Upload QWQ_infer.py with huggingface_hub
Browse files- QWQ_infer.py +86 -0
QWQ_infer.py
ADDED
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
|
3 |
+
import json
|
4 |
+
import pprint
|
5 |
+
|
6 |
+
|
7 |
+
def read_json(file_path):
|
8 |
+
with open(file_path, 'r', encoding='utf-8') as file:
|
9 |
+
data = json.load(file)
|
10 |
+
return data
|
11 |
+
|
12 |
+
def write_json(file_path, data):
|
13 |
+
with open(file_path, 'w', encoding='utf-8') as file:
|
14 |
+
json.dump(data, file, ensure_ascii=False, indent=4)
|
15 |
+
|
16 |
+
data = read_json("DataSet/train_samples_all_tuning.json")
|
17 |
+
|
18 |
+
|
19 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
20 |
+
|
21 |
+
model_name = "Model/QwQ-32B-Preview"
|
22 |
+
model = AutoModelForCausalLM.from_pretrained(
|
23 |
+
model_name,
|
24 |
+
torch_dtype="auto",
|
25 |
+
device_map="auto"
|
26 |
+
)
|
27 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
28 |
+
|
29 |
+
|
30 |
+
|
31 |
+
def chat_QwQ(prompt):
|
32 |
+
|
33 |
+
messages = [
|
34 |
+
{"role": "system", "content": "You are a helpful and harmless assistant. You are Qwen developed by Alibaba. You should think step-by-step."},
|
35 |
+
{"role": "user", "content": prompt}
|
36 |
+
]
|
37 |
+
|
38 |
+
text = tokenizer.apply_chat_template(
|
39 |
+
messages,
|
40 |
+
tokenize=False,
|
41 |
+
add_generation_prompt=True
|
42 |
+
)
|
43 |
+
|
44 |
+
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
45 |
+
|
46 |
+
generated_ids = model.generate(
|
47 |
+
**model_inputs,
|
48 |
+
max_new_tokens=512
|
49 |
+
)
|
50 |
+
|
51 |
+
generated_ids = [
|
52 |
+
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
53 |
+
]
|
54 |
+
|
55 |
+
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
56 |
+
# print(response)
|
57 |
+
return response
|
58 |
+
|
59 |
+
|
60 |
+
|
61 |
+
|
62 |
+
# from transformers import MarianMTModel, MarianTokenizer
|
63 |
+
|
64 |
+
# model_name = "Model/opus-mt-en-zh"
|
65 |
+
# tokenizer = MarianTokenizer.from_pretrained(model_name)
|
66 |
+
# model = MarianMTModel.from_pretrained(model_name)
|
67 |
+
|
68 |
+
|
69 |
+
|
70 |
+
|
71 |
+
|
72 |
+
for i in data:
|
73 |
+
sent1 = i['conversations'][0]['value']
|
74 |
+
sent2 = i['conversations'][1]['value']
|
75 |
+
sentence = sent1 + sent2
|
76 |
+
prompt = "This is a question-answering datapoint based on image information. Determine whether the answer can be judged without relying on the image. If it can, this is considered bad data; if it requires the image, it is considered good data. Rate this datapoint on a scale from 1 (bad) to 5 (good). ###### " + sentence
|
77 |
+
answer = chat_QwQ(prompt)
|
78 |
+
|
79 |
+
|
80 |
+
# english_text = answer
|
81 |
+
# inputs = tokenizer.encode(english_text, return_tensors="pt", truncation=True)
|
82 |
+
# translated = model.generate(inputs, max_length=40, num_beams=4, early_stopping=True)
|
83 |
+
# chinese_translation = tokenizer.decode(translated[0], skip_special_tokens=True)
|
84 |
+
pprint.pprint(prompt)
|
85 |
+
pprint.pprint(answer)
|
86 |
+
break
|