ChimeraLlama-3-8B

ChimeraLlama-3-8B outperforms Llama 3 8B Instruct on Nous' benchmark suite.

ChimeraLlama-3-8B is a merge of the following models using LazyMergekit:

• NousResearch/Meta-Llama-3-8B-Instruct
• mlabonne/OrpoLlama-3-8B
• Locutusque/Llama-3-Orca-1.0-8B
• abacusai/Llama-3-Smaug-8B

🏆 Evaluation

Nous

Evaluation performed using LLM AutoEval, see the entire leaderboard here.

Model	Average	AGIEval	GPT4All	TruthfulQA	Bigbench
mlabonne/ChimeraLlama-3-8B 📄	51.58	39.12	71.81	52.4	42.98
meta-llama/Meta-Llama-3-8B-Instruct 📄	51.34	41.22	69.86	51.65	42.64
mlabonne/OrpoLlama-3-8B 📄	48.63	34.17	70.59	52.39	37.36
meta-llama/Meta-Llama-3-8B 📄	45.42	31.1	69.95	43.91	36.7

🧩 Configuration

models:
  - model: NousResearch/Meta-Llama-3-8B
    # No parameters necessary for base model
  - model: NousResearch/Meta-Llama-3-8B-Instruct
    parameters:
      density: 0.58
      weight: 0.4
  - model: mlabonne/OrpoLlama-3-8B
    parameters:
      density: 0.52
      weight: 0.2
  - model: Locutusque/Llama-3-Orca-1.0-8B
    parameters:
      density: 0.52
      weight: 0.2
  - model: abacusai/Llama-3-Smaug-8B
    parameters:
      density: 0.52
      weight: 0.2
merge_method: dare_ties
base_model: NousResearch/Meta-Llama-3-8B
parameters:
  int8_mask: true
dtype: float16

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "mlabonne/ChimeraLlama-3-8B"
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"])