Below are Colab notebooks, organized by model. You can also view all notebooks in our docs.
The notebooks run locally and feature data prep, training and inference. Read our fine-tuning guide.

Model	Type	Notebook Link
Qwen3.5 (4B)	Vision
Qwen3.5 (2B)	Vision
gpt-oss (20B)	Fine-tuning
gpt-oss (20B)	GRPO
Qwen3 (14B)	Conversational
Qwen3-VL (8B)	Vision
Qwen3-Embedding (0.6B)	Embeddings
Qwen3: Advanced GRPO	GRPO
Gemma 3 (4B)	Vision
Gemma 3N (4B)	Audio
embeddinggemma (300M)	Embeddings
Mistral Ministral 3 (3B)	Vision
Mistral v0.3 (7B)	Vision
Llama 3.1 (8B) Alpaca	Alpaca
Llama 3.2 (1B + 3B)	Conversational
Phi-4 (14B)	Conversational
Orpheus-TTS (3B)	TTS

Model	Type	Notebook Link
(A100) gpt oss (20B)	GRPO
gpt oss (20B)	GRPO
gpt oss (20B)	GRPO
Phi 4 (14B)	GRPO
Llama3.1 (8B)	GRPO
Qwen3 (4B)	GRPO
Gemma3 (1B)	GRPO
Qwen2.5 (3B)	GRPO
LFM2.5 (1.2B)	GRPO
DeepSeek R1 0528 Qwen3 (8B)	GRPO
Mistral v0.3 (7B)	GRPO

Model	Type	Notebook Link
Orpheus (3B)	TTS
Llasa TTS (3B)	TTS
Sesame CSM (1B)	TTS
Oute TTS (1B)	TTS
Llasa TTS (1B)	TTS
Spark TTS (0.5B)	TTS

Model	Type	Notebook Link
Pixtral (12B)	Vision
ERNIE 4 5 VL 28B A3B PT	Vision
Llama3.2 (11B)	Vision
Qwen3 VL (8B)	Vision
Qwen3 VL (8B)	Vision GRPO
Qwen3 5 (4B)	Vision
Qwen3 5 (2B)	Vision
Qwen3 5 (0 8B)	Vision
Ministral3 VL (3B)	Vision
Gemma3N (4B)	Vision
Gemma3 (4B)	Vision
Gemma3 (4B)	Vision GRPO
Qwen2.5 VL (7B)	Vision
Qwen2.5 VL (7B)	Vision GRPO
LFM2.5 VL (1.6B)	Vision
Qwen2 VL (7B)	Vision

Model	Type	Notebook Link
EmbeddingGemma (300M)
All MiniLM L6 v2
Qwen3 Embedding (4B)
Qwen3 Embedding (0 6B)
BGE M3
ModernBert
ModernBERT (Large)	Classification

Model	Type	Notebook Link
Whisper (Large)	Fine Tuning

Model	Type	Notebook Link
Deepseek OCR (3B)	Fine Tuning
Deepseek OCR (3B)	Evaluation
Deepseek OCR (3B)	Eval
Deepseek OCR 2 (3B)
Paddle OCR (1B)	Vision

Model	Type	Notebook Link
ModernBert
ModernBERT (Large)	Classification

Model	Type	Notebook Link
Deepseek OCR (3B)	Fine Tuning
Deepseek OCR (3B)	Evaluation
Deepseek OCR (3B)	Eval
Deepseek OCR 2 (3B)

Model	Type	Notebook Link
ERNIE 4 5 VL 28B A3B PT	Vision
ERNIE 4 5 21B A3B PT	Conversational

Model	Type	Notebook Link
(A100) GLM Flash(80GB)

Model	Type	Notebook Link
(A100) gpt oss (120B)	Fine Tuning
gpt oss (20B)	GRPO 2048
gpt oss (20B)	Fine Tuning
gpt oss (20B)	Fine Tuning
gpt oss BNB (20B)	Inference
(OpenEnv) gpt oss (20B)	GRPO 2048
(DGX Spark) gpt oss (20B)	GRPO 2048
gpt oss BF16 (20B)	GRPO 2048
(OpenEnv) gpt oss BF16 (20B)	GRPO 2048
gpt oss MXFP4 (20B)	Inference

Model	Type	Notebook Link
EmbeddingGemma (300M)
FunctionGemma (270M)	Tool Calling
FunctionGemma (270M)	Mobile Actions
FunctionGemma (270M)	Inference
FunctionGemma (270M)	Conversational
(A100) Gemma3 (27B)	Conversational
CodeGemma (7B)	Conversational
Gemma3N (4B)	Vision
Gemma3N (4B)	Multimodal
Gemma3N (4B)	Audio
Gemma3N (2B)	Inference
Gemma3 (4B)	Vision
Gemma3 (4B)	Vision GRPO
Gemma3 (4B)	Conversational
Gemma3 (270M)	Conversational
Gemma3 (270M)
Gemma2 (9B)	Alpaca
Gemma2 (2B)	Alpaca

Model	Type	Notebook Link
Granite4.0 (3B)	Conversational
Granite4.0 (350M)	Conversational

Model	Type	Notebook Link
Liquid LFM2 (1.2B)	Conversational
Liquid LFM2	Conversational
Falcon H1 (0.5B)	Alpaca
Falcon H1	Alpaca

Model	Type	Notebook Link
(A100) Llama3.3 (70B)	Conversational
Llama3.2 (1B)	RAFT
Llama3.2 (1B)	FP8 GRPO
Llama3.2 (1B and 3B)	Conversational
Llama3.2 (11B)	Vision
Llama3.1 (8B)	Inference
Llama3.1 (8B)	Alpaca
Llama3 (8B)	Ollama
Llama3 (8B)	ORPO
Llama3 (8B)	Conversational
Llama3 (8B)	Alpaca
TinyLlama (1.1B)	Alpaca

Model	Type	Notebook Link
Magistral (24B)	Reasoning Conversational
Mistral Small (22B)	Alpaca
Pixtral (12B)	Vision
Mistral Nemo (12B)	Alpaca
Zephyr (7B)	DPO
Mistral (7B)	Text Completion
Ministral3 (3B)	GRPO Sudoku
Ministral3 VL (3B)	Vision
Mistral v0.3 (7B)	Conversational
Mistral v0.3 (7B)	CPT
Mistral v0.3 (7B)	Alpaca

Model	Type	Notebook Link
(A100) Nemotron Nano 3 30B A3B
(A100) Nemotron 3 Nano 30B A3B

Model	Type	Notebook Link
Paddle OCR (1B)	Vision

Model	Type	Notebook Link
Phi 4	Conversational
Phi 3.5 Mini	Conversational
Phi 3 Medium	Conversational

Model	Type	Notebook Link
(A100) Qwen3 (32B)	Reasoning Conversational
(A100) Qwen 3 5 27B(80GB)
TinyQwen3 MoE
Qwen3 (8B)	FP8 GRPO
Qwen3 (4B)	Thinking
Qwen3 (4B)	QAT
Qwen3 (4B)	Conversational
Qwen3 (14B)	Reasoning Conversational
Qwen3 (14B)	Alpaca
Qwen3 (14B)
Qwen3 (0.6B)	Reasoning Conversational
Qwen3 (0 6B)
Qwen3 VL (8B)	Vision
Qwen3 VL (8B)	Vision GRPO
Qwen3 MoE
Qwen3 Embedding (4B)
Qwen3 Embedding (0 6B)
Qwen3 5 (4B)	Vision
Qwen3 5 (2B)	Vision
Qwen3 5 (0 8B)	Vision
Qwen3 5 MoE
Qwen2.5 (7B)	Alpaca
Qwen2.5 VL (7B)	Vision
Qwen2.5 VL (7B)	Vision GRPO
Qwen2.5 Coder (14B)	Conversational
Qwen2.5 Coder (1.5B)	Tool Calling
Qwen2 (7B)	Alpaca
Qwen2 VL (7B)	Vision

Usecase	Model	Notebook Link
Text Classification	Llama 3.1 (8B)
Tool Calling	Qwen2.5-Coder (1.5B)
Multiple Datasets
KTO	Qwen2.5-Instruct (1.5B)
Inference Chat UI	LLaMa 3.2 Vision
Conversational	LLaMa 3.2 (1B and 3B)
ChatML	Mistral (7B)
Text Completion	Mistral (7B)

Model	Type	Notebook Link
LFM2.5 (1.2B)	Text Completion
LFM2.5 (1.2B)	Conversational
LFM2.5 (1.2B)
LFM2.5 VL (1.6B)	Vision
Unsloth	Studio
Synthetic Data Hackathon	Synthetic Data
NeMo Gym Sudoku
NeMo Gym Multi Environment
CodeForces cot Finetune for Reasoning on CodeForces	Reasoning

Click for all our Kaggle notebooks categorized by model:

GRPO & Reinforcement Learning Notebooks

Model	Type	Notebook Link
(A100) gpt oss (20B)	GRPO
gpt oss (20B)	GRPO
gpt oss (20B)	GRPO
Phi 4 (14B)	GRPO
Llama3.1 (8B)	GRPO
Qwen3 (4B)	GRPO
Gemma3 (1B)	GRPO
Qwen2.5 (3B)	GRPO
DeepSeek R1 0528 Qwen3 (8B)	GRPO
Mistral v0.3 (7B)	GRPO

Text-to-Speech (TTS) Notebooks

Model	Type	Notebook Link
Orpheus (3B)	TTS
Llasa TTS (3B)	TTS
Sesame CSM (1B)	TTS
Oute TTS (1B)	TTS
Llasa TTS (1B)	TTS
Spark TTS (0.5B)	TTS

Vision (Multimodal) Notebooks

Model	Type	Notebook Link
Pixtral (12B)	Vision
ERNIE 4 5 VL 28B A3B PT	Vision
Llama3.2 (11B)	Vision
Qwen3 VL (8B)	Vision
Qwen3 VL (8B)	Vision GRPO
Ministral3 VL (3B)	Vision
Gemma3N (4B)	Vision
Gemma3 (4B)	Vision
Gemma3 (4B)	Vision GRPO
Qwen2.5 VL (7B)	Vision
Qwen2.5 VL (7B)	Vision GRPO
Qwen2 VL (7B)	Vision

Embedding Notebooks

Model	Type	Notebook Link
EmbeddingGemma (300M)
All MiniLM L6 v2
Qwen3 Embedding (4B)
Qwen3 Embedding (0 6B)
BGE M3
ModernBert
ModernBERT (Large)	Classification

Speech-to-Text (STT) Notebooks

Model	Type	Notebook Link
Whisper (Large)	Fine Tuning

OCR Notebooks

Model	Type	Notebook Link
Deepseek OCR (3B)	Fine Tuning
Deepseek OCR (3B)	Evaluation
Deepseek OCR (3B)	Eval
Deepseek OCR 2 (3B)
Paddle OCR (1B)	Vision

BERT Notebooks

Model	Type	Notebook Link
ModernBert
ModernBERT (Large)	Classification

Deepseek Notebooks

Model	Type	Notebook Link
Deepseek OCR (3B)	Fine Tuning
Deepseek OCR (3B)	Evaluation
Deepseek OCR (3B)	Eval
Deepseek OCR 2 (3B)

ERNIE Notebooks

Model	Type	Notebook Link
ERNIE 4 5 VL 28B A3B PT	Vision
ERNIE 4 5 21B A3B PT	Conversational

GPT-OSS Notebooks

Model	Type	Notebook Link
(A100) gpt oss (120B)	Fine Tuning
gpt oss (20B)	Fine Tuning
gpt oss (20B)	Fine Tuning
gpt oss BNB (20B)	Inference
gpt oss MXFP4 (20B)	Inference

Gemma Notebooks

Model	Type	Notebook Link
EmbeddingGemma (300M)
(A100) Gemma3 (27B)	Conversational
CodeGemma (7B)	Conversational
Gemma3N (4B)	Vision
Gemma3N (4B)	Multimodal
Gemma3N (4B)	Audio
Gemma3N (2B)	Inference
Gemma3 (4B)	Vision
Gemma3 (4B)	Vision GRPO
Gemma3 (4B)	Conversational
Gemma3 (270M)	Conversational
Gemma2 (9B)	Alpaca
Gemma2 (2B)	Alpaca

Granite Notebooks

Model	Type	Notebook Link
Granite4.0 (3B)	Conversational
Granite4.0 (350M)	Conversational

Linear Attention Notebooks

Model	Type	Notebook Link
Liquid LFM2 (1.2B)	Conversational
Falcon H1 (0.5B)	Alpaca

Llama Notebooks

Model	Type	Notebook Link
(A100) Llama3.3 (70B)	Conversational
Llama3.2 (1B)	RAFT
Llama3.2 (1B)	FP8 GRPO
Llama3.2 (1B and 3B)	Conversational
Llama3.2 (11B)	Vision
Llama3.1 (8B)	Inference
Llama3.1 (8B)	Alpaca
Llama3 (8B)	Ollama
Llama3 (8B)	ORPO
Llama3 (8B)	Conversational
Llama3 (8B)	Alpaca
TinyLlama (1.1B)	Alpaca

Mistral Notebooks

Model	Type	Notebook Link
Magistral (24B)	Reasoning Conversational
Mistral Small (22B)	Alpaca
Pixtral (12B)	Vision
Mistral Nemo (12B)	Alpaca
Zephyr (7B)	DPO
Mistral (7B)	Text Completion
Ministral3 (3B)	GRPO Sudoku
Ministral3 VL (3B)	Vision
Mistral v0.3 (7B)	Conversational
Mistral v0.3 (7B)	CPT
Mistral v0.3 (7B)	Alpaca

Nemotron Notebooks

Model	Type	Notebook Link
(A100) Nemotron Nano 3 30B A3B
(A100) Nemotron 3 Nano 30B A3B

Paddle Notebooks

Model	Type	Notebook Link
Paddle OCR (1B)	Vision

Phi Notebooks

Model	Type	Notebook Link
Phi 4	Conversational
Phi 3.5 Mini	Conversational
Phi 3 Medium	Conversational

Qwen Notebooks

Model	Type	Notebook Link
(A100) Qwen3 (32B)	Reasoning Conversational
Qwen3 (8B)	FP8 GRPO
Qwen3 (4B)	Thinking
Qwen3 (4B)	QAT
Qwen3 (4B)	Conversational
Qwen3 (14B)	Reasoning Conversational
Qwen3 (14B)	Alpaca
Qwen3 (14B)
Qwen3 VL (8B)	Vision
Qwen3 VL (8B)	Vision GRPO
Qwen3 Embedding (4B)
Qwen3 Embedding (0 6B)
Qwen2.5 (7B)	Alpaca
Qwen2.5 VL (7B)	Vision
Qwen2.5 VL (7B)	Vision GRPO
Qwen2.5 Coder (14B)	Conversational
Qwen2.5 Coder (1.5B)	Tool Calling
Qwen2 (7B)	Alpaca
Qwen2 VL (7B)	Vision

Other Notebooks

Model	Type	Notebook Link
Unsloth	Studio
CodeForces cot Finetune for Reasoning on CodeForces	Reasoning

• NumPy 2.x ↔ soxr: NumPy 2.x breaks soxr, causing Unsloth import failures. Pin numpy<2 to resolve. Use pip install --force-reinstall "numpy<2" if needed. Impact: Prevents Unsloth from running.
• soxr reinstall: pip install --force-reinstall soxr can pull NumPy 2.x back unless using --no-deps. Use pip install --force-reinstall --no-deps soxr to avoid this. Impact: May reintroduce NumPy 2.x and break Unsloth imports.
• typing_extensions: Older typing_extensions can break torch import (TypeIs missing) until upgraded. Upgrade with pip install --upgrade typing_extensions. Impact: Prevents PyTorch from importing correctly.
• Resolver warnings: Pinning numpy<2 can cause pip resolver warnings with SciPy/Numba; typically non-fatal. Impact: Cosmetic warnings only, does not affect functionality.
• ROCm / triton_key: LoRA backward can crash under torch.compile if Triton lacks triton_key; workaround is to disable Inductor/compile on ROCm (handled in code now, but worth noting). Impact: May cause training crashes on AMD GPUs when using torch.compile.

If you'd like to contribute to our notebooks, here's a guide to get you started:

1. Find the Template: We've provided a template notebook called Template_Notebook.ipynb in the root directory of this project. This template contains the basic structure and formatting guidelines for all notebooks in this collection.
2. Create Your Notebook:
   • Make a copy of Template_Notebook.ipynb.
   • Rename the copied file to follow this naming convention:
     • LLM Notebooks: <Model Name>-<Type>.ipynb (e.g., Mistral_v0.3_(7B)-Alpaca.ipynb)
     • Vision Notebooks: <Model Name>-Vision.ipynb (e.g., Llava_v1.6_(7B)-Vision.ipynb)
     • Example of <Type>: Alpaca, Conversational, CPT, DPO, ORPO, Text_Completion, CSV, Inference, Unsloth_Studio
3. Place in original_template: Once your notebook is ready, move it to the original_template directory.
4. Update Notebooks: Run the following command in your terminal:

   python update_all_notebooks.py

   This script will automatically:
   • Copy your notebook from original_template to the notebooks directory.
   • Update the notebook's internal sections (like Installation, News) to ensure consistency.
   • Add your notebook to the appropriate list in this README.md file.
5. Create a Pull Request: After that, just create a pull request (PR) to merge your changes, making it available for everyone!
   • We appreciate your contributions and look forward to reviewing your notebooks!