efort
/
glm4
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

update:trans_batch_demo

This commit is contained in:
zhipuch 2024-12-31 16:07:08 +08:00
parent 81ba7e087c
commit 4dc2f76e68
1 changed files with 4 additions and 4 deletions

8
basic_demo/trans_batch_demo.py
View File

@ -46,7 +46,7 @@ def batch(
padding="max_length",
truncation=True,
max_length=max_input_tokens).to(model.device)
gen_kwargs = {
"max_new_tokens": max_new_tokens,
"num_beams": num_beams,
@ -75,14 +75,14 @@ if __name__ == "__main__":
]
batch_inputs = []
max_input_tokens = 1024
max_input_tokens = 128
for i, messages in enumerate(batch_message):
new_batch_input = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
new_batch_input = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)[12:]
max_input_tokens = max(max_input_tokens, len(new_batch_input))
batch_inputs.append(new_batch_input)
gen_kwargs = {
"max_input_tokens": max_input_tokens,
"max_new_tokens": 8192,
"max_new_tokens": 256,
"do_sample": True,
"top_p": 0.8,
"temperature": 0.8,