Rare find

LiteRT-LM. Run AI models directly on your phone, laptop, or Raspberry Pi without the cloud.

ai.google.dev/edge/litert-lm
by @google-ai-edge

Vaya's read on this project

Problem, audience, market, and the verdict — sign in to see it.

Updates

July 2026
  • No public description
  • Add multimodal support to Apple FM adapter merging from the external …
  • feat(cli): add $comment version metadata to config schema fields
  • Add support for different attention mask policies.
  • feat(cli): make --ringbuffers-local-attention a hidden CLI flag
  • Log executor mark durations in ms for LiteRT-LM.
  • docs: add version comments to public C API in engine.h
  • feat(cli): add support for --thinking and --thinking-budget in config…
  • Introduce StateInterface for KV cache management.
  • feat(cli): add --config option to use custom configuration file
  • No public description
  • feat(cli): make --gpu-decode-steps-per-sync available in run command …
  • Import PR #2688: Support Data processor for vision encoder.
  • feat(cli): add --chat-template option to pack and unpack commands
  • Rename KV cache to state.
  • Add proto definitions for LLM executor metadata.
  • Initialize litert_lm_cli Kokoro macOS pipeline
  • No public description
  • Fix naming regression for weight cache
  • Expose GPU performance-sensitive flags to LiteRT-LM's PyThon layer an…
  • Release —v0.15.0-alpha0
  • Release —v0.14.0
June 2026
  • Release —v0.14.0-alpha.0
  • Release —v0.13.1
  • Release —v0.13.0
May 2026
  • Release —v0.12.0
  • Release —v0.11.0
April 2026
  • Release —v0.11.0-rc.1
  • Release —v0.10.2
  • Release —v0.10.1