Add contributing guide

CONTRIBUTING.md

@@ -0,0 +1,75 @@
+# Contributing to bitnet.cpp
+
+Thanks for helping improve bitnet.cpp. This project provides optimized CPU and
+GPU inference support for 1-bit LLMs, including BitNet b1.58 models and the
+POWER8/PowerPC port maintained in this fork.
+
+## Good First Contributions
+
+Helpful contributions include:
+
+- Documentation fixes for setup, model conversion, inference, and benchmarking.
+- Reproducible notes for specific operating systems, compilers, CPUs, or GPUs.
+- Small tests for pure Python helpers and command-line behavior.
+- Build script improvements for CMake, PowerPC, POWER8, or GPU workflows.
+- Clarifications to kernel documentation, tuning guidance, and benchmark notes.
+
+Please keep each pull request focused on one topic. For larger changes, open an
+issue first so maintainers can discuss the target hardware, model format,
+dependencies, and expected validation.
+
+## Working Locally
+
+Start from a fresh branch:
+
+```bash
+git checkout -b your-change-name
+```
+
+Install Python dependencies before working on scripts:
+
+```bash
+pip install -r requirements.txt
+```
+
+If your change touches GPU utilities, also review `gpu/requirements.txt`.
+Do not commit downloaded models, generated benchmark outputs, local build
+directories, or machine-specific logs.
+
+## Validation
+
+Choose checks that match the files you changed:
+
+- Documentation-only changes: run `git diff --check` and review Markdown
+  rendering where practical.
+- Python helper changes: run the focused `pytest` command for the affected
+  tests, or add a small test when behavior changes.
+- Build or kernel changes: include the CMake or build command used, compiler
+  version, target architecture, and hardware tested.
+- Benchmark changes: include the model, quantization type, thread count, and
+  command used to reproduce the result.
+
+If you cannot run a full model download or hardware-specific benchmark, state
+that clearly in the pull request and describe the lighter validation you did
+run.
+
+## Pull Request Checklist
+
+- Explain the problem and the scope of the fix.
+- List the exact validation commands you ran.
+- Keep unrelated formatting changes out of the diff.
+- Avoid committing generated files, model weights, caches, or local logs.
+- Follow the Microsoft Open Source Code of Conduct in `CODE_OF_CONDUCT.md`.
+
+## Issue Reports
+
+For ordinary build, documentation, or usage issues, include:
+
+- Operating system and architecture.
+- Compiler and CMake versions.
+- Python version and environment manager, if relevant.
+- Model name, quantization type, and command used.
+- The complete error message or the smallest useful excerpt.
+
+Specific reproduction details are especially important for PowerPC, POWER8,
+GPU, and model-conversion issues, where behavior can vary by platform.