| # SBD |
|
|
| **SBD** is an independent AI research and development project focused on building practical, accessible language models for **coding, conversation, and software development**. |
|
|
| Our goal is to experiment with smaller, efficient models that can run locally while still providing useful capabilities for developers and everyday users. |
|
|
| ## 🤖 Models |
|
|
| ### SBD4.5 |
|
|
| **SBD4.5** is a ~3B parameter coding and conversational language model. |
|
|
| It is designed for: |
|
|
| - 💻 Code generation |
| - 🐛 Debugging |
| - 🦀 Rust programming |
| - 💬 General conversation |
| - 📚 Code explanation |
| - 🔧 Software-development assistance |
| - ✨ Code completion |
| - 🧠 Technical question answering |
|
|
|
|
| ## 🎯 Goals |
|
|
| SBD focuses on: |
|
|
| - Creating capable small language models |
| - Improving coding assistance |
| - Making models easier to run locally |
| - Supporting open model formats such as GGUF |
| - Experimenting with efficient training and fine-tuning |
| - Building models that work well with tools such as LM Studio and llama.cpp |
| - Making AI development more accessible |
|
|
| ## 🧪 Research & Development |
|
|
| Models published by SBD may be experimental. |
|
|
| We regularly test: |
|
|
| - New datasets |
| - Training strategies |
| - Fine-tuning techniques |
| - Quantization methods |
| - Prompt formats |
| - Coding performance |
| - Local inference |
|
|
| Because of this, capabilities may differ between model versions. |
|
|
| ## 📦 Model Formats |
|
|
| Depending on the release, SBD models may be available in: |
|
|
| - Hugging Face Transformers / SafeTensors |
| - GGUF |
| - Quantized GGUF variants |
| - Full-precision checkpoints |
|
|
| Common GGUF releases may include: |
|
|
| - `Q8_0` |
| - `Q6_K` |
| - `Q5_K_M` |
| - `Q4_K_M` |
|
|
| ## 📊 Datasets |
|
|
| SBD experiments with coding and conversational datasets. |
|
|
| SBD4.5 uses coding data including: |
|
|
| **Convence/Rust-Coder** |
|
|
| https://huggingface.co/datasets/Convence/Rust-Coder |
|
|
| Additional datasets may be used for future model versions. |
|
|
| ## 🔓 Open Source |
|
|
| We aim to make SBD models easy to experiment with and run locally. |
|
|
| Unless stated otherwise on an individual model page, SBD models are released under permissive open-source licenses. |
|
|
| For example, **SBD4.5 is released under Apache License 2.0.** |
|
|
| Always check the individual model card for the exact license and usage conditions. |
|
|
| ## ⚠️ Experimental Models |
|
|
| SBD models can produce: |
|
|
| - Incorrect information |
| - Non-compiling code |
| - Bugs |
| - Hallucinated APIs |
| - Insecure code |
| - Unexpected responses |
|
|
| Generated code should always be reviewed and tested before being used in production. |
|
|
| ## 🚀 SBD Model Series |
|
|
| The SBD model family represents our ongoing experimentation with local and developer-focused AI. |
|
|
| Current generation: |
|
|
| > **SBD4.5 — 3B Coding & Chat Model** |
|
|
| More models, datasets, quantizations, and tools may be released as the project develops. |
|
|
| --- |
|
|
| ### Build. Experiment. Improve. |
|
|
| **SBD** |
|
|
