NeuralMarcoro14-7B/README.md

license, tags, datasets, base_model, model-index

license	tags	datasets	base_model	model-index
cc-by-nc-4.0	mlabonne/Marcoro14-7B-slerp dpo rlhf merge mergekit lazymergekit	mlabonne/chatml_dpo_pairs	mlabonne/Marcoro14-7B-slerp	name results NeuralMarcoro14-7B task dataset metrics source type name text-generation Text Generation name type config split args AI2 Reasoning Challenge (25-Shot) ai2_arc ARC-Challenge test num_few_shot 25 type value name acc_norm 71.42 normalized accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralMarcoro14-7B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type split args HellaSwag (10-Shot) hellaswag validation num_few_shot 10 type value name acc_norm 87.59 normalized accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralMarcoro14-7B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args MMLU (5-Shot) cais/mmlu all test num_few_shot 5 type value name acc 64.84 accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralMarcoro14-7B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args TruthfulQA (0-shot) truthful_qa multiple_choice validation num_few_shot 0 type value mc2 65.64 url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralMarcoro14-7B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args Winogrande (5-shot) winogrande winogrande_xl validation num_few_shot 5 type value name acc 81.22 accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralMarcoro14-7B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args GSM8k (5-shot) gsm8k main test num_few_shot 5 type value name acc 70.74 accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralMarcoro14-7B Open LLM Leaderboard

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 Q4_K_M).

⚡ Quantized models

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

🏆 Evaluation

Open LLM Leaderboard

Nous

Model	AGIEval	GPT4ALL	TruthfulQA	Bigbench	Average
NeuralMarcoro14-7B	44.59	76.17	65.94	46.9	58.4
Marcoro14-7B-slerp	44.66	76.24	64.15	45.64	57.67
Change	-0.07	-0.07	+1.79	+1.26	+0.73

🧩 Training hyperparameters

LoRA:

• r=16
• lora_alpha=16
• lora_dropout=0.05
• bias="none"
• task_type="CAUSAL_LM"
• target_modules=['k_proj', 'gate_proj', 'v_proj', 'up_proj', 'q_proj', 'o_proj', 'down_proj']

Training arguments:

• per_device_train_batch_size=4
• gradient_accumulation_steps=4
• gradient_checkpointing=True
• learning_rate=5e-5
• lr_scheduler_type="cosine"
• max_steps=200
• optim="paged_adamw_32bit"
• warmup_steps=100

DPOTrainer:

• beta=0.1
• max_prompt_length=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"])