---
library_name: transformers
base_model:
  - nbeerbower/llama-3-wissenschaft-8B
datasets:
- jondurbin/truthy-dpo-v0.1
- kyujinpy/orca_math_dpo
license: other
license_name: llama3
---

![image/png](https://huggingface.co/nbeerbower/bophades-mistral-7B/resolve/main/bophades.png)

# llama-3-bophades-v3-8B

This model is based on Llama-3-8b, and is governed by [META LLAMA 3 COMMUNITY LICENSE AGREEMENT](LICENSE)

[nbeerbower/llama-3-wissenschaft-8B](https://huggingface.co/nbeerbower/llama-3-wissenschaft-8B) finetuned on [jondurbin/truthy-dpo-v0.1](https://huggingface.co/datasets/jondurbin/truthy-dpo-v0.1) and [kyujinpy/orca_math_dpo](https://huggingface.co/datasets/kyujinpy/orca_math_dpo). 

### Method

Finetuned using an A100 on Google Colab.

[Fine-tune a Mistral-7b model with Direct Preference Optimization](https://towardsdatascience.com/fine-tune-a-mistral-7b-model-with-direct-preference-optimization-708042745aac) - [Maxime Labonne](https://huggingface.co/mlabonne)

### Configuration

Dataset preperation and message formatting:

```python
def chatml_format(example):
    # Initialize formatted system message
    system = ""

    # Check if 'system' field exists and is not None
    if example.get('system'):
        system = "<|im_start|>system\n" + example['system'] + "<|im_end|>\n"

    # Format instruction
    instruction = ""
    if example.get('prompt'):
        instruction = example['prompt']
    if example.get('question'):
        instruction = example['question']
    prompt = "<|im_start|>user\n" + instruction + "<|im_end|>\n<|im_start|>assistant\n"

    # Format chosen answer
    chosen = example['chosen'] + "<|im_end|>\n"

    # Format rejected answer
    rejected = example['rejected'] + "<|im_end|>\n"

    return {
        "prompt": system + prompt,
        "chosen": chosen,
        "rejected": rejected,
    }

# Array of datasets to concat
ds = [
    "jondurbin/truthy-dpo-v0.1",
    "kyujinpy/orca_math_dpo"
]

# load_dataset and combine all
loaded_datasets = [load_dataset(dataset_name, split='train') for dataset_name in ds]
dataset = concatenate_datasets(loaded_datasets)

# Save columns
original_columns = dataset.column_names

# Tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_name)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "left"

# Format dataset
dataset = dataset.map(
    chatml_format,
    remove_columns=original_columns
)
```

LoRA, model, and training settings:

```python
# LoRA configuration
peft_config = LoraConfig(
    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']
)

# Model to fine-tune
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    load_in_4bit=True
)
model.config.use_cache = False

# Reference model
ref_model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    load_in_4bit=True
)

# Training arguments
training_args = TrainingArguments(
    per_device_train_batch_size=2,
    gradient_accumulation_steps=4,
    gradient_checkpointing=True,
    learning_rate=5e-5,
    lr_scheduler_type="cosine",
    max_steps=1000,
    save_strategy="no",
    logging_steps=1,
    output_dir=new_model,
    optim="paged_adamw_32bit",
    warmup_steps=100,
    bf16=True,
    report_to="wandb",
)

# Create DPO trainer
dpo_trainer = DPOTrainer(
    model,
    ref_model,
    args=training_args,
    train_dataset=dataset,
    tokenizer=tokenizer,
    peft_config=peft_config,
    beta=0.1,
    max_prompt_length=2048,
    max_length=4096,
    force_use_ref_model=True
)

# Fine-tune model with DPO
dpo_trainer.train()
```