🦥
UNSLOTH.AI STUDIO Fast LLM Fine-Tuning & Ollama Export Platform
Unsloth AVX-512 Ready
Compute Hardware ⚡
30 vCPUs
Intel Xeon Gold · 0% usage
System Memory 🧠
125 GB
5.4 GB Used (4.3%)
Active Training Jobs 🎯
0 Active
0 Total Jobs Executed
Ollama Gateway Link 🔗
Connected
Local Modelfile Auto-Sync

🛠️ Fine-Tuning Job Builder

Unsloth 5x Fast SFT
You can select a preset above or type any HuggingFace model ID.

⚡ Unsloth Optimizations

  • ✔
    2x - 5x Faster Training Speed: Hand-written OpenAI Triton and AVX-512 kernels bypass PyTorch autograd bottlenecks.
  • ✔
    70% - 80% Memory Reduction: QLoRA 4-bit quantization allows training large 7B/14B/32B models with minimal RAM footprint.
  • ✔
    0% Loss in Accuracy: Mathematically identical outputs to standard full precision backpropagation.
  • ✔
    Direct Ollama & GGUF Export: Export fine-tuned LoRA weights directly into GGUF or generate an Ollama Modelfile.

📋 Training Jobs History

Job Name Base Model Dataset Status Loss Progress Started Actions
Loading jobs...

Fine-Tuned Checkpoints & Ollama Exporter

Register your completed LoRA fine-tuned models directly into the local Ollama instance.

Loading checkpoints...

📂 Available Datasets

Filename Samples Size Action
Loading datasets...

📤 Upload Custom Dataset

Supported format: Alpaca (instruction, input, output) or Chat (conversations: [{role, content}]).

📓 Unsloth Interactive JupyterLab Workspace

Open pre-configured interactive notebooks for full code-level fine-tuning control.

01_Qwen_2_5_Fine_Tuning.ipynb
Fine-tune Qwen 2.5 1.5B/7B/14B/32B with Unsloth Fast SFT and LoRA adapters.
02_Llama_3_3_Fine_Tuning.ipynb
Fine-tune Meta Llama 3.2 3B and Llama 3.3 70B efficiently.
03_Export_to_GGUF_and_Ollama.ipynb
Convert trained LoRA weights into GGUF quantization format for local Ollama serving.
Action completed!