Google GitGoogle Git
Sign in
chromium/external/github.com/google-ai-edge/LiteRT-LM/HEAD
  1. 4a363c0 New session configs based on protobuf by Byungchul Kim · 20 hours ago upstream/main
  2. 798d51e Support FLUX.2-klein engine/session by Byungchul Kim · 21 hours ago
  3. 2d872e0 Add Android benchmarks for Gemma 4 2B LoRA on CPU and GPU. by Salil Tambe · 27 hours ago
  4. 0063187 Allow creating embedding engines for models without a vision encoder. by Google AI Edge · 27 hours ago
  5. b2f3d3c Add speculative decoding statistics to the MTP drafter, the executor and the by Mohammadreza Heydary · 28 hours ago
  6. 8b7bb5c When the thought channel is not prefilled and the model skips thinking, by Google AI Edge · 28 hours ago
  7. 4697c72 - Write all-ones into `kPrevMaskName` in `input_buffers_map_` and clear `buffered_spectrogram_` inside `AudioLiteRtCompiledModelExecutor::AudioStreamingEncoder::CreateNewContext` so new streaming contexts start cleanly. by Google AI Edge · 31 hours ago
  8. e16809a Handle partial chunks correctly in AudioInputSource and TextInputSource. by Byungchul Kim · 32 hours ago
  9. ea11963 Document in the C API that sessions and conversations must be destroyed before the engine they were created from. by Tenghui Zhu · 34 hours ago
  10. 74e702d Text2Image engine and session by Byungchul Kim · 34 hours ago
  11. d07a923 Fix Gemma-4 GPU LoRA support for quantized models in MLDrift delegate. by Salil Tambe · 2 days ago
  12. 4b15b34 Report LoRA inputs that don't match any tensor in the LoRA data. by Google AI Edge · 2 days ago
  13. e9fb7ff Open-source the EmbeddingGemma 2 multimodal search web demo by Google AI Edge · 2 days ago
  14. 97e6dc7 [LiteRT-LM] Eagerly filter channel content from KV cache after decode. by Google AI Edge · 2 days ago
  15. b8b3131 Fix text scoring rewinding the KV-cache position twice after prefill. by Google AI Edge · 2 days ago
  16. 7840057 Isolate espeak_ng genrule working directories to fix parallel build race in standalone Bazel builds by Salil Tambe · 2 days ago
  17. 7233a2d Create LoRA buffers per signature so prefill gets buffer types it supports by Google AI Edge · 2 days ago
  18. 3a1f60d Rename Omni ASR/TTS session targets and standardize stage/factory signatures. by Byungchul Kim · 2 days ago
  19. d4beba4 Fix multibyte UTF-8 handling and byte-fallback detokenization in BufferedStreamingDetokenizer. by Mohammadreza Heydary · 2 days ago
  20. 1f7fa76 Add FLUX.2 model config, math helpers, denoiser stage, and VAE decoder stage. by Byungchul Kim · 2 days ago
  21. cb1495f update benchmark script by Google AI Edge · 2 days ago
  22. 4ecac4f Add DMA-BUF memory metrics to LiteRT LM metrics proto. by Andrew Zhang · 3 days ago
  23. 4c72b51 Unify ASR and TTS session orchestration under MultiStagedSession. by Byungchul Kim · 3 days ago
  24. f0f235e Add embedding demo link to litert lm readme by Google AI Edge · 3 days ago
  25. a926bb0 Fill lora_param_tensor so GPU runtime-BMM LoRA covers all channels by Salil Tambe · 3 days ago
  26. 5b87f5b Add TextEncoderStage under omni/text2image. by Byungchul Kim · 3 days ago
  27. 93d7412 Bump LiteRT-LM version to 0.19.0 by Google AI Edge · 3 days ago
  28. deaaa64 Re-prime the MTP drafter after any non-speculative step. by Mohammadreza Heydary · 3 days ago
  29. 8eca57a Add Conversation::CountTokens() API to measure prospective message token footprint. by Google AI Edge · 3 days ago
  30. c9848c4 add v0.18.0 release notes by Google AI Edge · 3 days ago
  31. 0bb486b Load Gemma 4 audio encoder LoRAs under the encoder's own input names by Salil Tambe · 3 days ago
  32. 84d9669 Pipe through max_top_k_ setting when set by Google AI Edge · 4 days ago
  33. dbdb99b Internal change by Google AI Edge · 4 days ago
  34. 525de7e Add an --output_size flag to embedding_litert_lm_main for truncation of the output embedding. by Marissa Ikonomidis · 4 days ago
  35. 0540e5c Replace `# pytype: disable` suppressions with `# pyrefly: ignore` by Oleh Prypin · 5 days ago
  36. d17a52f Add ImageDecoder and PromptSource interface for image gen model inference by Byungchul Kim · 7 days ago
  37. 10245c8 Internal change by Wai Hon Law · 7 days ago
  38. 8092fe5 Extend the Kotlin `benchmark` API and JNI bindings (`nativeCreateBenchmark`) to accept optional `visionBackend`, `audioBackend`, and `contents` parameters, and expose `BenchmarkInfo.markDurationsInSecond` from `BenchmarkInfo::GetMarkDurations()`. by Salil Tambe · 7 days ago
  39. 5befccd Restrict espeak-ng dictionaries to Kokoro-supported languages. by Tenghui Zhu · 7 days ago
  40. f136f4a Handle silence in more robust way by Byungchul Kim · 7 days ago
  41. 400273c Add pre-compiled model support to LiteRT-LM CompiledModelExecutor by Google AI Edge · 7 days ago
  42. b320801 Internal change by Tenghui Zhu · 7 days ago
  43. 956de4b Strengthen input validation and null pointer checks in LiteRT LM C APIs. by Mohammadreza Heydary · 7 days ago
  44. 58fcc1c Preserve prefill start token and truncate repetition loops in LmDecoder. by Byungchul Kim · 7 days ago
  45. b34761f Stitch overlapping chunks at longest contiguous match in LevenshteinTextMerger. by Byungchul Kim · 7 days ago
  46. c8ff070 Retain unconfirmed words on empty chunks in TimestampTextMerger. by Byungchul Kim · 7 days ago
  47. 3dbb23e Skip auxiliary mask and cache update warmup in LiteRT-LM NPU compiled model executor when using hardware update (kWH). by Andrew Zhang · 8 days ago
  48. 1a14a11 Remove litert_lm_get_last_error_code from the LiteRT-LM C API. by Mohammadreza Heydary · 8 days ago
  49. 5775c07 Zero-pad wrapped ring buffer frames at EOF in FileAudioSource. by Byungchul Kim · 8 days ago
  50. 4af508a Document the LiteRT-LM C API 1.0.0 return-shape contract by Mohammadreza Heydary · 8 days ago
  51. fc6ca8a Return status codes from the experimental LiteRT-LM C API by Mohammadreza Heydary · 8 days ago
  52. a5d53ea Return status codes from all LiteRT-LM model info C API functions.\ by Mohammadreza Heydary · 8 days ago
  53. 216b00b Add ASR, TTS, and ImageGen in litert_lm_builder. by Byungchul Kim · 8 days ago
  54. 5190782 Return status codes from all LiteRT-LM embedding engine C API functions. by Mohammadreza Heydary · 8 days ago
  55. b4a9d0e Return status codes from LiteRT-LM C API conversation functions by Mohammadreza Heydary · 8 days ago
  56. f4455da Sanitize tool and property names for Lark grammar rules in constrained decoding. by Google AI Edge · 8 days ago
  57. 8ab3da0 Return status codes from the remaining LiteRT-LM C engine.h accessors by Mohammadreza Heydary · 8 days ago
  58. 7d9a2a3 Return status codes from LiteRT-LM C API result-producing functions by Mohammadreza Heydary · 8 days ago
  59. 1dc4633 Return status codes from LiteRT-LM C API engine constructors by Mohammadreza Heydary · 8 days ago
  60. d101ec9 Make the dummy test model's output shape match its declared shape. by Tommy Chiang · 8 days ago
  61. 9369b17 Fix support for Gemma3 270M in LiteRT LM runtime and builder. by Andrew Zhang · 9 days ago
  62. 12ba057 Also skip the host local attention mask fill when both masks are pruned by the GPU delegate. by Fengwu Yao · 9 days ago
  63. b102de2 This is an internal change by Google AI Edge · 9 days ago
  64. 526d965 Add DMA-BUF memory tracking utility and init time measurement for LiteRT-LM runners. by Andrew Zhang · 9 days ago
  65. 56c420b Add ImageGenMetadata in .litertlm by Byungchul Kim · 9 days ago
  66. 8fded1b Move TFLite mutable schema generation from patch step to build-time CMake target. by Mohammadreza Heydary · 9 days ago
  67. 96a6009 feat: add enableYnnpack experimental flag in Swift API by Wai Hon Law · 9 days ago
  68. 31c1745 Return canonical status codes instead of -1 from status-returning LiteRT-LM C API functions. by Mohammadreza Heydary · 9 days ago
  69. 72a3df7 Bump the LiteRT-LM C API version to 1.0.0 and document the status-return rule. by Mohammadreza Heydary · 9 days ago
  70. a21dcbe test(cli): add multimodal e2e tests and documentation for openai embeddings by Wai Hon Law · 9 days ago
  71. 78d7bd2 Add DMA-BUF and peak memory reporting to LiteRT tools and runner script for embedding models. by Andrew Zhang · 9 days ago
  72. 3bc51f1 Check LiteRT-LM C setter status codes in the Python, Swift and Go bindings by Mohammadreza Heydary · 9 days ago
  73. 34d1def Generate mutable TFLite schema header in litert_patcher.cmake. by Mohammadreza Heydary · 9 days ago
  74. 3431993 Return status codes from void setters in the LiteRT-LM C embedding API. by Mohammadreza Heydary · 9 days ago
  75. 0ea540e feat: add enableYnnpack experimental flag in Kotlin API by Wai Hon Law · 9 days ago
  76. a0331da Return status codes from void setters in the LiteRT-LM C conversation API. by Mohammadreza Heydary · 9 days ago
  77. b1bbf74 Return status codes from void setters in the LiteRT-LM C engine API. by Mohammadreza Heydary · 9 days ago
  78. a458481 Record the last error on every silent failure path in the LiteRT-LM C API by Mohammadreza Heydary · 9 days ago
  79. bfdec21 Support ASR models in .litertlm by Byungchul Kim · 9 days ago
  80. 122ca32 Consolidate omni IO types by Byungchul Kim · 10 days ago
  81. 6b8f5b5 Internal change by Mohammadreza Heydary · 10 days ago
  82. 26fdbaf Add benchmark image and audio inputs to LiteRT-LM runtime testdata. by Google AI Edge · 10 days ago
  83. f07055f Internal change for eval by Google AI Edge · 10 days ago
  84. 86a7e6d Add image item formatting (`<image>`) to the LFM2 Jinja prompt template and LlmMetadataProto.pbtext. by Salil Tambe · 10 days ago
  85. 34d7bde Add status-return helpers for the LiteRT-LM C API. by Mohammadreza Heydary · 10 days ago
  86. 3c04cd3 fix(cli): delete automatic model conversion from run command by Wai Hon Law · 10 days ago
  87. 615d193 Retain linear attention state buffers as inputs for GPU-optimized in-place mode. by Fengwu Yao · 10 days ago
  88. 8695dcf workspace config update. by Mohammadreza Heydary · 10 days ago
  89. fd68028 Skip CPU global causal attention mask initialization and fill when pruned by the GPU delegate. by Fengwu Yao · 10 days ago
  90. 556f824 - Ensure vision encoder input and output buffers are lazily allocated before execution in VisionLiteRtCompiledModelExecutor. by Google AI Edge · 10 days ago
  91. 3cbf770 - Ensure vision encoder input and output buffers are lazily allocated before execution in VisionLiteRtCompiledModelExecutor. by Salil Tambe · 10 days ago
  92. f76896c Automated Code Change by Amer Elsheikh · 10 days ago
  93. 46de5ea Add LiteRtLogitMaskRunner for accelerated logit masking. by Mohammadreza Heydary · 11 days ago
  94. 403fbee feat(model): add LFM2 multimodal metadata proto configuration by Salil Tambe · 11 days ago
  95. dc8facf Clean up LiteRT CMake shim and update SentencePiece build configuration. by Google AI Edge · 11 days ago
  96. 3ced337 Adds ImagePreprocessor::Preprocess(DecodedImage, ...) overload to accept pre-decoded pixel buffers directly. by Google AI Edge · 12 days ago
  97. 5e3bd63 Grow the NPU dynamic KV cache on demand during prefill and decode. by Matt Kreileder · 13 days ago
  98. 2f8284d Update dependencies of litert_lm by Byungchul Kim · 14 days ago
  99. 2fa7506 This is an internal change by Google AI Edge · 14 days ago
  100. aa2401b Fix Safari crash by splitting JSPI and Asyncify builds in LiteRT-LM Web by Google AI Edge · 14 days ago