Free Colab notebooks
— click and run
No local setup needed. Open in browser, connect the free T4 GPU, and run. All notebooks tested and working.
Fine-tune Llama 3.1 8B with Unsloth — 2x faster, free T4
Fine-tune Llama 3.1 8B on your own data in Google Colab. Unsloth makes it 2x faster and uses 60% less memory — fits on the free T4 GPU. Best notebook for custom LLM fine-tuning.
Fine-tune Llama 3.2 Vision — multimodal on free Colab
Fine-tune Llama 3.2's vision capabilities on your own image+text dataset. Build a custom medical imaging, document reading, or product recognition model for free.
Fine-tune Llama 3.2 for Chat — conversational format
Fine-tune Llama 3.2 (1B and 3B) specifically for chat using ShareGPT/ChatML format. Small models that run on the free T4 and can be deployed on CPU after training.
Faster-Whisper large-v3 — batch transcription with SRT output
Transcribe any audio or video file 4x faster than original Whisper using CTranslate2 optimisation. Outputs .txt, .srt subtitles, and .json with timestamps. Supports Hindi, Tamil, Telugu.
Orpheus TTS — free studio-quality voice generation on Colab
Generate studio-quality speech using Kokoro 82M, the best open-source TTS model. Multiple voices, emotion control, and Hindi support. Zero cost, runs on free T4.
WhisperX — speaker diarization and word-level timestamps
Full WhisperX tutorial: transcription with word-level timestamps and speaker diarization. Uses whisperx library directly. Perfect for meeting notes, interviews, and research.
Flux.1 Dev image generation — better than Midjourney, free
Generate photorealistic images with Flux.1 Dev, the best open-source image model. Runs on Colab T4. Prompt-following that beats Midjourney v6. Outputs at up to 1024x1024.
Stable Diffusion AUTOMATIC1111 WebUI — full UI on Colab
Run the full AUTOMATIC1111 Stable Diffusion WebUI in Colab. Access txt2img, img2img, inpainting, ControlNet, LoRA, and upscaling — all through a browser. Classic notebook, very reliable.
SDXL + ComfyUI on Colab — node-based workflow builder
Run ComfyUI (the most powerful image generation workflow tool) on Colab T4. Build any image generation pipeline visually with SDXL, ControlNet, IP-Adapter, and custom nodes.
LTX-Video — text to video on free Colab T4
Generate short videos from text prompts using LTX-Video, the fastest open-source video model. Runs on the free T4 GPU. 5-second videos in under 2 minutes.
Wan 2.1 Text-to-Video — best open-source video model
Generate high-quality 5-second videos with Wan 2.1, currently the best open-source video generation model. Beats Sora on most benchmarks. Requires Colab Pro for the 14B model.
FramePack — image to video with consistent motion
Turn any image into a smooth video clip with FramePack. Unlike other I2V models, FramePack maintains consistent character/object appearance across all frames. Free T4 compatible.
Build a RAG system with ChromaDB + LlamaIndex
Build a complete RAG pipeline from scratch — load documents, chunk them, embed with HuggingFace, store in ChromaDB, and query with LlamaIndex. Fully working in 30 minutes.
LangChain RAG over PDF documents
Build a PDF question-answering chatbot with LangChain. Upload any PDF and chat with it. Uses FAISS vector store and HuggingFace embeddings — no API key required for embedding.
Run Ollama on free Colab T4 — any local LLM in browser
Run Ollama on Colab's free T4 GPU and access models like Llama 3, Mistral, DeepSeek-R1 through a REST API. Full Ollama server running in the cloud — no local GPU needed.
Google ML Crash Course — official interactive notebooks
Google's own ML crash course as interactive Colab notebooks. Linear regression, neural networks, classification, fairness — complete curriculum from Google engineers.
HuggingFace Transformers — run any model in 5 lines
Official HuggingFace notebook for running transformers models. Sentiment analysis, text generation, summarisation, translation, question answering — all with the pipeline() API.
AI Stock Analysis — NSE/BSE data + LLM insights
Fetch NSE/BSE stock data with yfinance, compute technical indicators (RSI, MACD, Bollinger Bands), and generate plain-English analysis using an LLM. Adapted for Indian markets.
PandasAI — chat with your data using natural language
Ask questions about any CSV/Excel dataset in plain English. "What is the average salary by city?" — PandasAI translates your question to pandas code and runs it automatically.
AutoGen multi-agent chat — AI team that writes code
Build a multi-agent system where AI agents collaborate to solve coding tasks. One agent writes code, another reviews it, a third tests it — all automatically. Microsoft's AutoGen framework.
CrewAI — build a research + writing agent team
Build an AI crew where a researcher agent gathers information and a writer agent turns it into a blog post — automatically. Complete CrewAI tutorial from scratch.
Train your own reasoning model with GRPO on free Colab T4
Run DeepSeek-R1 7B (the reasoning model that shocked the world) on Colab's free T4. See chain-of-thought reasoning in action — the model shows its thinking process.