Edit model card

NeuralLLaMa-3-8b-DT-v0.1

image/png

NeuralLLaMa-3-8b-DT-v0.1 is a merge of the following models using LazyMergekit:

🧩 Configuration

models:
  - model: NousResearch/Meta-Llama-3-8B
    # No parameters necessary for base model
  - model: mlabonne/ChimeraLlama-3-8B-v2
    parameters:
      density: 0.33
      weight: 0.2
  - model: nbeerbower/llama-3-stella-8B
    parameters:
      density: 0.44
      weight: 0.4
  - model: uygarkurt/llama-3-merged-linear
    parameters:
      density: 0.55
      weight: 0.4
merge_method: dare_ties
base_model: NousResearch/Meta-Llama-3-8B
parameters:
  int8_mask: true
dtype: float16

πŸ—¨οΈ Chats

image/png

image/png

πŸ’» Usage

!pip install -qU transformers accelerate bitsandbytes

from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer, BitsAndBytesConfig
import torch

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)

MODEL_NAME = 'Kukedlc/NeuralLLaMa-3-8b-DT-v0.1'
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, device_map='cuda:0', quantization_config=bnb_config)

prompt_system = "You are an advanced language model that speaks Spanish fluently, clearly, and precisely.\
You are called Roberto the Robot and you are an aspiring post-modern artist."
prompt = "Create a piece of art that represents how you see yourself, Roberto, as an advanced LLm, with ASCII art, mixing diagrams, engineering and let yourself go."

chat = [
    {"role": "system", "content": f"{prompt_system}"},
    {"role": "user", "content": f"{prompt}"},
]

chat = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(chat, return_tensors="pt").to('cuda')
streamer = TextStreamer(tokenizer)
stop_token = "<|eot_id|>"
stop = tokenizer.encode(stop_token)[0]

_ = model.generate(**inputs, streamer=streamer, max_new_tokens=1024, do_sample=True, temperature=0.7, repetition_penalty=1.2, top_p=0.9, eos_token_id=stop)

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 21.12
IFEval (0-Shot) 43.71
BBH (3-Shot) 28.01
MATH Lvl 5 (4-Shot) 7.25
GPQA (0-shot) 7.05
MuSR (0-shot) 9.69
MMLU-PRO (5-shot) 31.02
Downloads last month
6,230
Safetensors
Model size
8.03B params
Tensor type
FP16
Β·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for Kukedlc/NeuralLLaMa-3-8b-DT-v0.1

Spaces using Kukedlc/NeuralLLaMa-3-8b-DT-v0.1 8

Evaluation results