• Supports Llama (7, 13, 70b), Yi (6, 34b), Mistral (7b), Tinyllama, CodeLlama (7, 13, 34b), and all Llama / Mistral derived architectures!
• All kernels written in OpenAI's Triton language.
• 0% loss in accuracy - no approximation methods - all exact.
• No change of hardware necessary. Supports NVIDIA GPUs since 2018+. Minimum CUDA Compute Capability 7.0 (V100, T4, Titan V, RTX 20, 30, 40x, A100, H100, L40 etc) Check your GPU
• NEW! Works on Linux and Windows via WSL.
• NEW! Experimental support for DPO (Direct Preference Optimization)!
• Supports 4bit and 16bit QLoRA / LoRA finetuning via bitsandbytes.
• Open source version trains 5x faster or you can check out Unsloth Pro and Max codepaths for 30x faster training!

Join our Discord! If you trained a model with Unsloth, we made a cool sticker!!

Unsloth currently only supports Linux distros and Pytorch == 2.1.

conda install cudatoolkit xformers bitsandbytes pytorch pytorch-cuda=12.1 \
  -c pytorch -c nvidia -c xformers -c conda-forge -y
pip install "unsloth[kaggle] @ git+https://github.com/unslothai/unsloth.git"

1. Find your CUDA version via

import torch; torch.version.cuda

2. We only support Pytorch 2.1 (2.1.1 bugs out for now): You can update Pytorch via Pip (interchange cu121 / cu118)

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

2. 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.

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"

Change cu121 to cu118 for CUDA version 11.8 or 12.1. Go to https://pytorch.org/ to learn more.

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

pip install --upgrade pip

We support Huggingface's TRL, Trainer, Seq2SeqTrainer or even Pytorch code!

from unsloth import FastLlamaModel, FastMistralModel
import torch
max_seq_length = 2048 # Can change to any number <= 4096
dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
load_in_4bit = True # Use 4bit quantization to reduce memory usage. Can be False.

# Load Llama model
model, tokenizer = FastLlamaModel.from_pretrained(
    model_name = "unsloth/llama-2-7b", # Supports any llama model eg meta-llama/Llama-2-7b-hf
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
    # token = "hf_...", # use one if using gated models like meta-llama/Llama-2-7b-hf
)

# Do model patching and add fast LoRA weights
model = FastLlamaModel.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, # Currently only supports dropout = 0
    bias = "none",    # Currently only supports bias = "none"
    use_gradient_checkpointing = True,
    random_state = 3407,
    max_seq_length = max_seq_length,
)

trainer = .... Use Huggingface's Trainer and dataset loading (TRL, transformers etc)

152334H hacked Unsloth to work with DPO via TRL!

1. Hack the model's config.json to be llama model. Example gist.
2. Use Unsloth for DPO for both base and reference models. Example gist.

1. Support Mixtral.
2. Does not support non Llama models - we do so in the future.

Time taken for 1 epoch

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

Time taken for 1 epoch

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.

Click "Code" for a fully reproducible example. "Unsloth Equal" is a preview of our PRO version, with code stripped out. All settings and the loss curve remains identical.

Manual autograd, Triton kernels etc. See our Benchmark Breakdown for more info!

$$ \begin{align} y &= \frac{x_i}{\sqrt{\frac{1}{n}\sum{x_i^2}+\epsilon}} \cdot w \\ y &= \frac{x_i}{\sqrt{\frac{1}{n}\sum{x_i^2}+\epsilon}} \cdot w \\ r &= \frac{1}{\sqrt{\frac{1}{n}\sum{x_i^2}+\epsilon}} \\ \frac{dC}{dX} &= \frac{1}{n} r \bigg( n (dY \cdot w) - \bigg( x_i \cdot r \cdot \sum{dY \cdot y_i } \bigg) \bigg) \end{align} $$

1. Sometimes bitsandbytes or xformers does not link properly. Try running:

!ldconfig /usr/lib64-nvidia

2. Windows is not supported as of yet - we rely on Xformers and Triton support, so until both packages support Windows officially, Unsloth will then support Windows.

3. If it doesn't install - maybe try updating pip.

1. RandomInternetPreson for confirming WSL support
2. 152334H for experimental DPO support
3. atgctg for syntax highlighting