metadata

base_model: teknium/OpenHermes-2.5-Mistral-7B
tags:
  - mistral
  - instruct
  - finetune
  - chatml
  - gpt4
  - synthetic data
  - distillation
  - dpo
  - rlhf
license: apache-2.0
language:
  - en
datasets:
  - mlabonne/chatml_dpo_pairs

NeuralHermes 2.5 - Mistral 7B - GGUF

NeuralHermes is an OpenHermes-2.5-Mistral-7B model that has been further fine-tuned with Direct Preference Optimization (DPO) using the mlabonne/chatml_dpo_pairs dataset.

It is directly inspired by the RLHF process described by neural-chat-7b-v3-1's authors to improve performance. I used the same dataset and reformatted it to apply the ChatML template. I haven't performed a comprehensive evaluation of the model, but it works great, nothing broken apparently! :)

The code to train this model is available on Google Colab and GitHub. It required an A100 GPU for about an hour.

Link to the original model: mlabonne/NeuralHermes-2.5-Mistral-7B.

Article and code to quantize your own LLMs: Quantize Llama models with GGUF and llama.cpp

Usage

You can run this model using LM Studio or any other frontend.

You can also run this model using the following code:

import transformers
from transformers import AutoTokenizer

# Format prompt
message = [
    {"role": "system", "content": "You are a helpful assistant chatbot."},
    {"role": "user", "content": "What is a Large Language Model?"}
]
tokenizer = AutoTokenizer.from_pretrained(new_model)
prompt = tokenizer.apply_chat_template(message, add_generation_prompt=True, tokenize=False)

# Create pipeline
pipeline = transformers.pipeline(
    "text-generation",
    model=new_model,
    tokenizer=tokenizer
)

# Generate text
sequences = pipeline(
    prompt,
    do_sample=True,
    temperature=0.7,
    top_p=0.9,
    num_return_sequences=1,
    max_length=200,
)
print(sequences[0]['generated_text'])

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,