shunxing1234
commited on
Commit
•
d53e5a3
1
Parent(s):
811643a
Upload test_chat.py
Browse files- test_chat.py +21 -0
test_chat.py
ADDED
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import AutoTokenizer, AquilaForCausalLM
|
2 |
+
import torch
|
3 |
+
from cyg_conversation import default_conversation, covert_prompt_to_input_ids_with_history
|
4 |
+
|
5 |
+
tokenizer = AutoTokenizer.from_pretrained("BAAI/AquilaChat-7B")
|
6 |
+
model = AquilaForCausalLM.from_pretrained("BAAI/AquilaChat-7B")
|
7 |
+
model.eval()
|
8 |
+
model.to("cuda:4")
|
9 |
+
vocab = tokenizer.vocab
|
10 |
+
print(len(vocab))
|
11 |
+
|
12 |
+
text = "请给出10个要到北京旅游的理由。"
|
13 |
+
|
14 |
+
tokens = covert_prompt_to_input_ids_with_history(text, history=[], tokenizer=tokenizer, max_token=512)
|
15 |
+
|
16 |
+
tokens = torch.tensor(tokens)[None,].to("cuda:4")
|
17 |
+
|
18 |
+
out = model.generate(tokens, do_sample=True, max_length=512, eos_token_id=100007)[0]
|
19 |
+
out = tokenizer.decode(out.cpu().numpy().tolist())
|
20 |
+
|
21 |
+
print(out)
|