Add YAML configuration support to llama-cli with --config flag

[devin-ai-integration]

Make sure to read the contributing guidelines before submitting a PR

Summary

This PR adds comprehensive YAML configuration file support to llama-cli with a new --config flag. Users can now specify model parameters, sampling settings, and other options via YAML files instead of lengthy command lines, while maintaining full backward compatibility.

Link to Devin run: https://app.devin.ai/sessions/05f5de10f98f4d50a16fb685814bec88
Requested by: Jaime Mizrachi (@jaime-leo)

Key Features

• --config <path> flag: Load parameters from YAML configuration files
• Proper precedence: Command-line flags > YAML config > defaults
• Strict validation: Unknown YAML keys are rejected with helpful error messages
• Smart path resolution: Relative paths in YAML resolve relative to the config file directory
• Comprehensive coverage: Supports model, sampling, speculative decoding, and other parameter groups

Example Usage

# Use YAML config only
llama-cli --config configs/minimal.yaml

# Override YAML values with flags  
llama-cli --config configs/minimal.yaml -n 32 --temp 0.8

Example minimal.yaml:

model:
  path: models/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf
n_ctx: 256
sampling:
  seed: 42
  temp: 0.0
prompt: "Hello from YAML"
n_predict: 16
simple_io: true

Implementation Details

Architecture

• common/config.cpp: YAML parser with validation and path resolution
• Pre-scan approach: --config is processed before normal argument parsing to ensure proper precedence
• yaml-cpp dependency: Added via FetchContent with system package fallback

Testing

• Unit tests: Config parsing, error handling, path resolution (test-config-yaml.cpp)
• Integration tests: YAML-only, YAML+overrides, parity verification against equivalent flags
• CI workflow: Dedicated lightweight workflow with tiny model for testing

Files Changed

• common/config.{h,cpp} - YAML configuration loader (new)
• common/CMakeLists.txt - yaml-cpp dependency
• common/arg.cpp - --config flag and pre-scan logic
• tests/test-config-yaml.cpp - Unit tests (new)
• tests/CMakeLists.txt - Integration test registration
• tests/test-yaml-parity.sh - Parity verification script (new)
• configs/{minimal,override}.yaml - Example configurations (new)
• .github/workflows/config.yml - CI workflow for config tests (new)
• README.md - Usage documentation and examples

Review Checklist

High Priority:

• Verify pre-scan logic in arg.cpp doesn't interfere with existing argument parsing edge cases
• Check path resolution logic works correctly across platforms (Windows/Unix path separators)
• Confirm YAML key validation list in config.cpp covers all relevant common_params fields
• Test error handling for malformed YAML files and edge cases

Medium Priority:

• Verify CMake FetchContent integration works reliably across build environments
• Ensure integration tests are deterministic and don't depend on external model availability
• Check that precedence logic handles complex flag combinations correctly
• Review performance impact of YAML parsing on startup time

Low Priority:

• Validate example configs match documented usage patterns
• Confirm CI workflow is appropriately scoped and efficient

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring