Demo on Google Colab: https://colab.research.google.com/drive/1i5plJtq_6HIOuk_x7D-LkYDpcd3SADLf?usp=sharing
Similarly as Qwen-1.5-14B-Chat, you can always call this model from the AutoModel class.
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto
model = AutoModelForCausalLM.from_pretrained(
"ljsabc/Qwen-1.5-14B-Chat-Fujisaki",
torch_dtype="auto",
device_map="auto",
#load_in_4bit=True
)
tokenizer = AutoTokenizer.from_pretrained("ljsabc/Qwen-1.5-14B-Chat-Fujisaki")
prompt = "请撰写一条新的推文。"
messages = [
{"role": "system", "content": "你将扮演推特用户@ljsabc，你需要撰写你的原创推文或回复别人的推文。所有你的回复都应该使用简体中文书写。"},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)
generated_ids = model.generate(
model_inputs.input_ids,
max_new_tokens=512,
temperature=0.95,
top_p=0.99
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]