meta-Llama-3-8B-Instruct, Llama-3-youko-8b-instruct-chatvector, suzume- llama-3-8B-multilingual, and sa-v1-llama3-8b.
21
+ We would like to thank the model creators Meta, aixsatoshi, LightBlue, and Shisa-AI for allowing us to use their models for the merge.
22
+
23
+ ElyzaTasks100ベンチマークで平均点が3.85でした。(Llama3-70Bによる自動評価を3回行った平均点)
24
+
25
+ The average score was 3.85 on the ElyzaTasks100 benchmark. (Average score after 3 automatic evaluations by Llama3-70B)
26
+
27
+ ![image/png](https://cdn-uploads.huggingface.co/production/uploads/630420b4eedc089484c853e8/x4BbxfaW_wXPjDfv1Z4lJ.png)
28
+
29
+
30
+ ```python
31
+ from transformers import AutoTokenizer, AutoModelForCausalLM
32
+ import torch
33
+
34
+ model_id = "umiyuki/Llama-3-Umievo-itr014-Shizuko-8b"
35
+
36
+ tokenizer = AutoTokenizer.from_pretrained(model_id)
37
+ model = AutoModelForCausalLM.from_pretrained(
38
+ model_id,
39
+ torch_dtype=torch.bfloat16,
40
+ device_map="auto",
41
+ )
42
+
43
+ messages = [
44
+ {"role": "system", "content": "You must answer all responses in Japanese.あなたは役に立つ誠実な日本人のアシスタントです。あなたは全ての回答に日本語で答えなければならない。"},
45
+ {"role": "user", "content": "二人の少女が終末世界を旅する物語を書いてください。"},
46
+ ]
47
+
48
+ input_ids = tokenizer.apply_chat_template(
49
+ messages,
50
+ add_generation_prompt=True,
51
+ return_tensors="pt"
52
+ ).to(model.device)
53
+
54
+ terminators = [
55
+ tokenizer.eos_token_id,
56
+ tokenizer.convert_tokens_to_ids("<|eot_id|>")
57
+ ]
58
+
59
+ outputs = model.generate(
60
+ input_ids,
61
+ max_new_tokens=256,
62
+ eos_token_id=terminators,
63
+ do_sample=True,
64
+ temperature=0.6,
65
+ top_p=0.9,
66
+ )
67
+ response = outputs[0][input_ids.shape[-1]:]
68
+ print(tokenizer.decode(response, skip_special_tokens=True))
69
+ ```
70
+
71
+
72
 
73
  This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
74
 
 
233
  parameters:
234
  weight: 0.342193342214983
235
  ```
236
+
237
+ Built with Meta Llama 3
238
+
239
+ Meta Llama 3 is licensed under the Meta Llama 3 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved