robertknight/rten issues and pull requests

#411 - Add fast path for ArgMin / ArgMax when axis is contiguous

Pull Request - State: closed - Opened by robertknight about 8 hours ago

#410 - Optimize GatherND by avoiding redundant zeroing of output and adding fast path for contiguous inputs

Pull Request - State: closed - Opened by robertknight about 17 hours ago

#410 - Optimize GatherND by avoiding redundant zeroing of output and adding fast path for contiguous inputs

Pull Request - State: closed - Opened by robertknight about 17 hours ago

#409 - Optimize copying of non-contiguous tensors with 5+ dimensions

Pull Request - State: closed - Opened by robertknight about 18 hours ago

#409 - Optimize copying of non-contiguous tensors with 5+ dimensions

Pull Request - State: closed - Opened by robertknight about 18 hours ago

#408 - Reserve KV cache capacity after the first model run

Pull Request - State: closed - Opened by robertknight 1 day ago

#408 - Reserve KV cache capacity after the first model run

Pull Request - State: closed - Opened by robertknight 1 day ago

#407 - Support passing captures by-value to subgraphs

Pull Request - State: closed - Opened by robertknight 2 days ago

#406 - Prefer `Tensor::from` for creating vectors from array literals

Pull Request - State: closed - Opened by robertknight 3 days ago

#406 - Prefer `Tensor::from` for creating vectors from array literals

Pull Request - State: closed - Opened by robertknight 3 days ago

#406 - Prefer `Tensor::from` for creating vectors from array literals

Pull Request - State: closed - Opened by robertknight 3 days ago

#405 - Re-order execution plan to enable more operations to run in-place

Pull Request - State: closed - Opened by robertknight 4 days ago

#405 - Re-order execution plan to enable more operations to run in-place

Pull Request - State: closed - Opened by robertknight 4 days ago

#404 - Streamline NodeId -> usize conversions

Pull Request - State: closed - Opened by robertknight 5 days ago

#404 - Streamline NodeId -> usize conversions

Pull Request - State: closed - Opened by robertknight 5 days ago

#404 - Streamline NodeId -> usize conversions

Pull Request - State: closed - Opened by robertknight 5 days ago

#403 - Output transcription speed as a multiple of real-time in Whisper example

Pull Request - State: closed - Opened by robertknight 7 days ago

#403 - Output transcription speed as a multiple of real-time in Whisper example

Pull Request - State: closed - Opened by robertknight 7 days ago

#403 - Output transcription speed as a multiple of real-time in Whisper example

Pull Request - State: closed - Opened by robertknight 7 days ago

#402 - Support (de-)serialization of tensors using serde

Pull Request - State: closed - Opened by robertknight 7 days ago

#402 - Support (de-)serialization of tensors using serde

Pull Request - State: closed - Opened by robertknight 7 days ago

#402 - Support (de-)serialization of tensors using serde

Pull Request - State: closed - Opened by robertknight 7 days ago

#401 - Reuse existing output node when replacing operator with a fusion

Pull Request - State: closed - Opened by robertknight 7 days ago

#401 - Reuse existing output node when replacing operator with a fusion

Pull Request - State: closed - Opened by robertknight 7 days ago

#401 - Reuse existing output node when replacing operator with a fusion

Pull Request - State: closed - Opened by robertknight 7 days ago

#400 - Normalize and chunk audio inputs in wav2vec2 example

Pull Request - State: closed - Opened by robertknight 9 days ago

#400 - Normalize and chunk audio inputs in wav2vec2 example

Pull Request - State: closed - Opened by robertknight 9 days ago

#400 - Normalize and chunk audio inputs in wav2vec2 example

Pull Request - State: closed - Opened by robertknight 9 days ago

#399 - Support in-place `Concat` when input is a captured value

Issue - State: closed - Opened by robertknight 14 days ago - 3 comments
Labels: performance

#399 - Support in-place `Concat` when input is a captured value

Issue - State: closed - Opened by robertknight 14 days ago - 3 comments
Labels: performance

#399 - Support in-place `Concat` when input is a captured value

Issue - State: closed - Opened by robertknight 14 days ago - 3 comments
Labels: performance

#398 - Support fusing Transpose + MatMul where both inputs are transposed

Issue - State: open - Opened by robertknight 18 days ago
Labels: performance

#397 - Add Whisper speech recognition example

Pull Request - State: closed - Opened by robertknight 20 days ago

#396 - Add context to token generation errors

Pull Request - State: closed - Opened by robertknight 20 days ago

#395 - Support `cache_position` inputs in Hugging Face models

Pull Request - State: closed - Opened by robertknight 20 days ago

#394 - Add `token_id_filter` function to create logits filter from predicate

Pull Request - State: closed - Opened by robertknight 20 days ago

#393 - Add API to rten-generate for filtering/processing logits

Pull Request - State: closed - Opened by robertknight 21 days ago

#392 - Support newer `merges` format in tokenizer.json files

Pull Request - State: closed - Opened by robertknight 24 days ago

#391 - Parsing of `merges` field in tokenizer.json fails

Issue - State: closed - Opened by Tracreed 26 days ago - 1 comment

#390 - Enclose linked identifiers in backticks in API docs

Pull Request - State: closed - Opened by robertknight 29 days ago

#389 - WASM relaxed SIMD support

Issue - State: open - Opened by robertknight 29 days ago - 1 comment

#388 - Optimize Cast op when source and dest types are the same, add u8/i8 cast tests

Pull Request - State: closed - Opened by robertknight 30 days ago

#387 - int8 and uint8 support in Pad and other ops

Pull Request - State: closed - Opened by igor-yusupov about 1 month ago - 2 comments

#386 - Generalize `IntoOpResult` impls

Pull Request - State: closed - Opened by robertknight about 1 month ago

#385 - Use a macro to de-duplicate proxy layout impls

Pull Request - State: closed - Opened by robertknight about 1 month ago

#384 - Replace remaining uses of `check_dims` with `static_dims`

Pull Request - State: closed - Opened by robertknight about 1 month ago

#383 - Generalize `Output` => `Tensor`/`TensorView` convenience methods

Pull Request - State: closed - Opened by robertknight about 1 month ago

#382 - VALID auto_pad value

Issue - State: open - Opened by igor-yusupov about 1 month ago - 8 comments

#381 - Convert `NodeId` from an alias for `usize` to u32-sized opaque type

Pull Request - State: closed - Opened by robertknight about 1 month ago

#380 - Use "dep:" prefix to prevent implicit creation of features

Pull Request - State: closed - Opened by robertknight about 1 month ago

#379 - Support setting `RunOptions` for model inference when using `Generator`

Pull Request - State: closed - Opened by robertknight about 1 month ago

#378 - Remove self-dependency of crates in development

Pull Request - State: closed - Opened by robertknight about 1 month ago

#377 - Improve debug formatting of tensors

Pull Request - State: closed - Opened by robertknight about 1 month ago

#376 - Fix AVX-512 build of im2col.rs

Pull Request - State: closed - Opened by robertknight about 1 month ago

#375 - Use a Python virtualenv in CI

Pull Request - State: closed - Opened by robertknight about 2 months ago

#374 - Update YOLO example instructions to use YOLO v11

Pull Request - State: closed - Opened by robertknight about 2 months ago

#373 - Use `f32::round_ties_even` instead of `libm::rintf`

Pull Request - State: closed - Opened by robertknight about 2 months ago

#372 - Widen float16 constants to float32 in model converter

Pull Request - State: closed - Opened by robertknight about 2 months ago

#371 - Fuse pointwise operations into matmul / convolution operations

Issue - State: open - Opened by robertknight about 2 months ago
Labels: performance

#370 - Implement better depthwise convolution kernels

Issue - State: open - Opened by robertknight about 2 months ago
Labels: performance

#369 - Apply fusions correctly when fused ops are chained

Pull Request - State: closed - Opened by robertknight about 2 months ago

#368 - Add `--no-optimize` flag to rten CLI

Pull Request - State: closed - Opened by robertknight about 2 months ago

#367 - Replace old tensor slicing methods with new API

Pull Request - State: closed - Opened by robertknight about 2 months ago

#366 - Infer layout for `slice_copy` based on layout and range

Pull Request - State: closed - Opened by robertknight about 2 months ago

#365 - Add fallible variants of `TensorBase::slice_with`

Pull Request - State: closed - Opened by robertknight about 2 months ago

#364 - Use `zip` method instead of standalone function

Pull Request - State: closed - Opened by robertknight about 2 months ago

#363 - Support ranges with unknown index count in `slice_with`

Pull Request - State: closed - Opened by robertknight about 2 months ago

#362 - Replace most remaining uses of `slice` with `slice_with`

Pull Request - State: closed - Opened by robertknight about 2 months ago

#361 - Use `TensorBase::slice_with` in RNN operators

Pull Request - State: closed - Opened by robertknight about 2 months ago