v0.2.4

Added

• Hardware-Optimized Profiles for GPT-OSS Models: New built-in profiles optimized for different GPU architectures
  • gpt_oss_ampere: Optimized for NVIDIA A100 GPUs
  • gpt_oss_hopper: Optimized for NVIDIA H100/H200 GPUs
  • gpt_oss_blackwell: Optimized for NVIDIA Blackwell (B100/B200) GPUs
  • Based on official vLLM GPT recipes
• Shortcuts System: Save and quickly launch model + profile combinations
  • Quick launch from CLI: vllm-cli serve --shortcut NAME
  • Manage shortcuts through interactive mode or CLI commands
  • Import/export shortcuts for sharing configurations
• Ollama Model Support: Full integration with Ollama-downloaded models
  • Automatic discovery in user (~/.ollama) and system (/usr/share/ollama) directories
  • GGUF format detection and experimental serving support
• Environment Variable Management: Two-tier system for complete control
  • Universal environment variables for all servers
  • Profile-specific environment variables (override universal)
  • Clear indication of environment sources when launching
• GPU Selection: Select specific GPUs for model serving
  • CLI: --device 0,1 to use specific GPUs
  • Interactive UI for GPU selection in advanced settings
  • Automatic tensor_parallel_size adjustment
• Enhanced System Information: vLLM built-in feature detection
  • Detailed attention backend availability (Flash Attention 2/3, xFormers)
  • Feature compatibility checking per backend
• Server Cleanup Control: Configure server behavior on CLI exit
• Extended vLLM Arguments: Added 16+ new arguments for v1 engine
  • Performance, optimization, API, configuration, and monitoring options

Changed

• Enhanced Quick Serve menu shows last configuration and saved shortcuts
• Model field excluded from profiles for model-agnostic templates
• Model cache refresh properly respects TTL settings (>60s)
• Environment variables available in Custom Configuration menu

Fixed

• Fixed manual cache refresh functionality
• Fixed profile creation inconsistency between menus
• Fixed UI consistency issues with prompt formatting