flammen21 - Mistral-7B Open-source Large Language Model, Proficient in General Tasks Such as Role-playing and Creative Writing

Flammen21 Mistral 7B

Developed by flammenai

Based on the Mistral 7B large language model, fine-tuned through pre-trained model merging on the Date-DPO-v2 dataset, excelling in role-playing, creative writing, and general intelligent tasks.

Large Language Model

Transformers

Open Source License:Apache-2.0 #Role-playing Optimization #Creative Writing Enhancement #DPO Fine-tuning

Downloads 23

Release Time : 4/22/2024

Model Overview

Flammen21-mistral-7B is a large language model based on the Mistral 7B architecture, fine-tuned with Direct Preference Optimization (DPO) to enhance performance in role-playing, creative writing, and general intelligent tasks.

Model Features

Direct Preference Optimization Fine-tuning

Fine-tuned using the DPO method on the Date-DPO-v2 dataset, improving the model's performance on specific tasks.

LoRA Efficient Fine-tuning

Utilizes Low-Rank Adaptation (LoRA) technology, significantly reducing training resource requirements while maintaining model performance.

Long-context Processing

Supports a maximum context length of 4096 tokens, suitable for long-text tasks.

Model Capabilities

Role-playing

Creative Writing

Text Generation

Dialogue Systems

Content Creation

Use Cases

Entertainment

Role-playing Games

Engage in natural conversations as an AI character in games.

Provides an immersive role-playing experience.

Creative Writing Assistant

Helps writers generate creative content or overcome writer's block.

Inspires creative ideas and improves writing efficiency.

Education

Language Learning Partner

Acts as a conversational partner for language practice.

Provides a natural language exchange environment.

🚀 flammen21-mistral-7B

A Mistral 7B LLM built from merging pretrained models and finetuning on flammenai/Date-DPO-v2. Flammen specializes in exceptional character roleplay, creative writing, and general intelligence.

🚀 Quick Start

A Mistral 7B LLM is constructed by merging pretrained models and fine - tuning on flammenai/Date-DPO-v2. Flammen excels in character role - playing, creative writing, and general intelligence tasks.

✨ Features

Specializes in exceptional character roleplay.
Capable of high - quality creative writing.
Demonstrates general intelligence.

📦 Installation

No installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

No usage examples are provided in the original document, so this section is skipped.

📚 Documentation

Method

Finetuned using an L4 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

Configuration

LoRA, model, and training settings:

# 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=8,
    gradient_checkpointing=True,
    learning_rate=5e-5,
    lr_scheduler_type="cosine",
    max_steps=420,
    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()

🔧 Technical Details

No technical details beyond the finetuning method and configuration are provided in the original document, so this section is skipped.

📄 License

The project is licensed under the Apache - 2.0 license.

Property	Details
Library Name	transformers
License	apache - 2.0
Base Model	flammenai/flammen20 - mistral - 7B
Training Data	flammenai/Date - DPO - v2

image/png

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご