Commit
·
a7f2cd0
1
Parent(s):
91b112b
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,86 @@
|
|
1 |
---
|
2 |
license: mit
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
+
language: ja
|
4 |
+
|
5 |
---
|
6 |
+
|
7 |
+
```python
|
8 |
+
|
9 |
+
import torch
|
10 |
+
from peft import PeftModel, PeftConfig
|
11 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
12 |
+
|
13 |
+
model_name = "cyberagent/open-calm-large"
|
14 |
+
lora_weights = "Mizuiro-sakura/open-calm-large-finetuned-databricks-dolly"
|
15 |
+
|
16 |
+
# モデルの準備
|
17 |
+
model = AutoModelForCausalLM.from_pretrained(
|
18 |
+
model_name
|
19 |
+
)
|
20 |
+
|
21 |
+
# トークンナイザーの準備
|
22 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
23 |
+
|
24 |
+
# LoRAモデルの準備
|
25 |
+
model = PeftModel.from_pretrained(
|
26 |
+
model,
|
27 |
+
lora_weights,
|
28 |
+
adapter_name=lora_weights
|
29 |
+
)
|
30 |
+
|
31 |
+
# 評価モード
|
32 |
+
model.eval()
|
33 |
+
|
34 |
+
# プロンプトテンプレートの準備
|
35 |
+
def generate_prompt(data_point):
|
36 |
+
if data_point["input"]:
|
37 |
+
return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
|
38 |
+
|
39 |
+
### Instruction:
|
40 |
+
{data_point["instruction"]}
|
41 |
+
|
42 |
+
### Input:
|
43 |
+
{data_point["input"]}
|
44 |
+
|
45 |
+
### Response:"""
|
46 |
+
else:
|
47 |
+
return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
|
48 |
+
|
49 |
+
### Instruction:
|
50 |
+
{data_point["instruction"]}
|
51 |
+
|
52 |
+
### Response:"""
|
53 |
+
|
54 |
+
# テキスト生成関数の定義
|
55 |
+
def generate(instruction,input=None,maxTokens=256):
|
56 |
+
# 推論
|
57 |
+
prompt = generate_prompt({'instruction':instruction,'input':input})
|
58 |
+
input_ids = tokenizer(prompt, return_tensors="pt", truncation=True).input_ids.to('mps')
|
59 |
+
outputs = model.generate(
|
60 |
+
input_ids=input_ids,
|
61 |
+
max_new_tokens=maxTokens,
|
62 |
+
do_sample=True,
|
63 |
+
temperature=0.7,
|
64 |
+
top_p=0.75,
|
65 |
+
top_k=40,
|
66 |
+
no_repeat_ngram_size=2,
|
67 |
+
)
|
68 |
+
outputs = outputs[0].tolist()
|
69 |
+
|
70 |
+
# EOSトークンにヒットしたらデコード完了
|
71 |
+
if tokenizer.eos_token_id in outputs:
|
72 |
+
eos_index = outputs.index(tokenizer.eos_token_id)
|
73 |
+
else:
|
74 |
+
eos_index = len(outputs)
|
75 |
+
decoded = tokenizer.decode(outputs[:eos_index])
|
76 |
+
|
77 |
+
# レスポンス内容のみ抽出
|
78 |
+
sentinel = "### Response:"
|
79 |
+
sentinelLoc = decoded.find(sentinel)
|
80 |
+
if sentinelLoc >= 0:
|
81 |
+
print(decoded[sentinelLoc+len(sentinel):])
|
82 |
+
else:
|
83 |
+
print('Warning: Expected prompt template to be emitted. Ignoring output.')
|
84 |
+
|
85 |
+
generate("自然言語処理とは?")
|
86 |
+
```
|