Summary

This PR updates the Vision Language Model (VLM) from Qwen2.5-VL-7B to Qwen3-VL-4B across the entire codebase.

Changes

VLM Model Updates

• LemonadeClient: Update VLM model definition from qwen2.5-vl-7b to qwen3-vl-4b
  • Model ID: Qwen3-VL-4B-Instruct-GGUF (was Qwen2.5-VL-7B-Instruct-GGUF)
  • Update agent profiles: chat, rag, vlm, and mcp agents now use Qwen3-VL-4B
• Documentation: Update VLM references in CLAUDE.md, README.md, EMR agent README
• Tests: Update all VLM model references in integration tests
• CI/CD: Update RAG workflow to use Qwen3-VL-4B
• Scripts: Update PowerShell start-lemonade script examples
• Claude Agents: Update RAG specialist agent example

Additional Improvements

• Lint Tool: Add pre-check to verify GAIA package is installed before running import validation

Files Changed (11 total)

• .claude/agents/rag-specialist.md - Update RAG VLM example
• .github/workflows/test_rag.yml - Update CI VLM model
• CLAUDE.md - Update default VLM model reference
• README.md - Update vision models feature description
• scripts/start-lemonade.ps1 - Update example command
• src/gaia/agents/emr/README.md - Update VLM extraction reference
• src/gaia/llm/lemonade_client.py - Update VLM model definition and agent profiles
• tests/test_lemonade_client.py - Update test assertions
• tests/test_rag_integration.py - Update required models list
• tests/test_vlm_integration.py - Update VLM initialization
• util/lint.py - Add GAIA installation pre-check

Testing

• Verified VLM model references are consistent across codebase
• Updated all agent profiles (chat, rag, vlm, mcp)
• Updated test assertions to match new model
• Updated CI workflows to use Qwen3-VL-4B

Impact

• Users will need to download the new Qwen3-VL-4B-Instruct-GGUF model (~3.3GB)
• The smaller 4B model provides faster inference while maintaining good quality
• All vision-enabled agents (chat, rag, emr, mcp) will use the updated model