cactus 1.3.0

cactus: ^1.3.0 copied to clipboard

• Renamed CactusTelemetry to CactusConfig for configuration APIs
• Added CactusConfig.setProKey() to enable NPU acceleration
• Added forceTools parameter to CactusCompletionParams for forcing tool calling output

• Added reset() method to CactusLM and CactusSTT classes for clearing context without unloading models
• Added audioStream parameter support in CactusSTT transcription methods for streaming audio input
• Enhanced transcription methods to support both file path and audio stream inputs

• Move to cactus whisper
• Memory optimizations
• Sync binaries with latest cactus engine

• Synced binaries with the latest Cactus engine
• Added support for vision on LFM2 models

• Synced binaries with the latest Cactus engine
• Added support for tool calling on LFM2 models
• Drop vosk and default to whisper for STT

This release includes support for:

• Language Models: 270M to 1.7B parameters
  • Google Gemma-3 (270M, 1B)
  • Qwen3 (600M, 1.7B)
  • SmolLM2 (360M)
  • LiquidAI LFM2 (350M, 700M, 1.2B)
• Embedding Models:
  • Qwen3-Embedding (600M)
  • Nomic-Embed-Text-v2-MoE