Issues

• Request: provide Android wheels for the Python package (ctranslate2)

  OpenNMT/CTranslate2#2105 · Filter by author lastcodingbeast opened on Sep 29, 2026

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

  OpenNMT/CTranslate2#2102 · Filter by author the-cross-art opened on Sep 22, 2026

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

  OpenNMT/CTranslate2#2101 · Filter by author Nachoco000 opened on Sep 12, 2026

• Significant Korean accuracy degradation with Whisper inputs shorter than 30 seconds

  OpenNMT/CTranslate2#2100 · Filter by author feel012-web opened on Sep 11, 2026

• Chutki

  OpenNMT/CTranslate2#2093 · Filter by author kumarkunal18128-debug opened on Aug 30, 2026
  1

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

  OpenNMT/CTranslate2#2090 · Filter by author Krzysztofdaw opened on Aug 29, 2026
  2

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

  OpenNMT/CTranslate2#2088 · Filter by author paruparu opened on Aug 25, 2026
  2

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

  OpenNMT/CTranslate2#2085 · Filter by author Sethx86 opened on Aug 17, 2026

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

  OpenNMT/CTranslate2#2082 · Filter by author RemiArata opened on Aug 11, 2026

• Failure to compile on 32-bit x86 GNU/Linux

  OpenNMT/CTranslate2#2081 · Filter by author robertkirkman opened on Jul 31, 2026
  4

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

  OpenNMT/CTranslate2#2079 · Filter by author vannhph53095 opened on Jul 29, 2026
  1

• Allow no-timestamp Whisper decoding beyond 224 tokens

  OpenNMT/CTranslate2#2074 · Filter by author Emre-Akgul opened on Jul 11, 2026

• Error on using converted model

  OpenNMT/CTranslate2#2072 · Filter by author WhoMI7 opened on Jul 5, 2026
  5

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

  OpenNMT/CTranslate2#2064 · Filter by author christopher5106 opened on Jun 11, 2026
  1

• macOS Python 3.14 CPU backend: intra_threads=0 aborts on 4.7.2 and oversubscribes on 4.8.0

  OpenNMT/CTranslate2#2063 · Filter by author jmpnop opened on Jun 7, 2026
  1

• Expose per-step decoder cross-attention scores during generate() (for streaming / AlignAtt policies)

  OpenNMT/CTranslate2#2059 · Filter by author Ed1ks opened on Jun 3, 2026

• Add runtime LoRA adapter swapping

  OpenNMT/CTranslate2#2045 · Filter by author wonsup-shin opened on May 14, 2026

• del model deadlocks on ROCm 7.2.1 + gfx1100 (Windows) — HIP allocator free path

  OpenNMT/CTranslate2#2038 · Filter by author trasurc opened on May 8, 2026

• bfloat16 GEMM is noticeably slower than float16

  OpenNMT/CTranslate2#2033 · Filter by author BBC-Esq opened on Apr 29, 2026

• Feature request: intel GPU support

  OpenNMT/CTranslate2#2032 · Filter by author DDXDB opened on Apr 14, 2026
  2

• Is it possible to convert CTranslate2 weights back to Hugging Face Transformers?

  OpenNMT/CTranslate2#2031 · Filter by author tg-bomze opened on Apr 12, 2026

• 📋 Documentation Enhancement Suggestion

  OpenNMT/CTranslate2#2026 · Filter by author croviatrust opened on Mar 13, 2026

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

  OpenNMT/CTranslate2#2021 · Filter by author Unrest6585 opened on Feb 28, 2026
  5

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

  OpenNMT/CTranslate2#2017 · Filter by author MSJavaScript opened on Feb 19, 2026
  1

• AMD wheels for Windows built against ROCm 7.2, but Windows HIP SDK installer only goes to 7.1.1

  OpenNMT/CTranslate2#2016 · Filter by author PinW opened on Feb 14, 2026
  3