Views
No views yet
1slices:
2 - sources:
3 - model: shanchen/llama3-8B-slerp-med-chinese
4 layer_range: [0,32]
5 - model: shenzhi-wang/Llama3-8B-Chinese-Chat
6 layer_range: [0,32]
7merge_method: slerp
8base_model: shenzhi-wang/Llama3-8B-Chinese-Chat
9parameters:
10 t:
11 - filter: self_attn
12 value: [0.3, 0.5, 0.5, 0.7, 1]
13 - filter: mlp
14 value: [1, 0.7, 0.5, 0.5, 0.3]
15 - value: 0.5
16dtype: bfloat161!pip install -qU transformers accelerate
2
3from transformers import AutoTokenizer, AutoModelForCausalLM
4
5model_id = "shanchen/llama3-8B-slerp-biomed-chat-chinese"
6
7tokenizer = AutoTokenizer.from_pretrained(model_id)
8model = AutoModelForCausalLM.from_pretrained(
9 model_id, torch_dtype="auto", device_map="auto"
10)
11
12messages = [
13 {"role": "user", "content": "Can you speak Japanese?"},
14]
15
16input_ids = tokenizer.apply_chat_template(
17 messages, add_generation_prompt=True, return_tensors="pt"
18).to(model.device)
19
20outputs = model.generate(
21 input_ids,
22 max_new_tokens=192 max#8192,
23 do_sample=True,
24 temperature=0.6,
25 top_p=0.9,
26)
27response = outputs[0][input_ids.shape[-1]:]
28print(tokenizer.decode(response, skip_special_tokens=True))
29