mlabonne/NeuralMarcoro14-7B

NeuralMarcoro14-7B

This is a DPO fine-tuned version of mlabonne/Marcoro14-7B-slerp using the chatml_dpo_pairs preference dataset. It improves the performance of the model on Nous benchmark suite and the Open LLM Benchmark.

It is currently the best-performing 7B LLM on the Open LLM Leaderboard (08/01/24).

You can try it out in this Space (GGUF Q4KM).

⚡ Quantized models

• —GGUF: https://huggingface.co/mlabonne/NeuralMarcoro14-7B-GGUF

🏆 Evaluation

Open LLM Leaderboard

Nous

🧩 Training hyperparameters

LoRA:

• —r=16
• —lora_alpha=16
• —lora_dropout=0.05
• —bias="none"
• —tasktype="CAUSALLM"
• —targetmodules=['kproj', 'gateproj', 'vproj', 'upproj', 'qproj', 'oproj', 'downproj']

Training arguments:

• —perdevicetrainbatchsize=4
• —gradientaccumulationsteps=4
• —gradient_checkpointing=True
• —learning_rate=5e-5
• —lrschedulertype="cosine"
• —max_steps=200
• —optim="pagedadamw32bit"
• —warmup_steps=100

DPOTrainer:

• —beta=0.1
• —maxpromptlength=1024
• —max_length=1536

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "mlabonne/NeuralMarcoro14-7B"
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"])