Awesome

<a href="https://unsloth.ai"><picture> <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20logo%20white%20text.png"> <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20logo%20black%20text.png"> <img alt="unsloth logo" src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20logo%20black%20text.png" height="110" style="max-width: 100%;"> </picture></a>

<a href="https://colab.research.google.com/drive/135ced7oHytdxu3N2DNe1Z0kqjyYIkDXp?usp=sharing"><img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/start free finetune button.png" height="48"></a> <a href="https://discord.gg/u54VK8m8tk"><img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/Discord button.png" height="48"></a> <a href="https://ko-fi.com/unsloth"><img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/buy me a coffee button.png" height="48"></a>

Finetune Llama 3, Mistral & Gemma 2-5x faster with 80% less memory!

✨ Finetune for Free

All notebooks are beginner friendly! Add your dataset, click "Run All", and you'll get a 2x faster finetuned model which can be exported to GGUF, vLLM or uploaded to Hugging Face.

• Benchmarking compared to FA2 + Hugging Face combined.
• Kaggle Notebooks for Llama-3 8b, Gemma 7b, Mistral 7b
• This conversational notebook is useful for Llama-3. And ChatML for Mistral 7b.
• This text completion notebook is for continued pretraining / raw text.

🦥 Unsloth.ai News

• 📣 NEW! Qwen1.5-7B, Qwen1.5-14B, Qwen1.5-32B, Qwen1.5-72B now work, courtesy of Firefly's PR #428
• 📣 NEW! Llama-3 8b now works! Llama-3 70b also works (change the model name in the notebook).
• 📣 NEW! ORPO support is here!
• 📣 NEW! Phi-3 3.8b support is here!
• 📣 NEW! We cut memory usage by a further 30% and now support fine-tuning of LLMs with 4x longer context windows! No change required if you're using our notebooks. To enable, simply change 1 line:

model = FastLanguageModel.get_peft_model(
    model,
    use_gradient_checkpointing = "unsloth", # <<<<<<<
)

• 📣 CodeGemma now works along with Gemma 7b and Gemma 2b
• 📣 2x faster inference added for all our models

🔗 Links and Resources

⭐ Key Features

• All kernels written in OpenAI's Triton language. Manual backprop engine.
• 0% loss in accuracy - no approximation methods - all exact.
• No change of hardware. Supports NVIDIA GPUs since 2018+. Minimum CUDA Capability 7.0 (V100, T4, Titan V, RTX 20, 30, 40x, A100, H100, L40 etc) Check your GPU! GTX 1070, 1080 works, but is slow.
• Works on Linux and Windows via WSL.
• Supports 4bit and 16bit QLoRA / LoRA finetuning via bitsandbytes.
• Open source trains 5x faster - see Unsloth Pro for up to 30x faster training!
• If you trained a model with 🦥Unsloth, you can use this cool sticker!   <img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/made with unsloth.png" height="50" align="center" />

🥇 Performance Benchmarking

• For the full list of reproducable benchmarking tables, go to our website

• Benchmarking table below was conducted by 🤗Hugging Face.

💾 Installation Instructions

Conda Installation

Select either pytorch-cuda=11.8 for CUDA 11.8 or pytorch-cuda=12.1 for CUDA 12.1. If you have mamba, use mamba instead of conda for faster solving. See this Github issue for help on debugging Conda installs.

conda create --name unsloth_env python=3.10
conda activate unsloth_env

conda install pytorch-cuda=<12.1/11.8> pytorch cudatoolkit xformers -c pytorch -c nvidia -c xformers

pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"

pip install --no-deps trl peft accelerate bitsandbytes

Pip Installation

Do NOT use this if you have Anaconda. You must use the Conda install method, or else stuff will BREAK.

1. Find your CUDA version via

import torch; torch.version.cuda

2. For Pytorch 2.1.0: You can update Pytorch via Pip (interchange cu121 / cu118). Go to https://pytorch.org/ to learn more. Select either cu118 for CUDA 11.8 or cu121 for CUDA 12.1. If you have a RTX 3060 or higher (A100, H100 etc), use the "ampere" path. For Pytorch 2.1.1: go to step 3. For Pytorch 2.2.0: go to step 4.

pip install --upgrade --force-reinstall --no-cache-dir torch==2.1.0 triton \
  --index-url https://download.pytorch.org/whl/cu121

pip install "unsloth[cu118] @ git+https://github.com/unslothai/unsloth.git"
pip install "unsloth[cu121] @ git+https://github.com/unslothai/unsloth.git"
pip install "unsloth[cu118-ampere] @ git+https://github.com/unslothai/unsloth.git"
pip install "unsloth[cu121-ampere] @ git+https://github.com/unslothai/unsloth.git"

3. For Pytorch 2.1.1: Use the "ampere" path for newer RTX 30xx GPUs or higher.

pip install --upgrade --force-reinstall --no-cache-dir torch==2.1.1 triton \
  --index-url https://download.pytorch.org/whl/cu121

pip install "unsloth[cu118-torch211] @ git+https://github.com/unslothai/unsloth.git"
pip install "unsloth[cu121-torch211] @ git+https://github.com/unslothai/unsloth.git"
pip install "unsloth[cu118-ampere-torch211] @ git+https://github.com/unslothai/unsloth.git"
pip install "unsloth[cu121-ampere-torch211] @ git+https://github.com/unslothai/unsloth.git"

4. For Pytorch 2.2.0: Use the "ampere" path for newer RTX 30xx GPUs or higher.

pip install --upgrade --force-reinstall --no-cache-dir torch==2.2.0 triton \
  --index-url https://download.pytorch.org/whl/cu121

pip install "unsloth[cu118-torch220] @ git+https://github.com/unslothai/unsloth.git"
pip install "unsloth[cu121-torch220] @ git+https://github.com/unslothai/unsloth.git"
pip install "unsloth[cu118-ampere-torch220] @ git+https://github.com/unslothai/unsloth.git"
pip install "unsloth[cu121-ampere-torch220] @ git+https://github.com/unslothai/unsloth.git"

5. If you get errors, try the below first, then go back to step 1:

pip install --upgrade pip

6. For Pytorch 2.2.1:

# RTX 3090, 4090 Ampere GPUs:
pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
pip install --no-deps packaging ninja einops flash-attn xformers trl peft accelerate bitsandbytes

# Pre Ampere RTX 2080, T4, GTX 1080 GPUs:
pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
pip install --no-deps xformers trl peft accelerate bitsandbytes

7. To troubleshoot installs try the below (all must succeed). Xformers should mostly all be available.

nvcc
python -m xformers.info
python -m bitsandbytes

📜 Documentation

• Go to our Wiki page for saving to GGUF, checkpointing, evaluation and more!
• We support Huggingface's TRL, Trainer, Seq2SeqTrainer or even Pytorch code!
• We're in 🤗Hugging Face's official docs! Check out the SFT docs and DPO docs!

from unsloth import FastLanguageModel
import torch
from trl import SFTTrainer
from transformers import TrainingArguments
from datasets import load_dataset
max_seq_length = 2048 # Supports RoPE Scaling interally, so choose any!
# Get LAION dataset
url = "https://huggingface.co/datasets/laion/OIG/resolve/main/unified_chip2.jsonl"
dataset = load_dataset("json", data_files = {"train" : url}, split = "train")

# 4bit pre quantized models we support for 4x faster downloading + no OOMs.
fourbit_models = [
    "unsloth/mistral-7b-bnb-4bit",
    "unsloth/mistral-7b-instruct-v0.2-bnb-4bit",
    "unsloth/llama-2-7b-bnb-4bit",
    "unsloth/gemma-7b-bnb-4bit",
    "unsloth/gemma-7b-it-bnb-4bit", # Instruct version of Gemma 7b
    "unsloth/gemma-2b-bnb-4bit",
    "unsloth/gemma-2b-it-bnb-4bit", # Instruct version of Gemma 2b
    "unsloth/llama-3-8b-bnb-4bit", # [NEW] 15 Trillion token Llama-3
    "unsloth/Phi-3-mini-4k-instruct-bnb-4bit",
] # More models at https://huggingface.co/unsloth

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "unsloth/llama-3-8b-bnb-4bit",
    max_seq_length = max_seq_length,
    dtype = None,
    load_in_4bit = True,
)

# Do model patching and add fast LoRA weights
model = FastLanguageModel.get_peft_model(
    model,
    r = 16,
    target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
                      "gate_proj", "up_proj", "down_proj",],
    lora_alpha = 16,
    lora_dropout = 0, # Supports any, but = 0 is optimized
    bias = "none",    # Supports any, but = "none" is optimized
    # [NEW] "unsloth" uses 30% less VRAM, fits 2x larger batch sizes!
    use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
    random_state = 3407,
    max_seq_length = max_seq_length,
    use_rslora = False,  # We support rank stabilized LoRA
    loftq_config = None, # And LoftQ
)

trainer = SFTTrainer(
    model = model,
    train_dataset = dataset,
    dataset_text_field = "text",
    max_seq_length = max_seq_length,
    tokenizer = tokenizer,
    args = TrainingArguments(
        per_device_train_batch_size = 2,
        gradient_accumulation_steps = 4,
        warmup_steps = 10,
        max_steps = 60,
        fp16 = not torch.cuda.is_bf16_supported(),
        bf16 = torch.cuda.is_bf16_supported(),
        logging_steps = 1,
        output_dir = "outputs",
        optim = "adamw_8bit",
        seed = 3407,
    ),
)
trainer.train()

# Go to https://github.com/unslothai/unsloth/wiki for advanced tips like
# (1) Saving to GGUF / merging to 16bit for vLLM
# (2) Continued training from a saved LoRA adapter
# (3) Adding an evaluation loop / OOMs
# (4) Cutomized chat templates

<a name="DPO"></a>

DPO Support

DPO (Direct Preference Optimization), PPO, Reward Modelling all seem to work as per 3rd party independent testing from Llama-Factory. We have a preliminary Google Colab notebook for reproducing Zephyr on Tesla T4 here: notebook.

We're in 🤗Hugging Face's official docs! We're on the SFT docs and the DPO docs!

from unsloth import FastLanguageModel, PatchDPOTrainer
PatchDPOTrainer()
import torch
from transformers import TrainingArguments
from trl import DPOTrainer

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "unsloth/zephyr-sft-bnb-4bit",
    max_seq_length = max_seq_length,
    dtype = None,
    load_in_4bit = True,
)

# Do model patching and add fast LoRA weights
model = FastLanguageModel.get_peft_model(
    model,
    r = 64,
    target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
                      "gate_proj", "up_proj", "down_proj",],
    lora_alpha = 64,
    lora_dropout = 0, # Supports any, but = 0 is optimized
    bias = "none",    # Supports any, but = "none" is optimized
    # [NEW] "unsloth" uses 30% less VRAM, fits 2x larger batch sizes!
    use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
    random_state = 3407,
    max_seq_length = max_seq_length,
)

dpo_trainer = DPOTrainer(
    model = model,
    ref_model = None,
    args = TrainingArguments(
        per_device_train_batch_size = 4,
        gradient_accumulation_steps = 8,
        warmup_ratio = 0.1,
        num_train_epochs = 3,
        fp16 = not torch.cuda.is_bf16_supported(),
        bf16 = torch.cuda.is_bf16_supported(),
        logging_steps = 1,
        optim = "adamw_8bit",
        seed = 42,
        output_dir = "outputs",
    ),
    beta = 0.1,
    train_dataset = YOUR_DATASET_HERE,
    # eval_dataset = YOUR_DATASET_HERE,
    tokenizer = tokenizer,
    max_length = 1024,
    max_prompt_length = 512,
)
dpo_trainer.train()

🥇 Detailed Benchmarking Tables

• Click "Code" for fully reproducible examples
• "Unsloth Equal" is a preview of our PRO version, with code stripped out. All settings and the loss curve remains identical.
• For the full list of benchmarking tables, go to our website

Llama-Factory 3rd party benchmarking

• Link to performance table. TGS: tokens per GPU per second. Model: LLaMA2-7B. GPU: NVIDIA A100 * 1. Batch size: 4. Gradient accumulation: 2. LoRA rank: 8. Max length: 1024.

Performance comparisons between popular models

Mistral 7b

CodeLlama 34b

1 Tesla T4

2 Tesla T4s via DDP

Performance comparisons on 1 Tesla T4 GPU:

One Tesla T4 on Google Colab bsz = 2, ga = 4, max_grad_norm = 0.3, num_train_epochs = 1, seed = 3047, lr = 2e-4, wd = 0.01, optim = "adamw_8bit", schedule = "linear", schedule_steps = 10

Peak Memory Usage

Two Tesla T4s on Kaggle bsz = 2, ga = 4, max_grad_norm = 0.3, num_train_epochs = 1, seed = 3047, lr = 2e-4, wd = 0.01, optim = "adamw_8bit", schedule = "linear", schedule_steps = 10

Peak Memory Usage on a Multi GPU System (2 GPUs)

• Slim Orca bsz=1 for all benchmarks since bsz=2 OOMs. We can handle bsz=2, but we benchmark it with bsz=1 for consistency.

<br>

Thank You to

• HuyNguyen-hust for making RoPE Embeddings 28% faster
• RandomInternetPreson for confirming WSL support
• 152334H for experimental DPO support
• atgctg for syntax highlighting