OpenNMT/CTranslate2 issues and pull requests

#2105 - Request: provide Android wheels for the Python package (`ctranslate2`)

Issue - State: open - Opened by lastcodingbeast 7 days ago

#2104 - fix(cuda): guard exhausted candidates in TopK stage 2

Pull Request - State: open - Opened by mzggr0914 10 days ago - 3 comments

#2103 - Add UMT5 support (per-layer relative attention bias)

Pull Request - State: open - Opened by the-cross-art 12 days ago - 3 comments

#2102 - UMT5 support: per-layer relative attention bias without regressing T5/mT5

Issue - State: open - Opened by the-cross-art 14 days ago

#2101 - ROCm/HIP Windows wheel: Translator destruction hangs indefinitely, even in CPU-only mode with no AMD GPU present

Issue - State: open - Opened by Nachoco000 24 days ago

#2099 - Add Windows ARM64 build and CI support

Pull Request - State: open - Opened by Harishmcw 26 days ago

#2098 - Register nvidia-cublas-cu12/nvidia-cudnn-cu12 DLL dirs on Windows import

Pull Request - State: open - Opened by si-kui-a 27 days ago

#2097 - Guard dim_t overflow in compute_size, reshape and grow

Pull Request - State: open - Opened by kyo-zzz about 1 month ago - 1 comment

#2096 - remove runtime dependency on setuptools

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

#2095 - Version 4.8.2 with CHANGELOG

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

#2094 - Check for overflow in StorageView::reserve before allocation

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

#2093 - Chutki

Issue - State: open - Opened by kumarkunal18128-debug about 1 month ago - 1 comment

#2092 - Check for overflow in StorageView::reserve byte size

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

#2091 - Validate model variable sizes before allocation

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

#2090 - HIP: hipMallocAsync silently corrupts buffers on gfx1030 (RDNA2) although the device reports memory-pool support; default allocator should be cub_caching on HIP

Issue - State: open - Opened by Krzysztofdaw about 1 month ago - 2 comments

#2089 - Removed by author

Issue - State: closed - Opened by mzggr0914 about 1 month ago

#2088 - Host memory grows ~6.8 KB per inference on aarch64 + CUDA (not on x86_64, not on CPU)

Issue - State: open - Opened by paruparu about 1 month ago - 2 comments

#2087 - Fix decoder prefix preprocessing in alternatives mode

Pull Request - State: open - Opened by akira-in-tech about 2 months ago

#2086 - Add Windows ARM64 wheels

Pull Request - State: closed - Opened by ssss141414 about 2 months ago

#2085 - Windows 11 process-exit hang after ROCm Whisper inference with AMD GPU

Issue - State: open - Opened by Sethx86 about 2 months ago

#2084 - Update cxxopts to v3.3.1 (GCC-15 compatibility)

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

#2083 - Make flash_attention effective for the Whisper encoder

Pull Request - State: open - Opened by nilsreck about 2 months ago

#2082 - Source builds need workarounds for sdist, CUDA arches, non-x86 CPU dispatch, and setup.py hardcoding

Issue - State: open - Opened by RemiArata about 2 months ago

#2081 - Failure to compile on 32-bit x86 GNU/Linux

Issue - State: open - Opened by robertkirkman 2 months ago - 4 comments

#2080 - fix(python): import converters and specs lazily so inference does not load torch (memory reduction)

Pull Request - State: closed - Opened by Anai-Guo 2 months ago - 1 comment

#2079 - import ctranslate2 eagerly imports torch (via converters + specs), even for inference-only use

Issue - State: open - Opened by vannhph53095 2 months ago - 1 comment

#2078 - import ctranslate2 eagerly imports torch (via converters + specs), even for inference-only use

Issue - State: closed - Opened by vannhph53095 2 months ago

#2077 - Add native Apple Metal/MPS backend

Pull Request - State: open - Opened by TBO22 3 months ago - 14 comments

#2076 - Fix Windows shutdown deadlock and leak with the Ruy backend

Pull Request - State: closed - Opened by timo9378 3 months ago - 6 comments

#2075 - Allow Whisper decoding to use the full decoder context

Pull Request - State: open - Opened by Emre-Akgul 3 months ago - 20 comments

#2074 - Allow no-timestamp Whisper decoding beyond 224 tokens

Issue - State: open - Opened by Emre-Akgul 3 months ago

#2073 - Check lower ranks in storage_view

Pull Request - State: closed - Opened by jordimas 3 months ago

#2072 - Error on using converted model

Issue - State: open - Opened by WhoMI7 3 months ago - 5 comments

#2071 - Bump version to 4.8.1

Pull Request - State: closed - Opened by jordimas 3 months ago

#2070 - Version 4.8.1

Pull Request - State: closed - Opened by jordimas 3 months ago

#2069 - CI/CD compilation fixes for Windows

Pull Request - State: closed - Opened by jordimas 3 months ago

#2068 - Fix model load heap overflow

Pull Request - State: closed - Opened by jordimas 3 months ago

#2067 - Migrate CUDA build to first-class CMake CUDA language

Pull Request - State: open - Opened by drgaborlukacs 3 months ago

#2066 - Optimize attention softmax buffer reuse

Pull Request - State: closed - Opened by jordimas 4 months ago

#2065 - Fix process-killing integer division by zero when Whisper align() gets a window with no frames

Pull Request - State: closed - Opened by MAkcanca 4 months ago - 1 comment

#2064 - 4.8.0 regression: WhisperModel.align() fails with parallel_for failed: cudaErrorInvalidDevice on Tesla T4 (works on 4.7.0)

Issue - State: open - Opened by christopher5106 4 months ago - 1 comment

#2063 - macOS Python 3.14 CPU backend: intra_threads=0 aborts on 4.7.2 and oversubscribes on 4.8.0

Issue - State: open - Opened by jmpnop 4 months ago - 1 comment

#2060 - Support for gemma4 12b dense model

Pull Request - State: closed - Opened by jordimas 4 months ago

#2057 - Gaza

Issue - State: open - Opened by littralementEngage-hue 4 months ago

#2056 - Upgrade CI system dependencies and reduce build time

Pull Request - State: closed - Opened by jordimas 4 months ago

#2055 - Enables PACKED_GEMM by default for Intel MKL to improve Intel CPU performance

Pull Request - State: closed - Opened by jordimas 4 months ago

#2053 - Adopt latest BS_thread_pool libary version v5.1.0

Pull Request - State: closed - Opened by 3manifold 5 months ago - 4 comments

#2052 - [Feature] Adopt latest BS_thread_pool

Issue - State: closed - Opened by 3manifold 5 months ago

#2051 - Request for admin access to the prject

Issue - State: closed - Opened by jordimas 5 months ago

#2048 - Gemma4 support for dense model

Pull Request - State: closed - Opened by jordimas 5 months ago

#2036 - Harden legacy converter checkpoint loading

Pull Request - State: closed - Opened by igorsantosgrueiro 5 months ago - 1 comment

#2034 - Private security contact request

Issue - State: closed - Opened by igorsantosgrueiro 5 months ago - 2 comments

#2032 - Feature request: intel GPU support

Issue - State: open - Opened by DDXDB 6 months ago - 2 comments

#2025 - Return logits beamsearch

Pull Request - State: open - Opened by cmullovisl 7 months ago - 1 comment

#2023 - aarch64 docker images

Pull Request - State: open - Opened by NicolasBFR 7 months ago

#2022 - Make it possible to use a transformer model as encoder.

Pull Request - State: open - Opened by robinp 7 months ago

#2021 - ROCm (gfx1201/RX 9070 XT): faster-whisper crash in CTranslate2 4.7.1

Issue - State: open - Opened by Unrest6585 7 months ago - 5 comments

#2020 - fix issue with return_logits_vocab option

Pull Request - State: open - Opened by cmullovisl 7 months ago - 1 comment

#2019 - docs: add build instructions for NVIDIA Jetson (aarch64 + CUDA)

Pull Request - State: open - Opened by vimalk78 7 months ago - 1 comment

#2018 - Fix index calculation for logits_vocab

Pull Request - State: open - Opened by erip 7 months ago - 1 comment

#2017 - Cannot get correct translation results (NLLB-200-distilled-600M)

Issue - State: open - Opened by MSJavaScript 8 months ago - 1 comment

#2010 - Add return_attention support to Generator

Pull Request - State: open - Opened by QuentinFuxa 8 months ago - 1 comment

#2002 - Improve Primitives a Little Bit

Pull Request - State: closed - Opened by BBC-Esq 8 months ago - 2 comments

#1996 - Fix flaky transformers_wav2vec2 test

Pull Request - State: closed - Opened by 3manifold 8 months ago - 8 comments

#1986 - Integrate Clang ThreadSanitizer in tests

Pull Request - State: open - Opened by 3manifold 9 months ago

#1965 - Update to transfomers an add tests for Gemma 3 and Qwen3

Pull Request - State: closed - Opened by jordimas 9 months ago

#1964 - Fix license warning in installation

Pull Request - State: closed - Opened by jordimas 9 months ago

#1962 - T5Gemma conversion and inference

Pull Request - State: open - Opened by jordimas 9 months ago - 1 comment

#1960 - Enable generalized LayerNorm on CUDA

Pull Request - State: closed - Opened by BBC-Esq 10 months ago - 16 comments

#1950 - Making it easier for users to get started with CTranslate2 models

Issue - State: open - Opened by jordimas 10 months ago - 27 comments

#1949 - Conv1d pure CUDA implementation

Pull Request - State: closed - Opened by jordimas 10 months ago - 12 comments

#1940 - Update Docker to Cuda 12.8

Pull Request - State: open - Opened by Purfview 11 months ago - 4 comments

#1939 - Update to CUDA 12.8

Pull Request - State: open - Opened by Purfview 11 months ago - 8 comments

#1935 - Switch to Ubuntu ARM also run tests in mac

Pull Request - State: closed - Opened by jordimas 11 months ago

#1933 - Feature request: CUDA 13 support

Issue - State: open - Opened by AmaliaKalio 11 months ago - 10 comments

#1918 - The result for T5 is not the same

Issue - State: open - Opened by AndreyM0 about 1 year ago

#1916 - Hi, I'm unsure how to implement the ops::Div division operator

Issue - State: open - Opened by a2d8a4v about 1 year ago

#1914 - Feat:[riscv] Add rvv support to cpu/kernels.cc

Pull Request - State: open - Opened by little-creator about 1 year ago

#1909 - The pkg_resources package is slated for removal as early as 2025-11-30

Issue - State: open - Opened by surfincanoy about 1 year ago

#1907 - Fix CMake compatibility warnings

Pull Request - State: open - Opened by 3manifold about 1 year ago - 3 comments

#1905 - Fix incompatibility among manylinux2014 and rhel8

Pull Request - State: open - Opened by 3manifold about 1 year ago

#1904 - fix instructions for fairseq

Pull Request - State: open - Opened by frederik-uni about 1 year ago

#1900 - add build to release

Issue - State: open - Opened by frederik-uni over 1 year ago

#1899 - Compatibility with CMake < 3.5 has been removed from CMake.

Issue - State: open - Opened by JustFrederik over 1 year ago - 3 comments

#1896 - Performance comparison with TensorRT-LLM

Issue - State: open - Opened by Francis235 over 1 year ago

#1894 - pkg_resources needs to be replaced

Issue - State: open - Opened by BBC-Esq over 1 year ago

#1892 - how to build, so that auditwheel can bundle dependencies

Issue - State: open - Opened by Majdoddin over 1 year ago - 1 comment

#1891 - Add noexecstack linker flags

Pull Request - State: closed - Opened by teabuff over 1 year ago

#1890 - I can not get expected result following the quick start demo

Issue - State: open - Opened by nanfei01055 over 1 year ago

#1889 - Add biasing

Pull Request - State: open - Opened by anthonyrathe over 1 year ago

#1889 - Add support for shallow biasing of Whisper

Pull Request - State: open - Opened by anthonyrathe over 1 year ago - 3 comments

#1888 - AWQ is being deprecated

Issue - State: open - Opened by BBC-Esq over 1 year ago

#1887 - Development issues with `fused_layers`

Issue - State: open - Opened by Eyoel-gebre over 1 year ago

#1886 - Add support for lite-whisper

Pull Request - State: open - Opened by Eyoel-gebre over 1 year ago - 2 comments