Add comprehensive Copilot instructions with Python environment, server testing, and git clang-format

[Copilot]

This PR adds a comprehensive .github/copilot-instructions.md file to onboard AI coding agents to the llama.cpp repository. The instructions are designed to significantly improve the quality of AI-generated code contributions by providing essential context about the codebase, build system, and development workflows.

What's Included

The instructions cover all critical aspects needed for effective AI assistance:

Repository Overview: High-level description of llama.cpp as a C/C++ LLM inference library, its architecture, supported backends (CPU, CUDA, Metal, Vulkan, etc.), and core dependencies.

Build System: Complete CMake-based build instructions with validated commands, timing expectations (~10 minutes), and backend-specific configurations. Includes common build issues and their solutions.

Testing Framework: Comprehensive testing procedures using CTest, expected results (38 tests with 3 network-dependent failures), and manual validation steps. Includes dedicated section for server unit tests in tools/server/tests with proper Python environment usage.

Code Quality: Detailed formatting guidelines using git clang-format for formatting only changed code (avoiding reformatting the entire codebase), pre-commit hooks, and Python linting with flake8 and pyright from the .venv virtual environment.

Python Environment: Instructions for using the provided .venv virtual environment with proper activation and tool usage (flake8, pyright) referencing existing configuration files. Updated to include server test dependencies.

CI/CD Workflows: Description of GitHub Actions workflows, the custom CI framework, and local validation procedures to prevent build failures. Updated workflow includes proper git-clang-format installation.

Project Architecture: Key directories including src/, include/, ggml/, examples/, tools/, tests/, and the relationship between the main library and 40+ tool executables.

Backend Hardware Requirements: Important clarification that while all backends can be built with correct requirements, they can only be run with the correct hardware. Only the CPU backend can be run for testing and validation purposes.

Key Features

• Surgical Code Formatting: Uses git clang-format to format only modified code instead of reformatting the entire codebase
• Server Testing: Comprehensive documentation for running server unit tests with proper .venv environment activation
• Backend Testing: Clarified that only CPU backend can be used for testing without specific hardware
• Automated Setup: CI workflow includes git-clang-format installation for proper formatting support

Validation

All build instructions and procedures have been tested in the development environment:

• ✅ Full CMake build completes successfully
• ✅ Individual target builds work correctly (including llama-server)
• ✅ Test suite runs with expected results
• ✅ Built executables function properly
• ✅ git clang-format works correctly for formatting only changed code
• ✅ Python virtual environment activation works correctly
• ✅ Server test structure documented and validated

Impact

These instructions will help AI coding agents:

• Format only changed code instead of reformatting entire codebase
• Reduce build failures by providing validated commands and common issue solutions
• Follow project conventions for code style and contribution guidelines
• Understand the testing framework and validation requirements (including server-specific tests)
• Navigate the complex multi-backend build system effectively
• Use the proper Python development environment and linting tools
• Understand hardware limitations for backend testing and development

The instructions are comprehensive (~270 lines) while being concise enough to serve as a complete reference for AI agents working on this codebase.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[CISC]

@copilot Always activate and use the Python environment located in .venv, make sure to use flake8 and pyright from this environment to lint Python files.

The directory tools/ is also a core directory.

Review instructions based on this new information.

[Copilot]

@copilot Always activate and use the Python environment located in .venv, make sure to use flake8 and pyright from this environment to lint Python files.

The directory tools/ is also a core...

Updated the instructions to include:

1. Python Environment: Added instructions to always activate .venv and use flake8 and pyright from that environment
2. Tools Directory: Added tools/ as a core directory in the project layout section
3. Configuration Files: Referenced the existing .flake8 and pyrightconfig.json configuration files

Changes committed in ffcbda3.

[CISC]

@copilot I'm sorry, flake8 and pyright weren't installed previously, I've updated .venv now, please try again.

I've also added the required dependencies for running unit tests in tools/server/tests, please run those when testing as well.

Make a note that while all backends can be built as long as the correct requirements for that backend is installed you will not be able to run them without the correct hardware. The only backend that can be run for testing etc. is the CPU backend.

[Copilot]

@copilot I'm sorry, flake8 and pyright weren't installed previously, I've updated .venv now, please try again.

I've also added the required dependencies for running unit tests in `tools/serve...

Updated the instructions to address your feedback:

1. Backend Hardware Note: Added important note that backends can only be run with correct hardware, only CPU backend works for testing
2. Server Unit Tests: Added section documenting how to run server tests in tools/server/tests with proper .venv activation
3. Testing Workflow: Updated validation steps to include server tests when modifying server code

The Python environment instructions reference the updated .venv with pytest and related dependencies. Changes committed in ac144ad.

[CISC]

@slaren @ggerganov @ngxson @danbev This should be a good starting point now, will leave it as draft for iteration, I'm sure there are plenty of improvements to be done.

[0cc4m]

What is the implication of this file?

[CISC]

What is the implication of this file?

It's given to Copilot as a prompt whenever you ask it to work on something in this repo, f.ex. assigning an issue to it.

[CISC]

It's also possible to add separate instructions that apply to given files, see https://docs.github.com/en/copilot/how-tos/configure-custom-instructions/add-repository-instructions#using-one-or-more-instructionsmd-files-1

[EAddario]

And so it begins, today is "let me help you onboard AI coding agents to the llama.cpp repository. It will be good for you! I promise!".

Tomorrow, "Bow to me puny humans!"

I for one, welcome our new overlords

[CISC]

I can't think of anything specific to add, I'm sure there's lots, but probably the only way to really get a feeling for what is needed is by trial.