-
Notifications
You must be signed in to change notification settings - Fork 0
/
1.py
47 lines (41 loc) · 1.95 KB
/
1.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
import json
import torch
from transformers import AutoTokenizer
from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel
device = "cuda"
tokenizer = AutoTokenizer.from_pretrained("/data/ruanjh/mamba-chat")
tokenizer.eos_token = "<|endoftext|>"
tokenizer.pad_token = tokenizer.eos_token
tokenizer.chat_template = "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'system' %}\n{{ '<|system|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n' + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}"
model = MambaLMHeadModel.from_pretrained(
"/data/ruanjh/mamba-chat",device=device, dtype=torch.float16
)
# user_message = """
# Translate this into German: {input}
# """
with open("/data/ruanjh/best_training_method/iwslt17/test.json") as f:
data = json.load(f)
en = [d["en"] for d in data]
result=[]
from tqdm import tqdm
with open('/data/ruanjh/best_training_method/iwslt17/mt_mamba_chat.de','w') as o:
for e in tqdm(en):
messages = []
user_message = f"Translate this into German: {e}"
messages.append(dict(role="user", content=user_message))
input_ids = tokenizer.apply_chat_template(
messages, return_tensors="pt", add_generation_prompt=True
).to("cuda")
out = model.generate(
input_ids=input_ids,
max_length=512,
temperature=0.9,
top_p=0.7,
eos_token_id=tokenizer.eos_token_id,
cg=True,
)
decoded = tokenizer.batch_decode(out, skip_special_tokens=True)[0].split("<|assistant|>\n")[-1]
messages.append(dict(role="assistant", content=decoded[0].split("<|assistant|>\n")[-1]))
decoded=decoded.replace('\n','\\n')
o.write(decoded+'\n')
# result.append()