mix-llama-3-8B-inst-line

mix-llama-3-8B-inst-line is a merge of the following models using LazyMergekit:

• NousResearch/Meta-Llama-3-8B-Instruct

🧩 Configuration

dtype: bfloat16
merge_method: linear
slices:
 - sources:
      - layer_range: [0, 32] # Assuming the first half of the model is more general and can be reduced more
        model: NousResearch/Meta-Llama-3-8B-Instruct
        parameters:
          weight: 1.0 # Reduce the weight of the first half to make room for the second half
      - layer_range: [0, 32] # Assuming the second half of the model is more specialized and can be reduced less
        model: NousResearch/Meta-Llama-3-8B-Instruct
        parameters:
          weight: 1.0 # Maintain the weight of the second half

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "JoPmt/mix-llama-3-8B-inst-line"
messages = [{"role": "user", "content": "What is a large language model?"}]

tokenizer = AutoTokenizer.from_pretrained(model)
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])