Zoyd/mlabonne_Meta-Llama-3-12B-Instruct-2_5bpw_exl2

Exllamav2 quant (exl2 / 2.5 bpw) made with ExLlamaV2 v0.0.21

Other EXL2 quants:

Quant	Model Size	lm_head
2.2	4176 MB	6
2.5	4519 MB	6
3.0	5143 MB	6
3.5	5766 MB	6
3.75	6077 MB	6
4.0	6391 MB	6
4.25	6703 MB	6
5.0	7637 MB	6
6.0	8992 MB	8
6.5	9616 MB	8
8.0	11473 MB	8

Meta-Llama-3-12B-Instruct

Meta-Llama-3-12B-Instruct is a merge of the following models using LazyMergekit:

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

🏆 Evaluation

Model	AGIEval	GPT4All	TruthfulQA	Bigbench	Average
Meta-Llama-3-12B-Instruct	41.7	67.71	52.75	40.58	50.69
Meta-Llama-3-12B	29.46	68.01	41.02	35.57	43.52

🧩 Configuration

slices:
  - sources:
    - model: NousResearch/Meta-Llama-3-8B-Instruct
      layer_range: [0,9]
  - sources:
    - model: NousResearch/Meta-Llama-3-8B-Instruct
      layer_range: [5,14]
  - sources:
    - model: NousResearch/Meta-Llama-3-8B-Instruct
      layer_range: [10,19]
  - sources:
    - model: NousResearch/Meta-Llama-3-8B-Instruct
      layer_range: [15,24]
  - sources:
    - model: NousResearch/Meta-Llama-3-8B-Instruct
      layer_range: [20,32]
merge_method: passthrough
dtype: bfloat16

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "mlabonne/Meta-Llama-3-12B-Instruct"
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"])