All pull requests

Pull requests

• build: update dependencies of litert_lm

  #4584·copybara-service[bot] opened Nov 13, 2026Bot·

  ·

• Default ConversationConfig::enable_rewinding to false on CPU and GPU backends.

  #4583·copybara-service[bot] opened Nov 13, 2026Bot·

  ·

• Support local/recurrent context checkpointing in LlmLiteRtCompiledModelExecutorBase.

  #4582·copybara-service[bot] opened Nov 13, 2026Bot·

  ·

• Upload streamed weights per model instead of through global state.

  #4581·copybara-service[bot] opened Nov 13, 2026Bot·

  ·

• Allow waiting for task and session completion inside terminal callbacks.

  #4580·copybara-service[bot] opened Nov 13, 2026Bot·

  ·

• feat: switch Android dynamic GPU accelerator to ml_drift_gpu_accelerator_so

  #4579·copybara-service[bot] opened Nov 13, 2026Bot·

  ·

• fix: report CLI usage and execution errors cleanly in litert_lm_main

  #4566·copybara-service[bot] opened Nov 12, 2026Bot·

  ·

• fix(cli): validate tools and tool_choice in OpenAI chat completions server

  #4565·copybara-service[bot] opened Nov 12, 2026Bot·

  ·

• feat: expose runTextScoring and session checkpointing in Kotlin SDK

  #4564·copybara-service[bot] opened Nov 12, 2026Bot·

  ·

• fix: honor session SamplerParameters in LlmLiteRtNpuCompiledModelExecutor

  #4563·copybara-service[bot] opened Nov 12, 2026Bot·

  ·

• feat: add litert_lm_conversation_config_set_channels and Python channels support

  #4562·copybara-service[bot] opened Nov 12, 2026Bot·

  ·

• fix: strip unclosed channel start tags from fallback history model prefixes

  #4561·copybara-service[bot] opened Nov 12, 2026Bot·

  ·

• feat(cli): add /clear and /reset commands to reset interactive conversation context

  #4560·copybara-service[bot] opened Nov 12, 2026Bot·

  ·

• fix: synchronize lazy ModelResources caches with absl::Mutex

  #4559·copybara-service[bot] opened Nov 12, 2026Bot·

  ·

• refactor: deduplicate LLM engine initialization and streaming helpers

  #4552·copybara-service[bot] opened Nov 11, 2026Bot·

  ·

• refactor: deduplicate EmbeddingEngineImpl initialization and multimodal helpers

  #4551·copybara-service[bot] opened Nov 11, 2026Bot·

  ·

• test: add NpuEmbedder::CreateForTest helper

  #4546·copybara-service[bot] opened Nov 11, 2026Bot·

  ·

• refactor: deduplicate buffer setup, warmup, and step helpers in NPU executor

  #4543·copybara-service[bot] opened Nov 11, 2026Bot·

  ·

• Support multi-prefill benchmark logging and append-chunk validation.

  #4541·copybara-service[bot] opened Nov 11, 2026Bot·

  ·

• Add multi-prefill Conversation benchmark runner in litert_lm_lib.

  #4540·copybara-service[bot] opened Nov 11, 2026Bot·

  ·

• Add Gemma 4 model config, weight declarations, and checkpoint hooks.

  #4539·copybara-service[bot] opened Nov 11, 2026Bot·

  ·

• Extend models/common with dual local/global RoPE, sliding-window KV ring buffers, and per-layer inputs.

  #4537·copybara-service[bot] opened Nov 11, 2026Bot·

  ·

• perf: defer CPU vision/audio executor compilation and evict after encoding

  #4533·copybara-service[bot] opened Nov 11, 2026Bot·

  ·

• Cancel and drain active session tasks in ~Conversation()

  #4528·copybara-service[bot] opened Nov 11, 2026Bot·

  ·
  1

• perf: add dimension-coalescing Highway SIMD SLICE CPU fallback kernel

  #4526·copybara-service[bot] opened Nov 11, 2026Bot·

  ·