metadata
language:
- zh
- en
ChatYuan-7B-merge
bilibili github kaggle huggingface
Based on LLAMA's latest Chinese-English dialogue language large model
You can see more detail in this repo
How to use
from transformers import LlamaForCausalLM, AutoTokenizer
import torch
ckpt = "tiansz/ChatYuan-7B-merge"
device = torch.device('cuda')
model = LlamaForCausalLM.from_pretrained(ckpt)
tokenizer = AutoTokenizer.from_pretrained(ckpt)
def answer(prompt):
prompt = f"用户：{prompt}\n小元："
input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to(device)
generate_ids = model.generate(input_ids, max_new_tokens=1024, do_sample = True, temperature = 0.7)
output = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
response = output[len(prompt):]
return response
result = answer("你好")
print(result)
int8:
from transformers import LlamaForCausalLM, AutoTokenizer
import torch
ckpt = "tiansz/ChatYuan-7B-merge"
device = torch.device('cuda')
max_memory = f'{int(torch.cuda.mem_get_info()[0]/1024**3)-1}GB'
n_gpus = torch.cuda.device_count()
max_memory = {i: max_memory for i in range(n_gpus)}
model = LlamaForCausalLM.from_pretrained(ckpt, device_map='auto', load_in_8bit=True, max_memory=max_memory)
tokenizer = AutoTokenizer.from_pretrained(ckpt)
def answer(prompt):
prompt = f"用户：{prompt}\n小元："
input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to(device)
generate_ids = model.generate(input_ids, max_new_tokens=1024, do_sample = True, temperature = 0.7)
output = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
response = output[len(prompt):]
return response
result = answer("你好")
print(result)