OpenNMT/CTranslate2 issues and pull requests

#1158 - Implement rotary embeddings and support the GPT-J model

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1157 - Positional Encoding RuntimeError

Issue - State: closed - Opened by VarunGumma almost 2 years ago - 1 comment

#1156 - how to convert other GPT-like models ? (or how to write my own converter)

Issue - State: closed - Opened by phineas-pta almost 2 years ago - 3 comments

#1155 - Docker: exec format error on linux/arm64/v8

Issue - State: open - Opened by gcervantes8 almost 2 years ago - 1 comment
Labels: enhancement

#1154 - Run error with ruy on AArch64

Issue - State: closed - Opened by MingyangW almost 2 years ago - 8 comments

#1153 - Support float16 on ARM CPUs with native float16 support

Issue - State: open - Opened by FlippFuzz almost 2 years ago - 3 comments
Labels: enhancement, cpu

#1152 - Does Dockerfile docker pull ghcr.io/opennmt/ctranslate2:latest-ubuntu20.04-cuda11.2 include cuDNN?

Issue - State: closed - Opened by mayeaux almost 2 years ago - 1 comment

#1151 - Unable to convert fairseq model

Issue - State: closed - Opened by bil-ash almost 2 years ago - 3 comments

#1150 - Fix synchronization issue when the model input is a CUDA storage

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1149 - Apply new OpenNMT-py option decoder_start_token

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1148 - Enable the Ruy backend in Python wheels for macOS ARM64

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1147 - Select the correct device when copying a StorageView instance

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1146 - Expose properties device and device_index for Whisper model instance

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1145 - Add generation option include_prompt_in_result

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1144 - Add method Whisper.encode to cache and reuse the encoded features

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1143 - Remove unnecessary wrapper for the StorageView Python bindings

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1142 - Use pybind11::keep_alive to tie the array lifetime with the storage

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1141 - Build a single instance of CudaAsyncAllocator

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1140 - Ideas for better performance

Issue - State: open - Opened by hobodrifterdavid almost 2 years ago - 12 comments

#1139 - issue converting user fine-tuned whisper models

Issue - State: closed - Opened by laphang almost 2 years ago - 2 comments

#1138 - Is there an example how to do CTranslate2 Whisper inference in C++ ?

Issue - State: closed - Opened by sv2 almost 2 years ago - 1 comment

#1137 - How to verify if my local build has openMP support

Issue - State: closed - Opened by maxcribe almost 2 years ago - 30 comments

#1136 - Fix condition in DTW backtrace

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1135 - Fix parsing of escaped characters \x.. in the Marian vocabulary file

Pull Request - State: closed - Opened by guillaumekln almost 2 years ago

#1134 - Marian converter: YAML parsing fails when NEL (\x85) present in vocab (spurious line break)

Issue - State: closed - Opened by TommiNieminen almost 2 years ago - 1 comment
Labels: bug

#1127 - Faster llama

Issue - State: open - Opened by maiquangtuan almost 2 years ago - 35 comments
Labels: enhancement

#1122 - Compilation error "‘CPU_FEATURES_VENDOR_GENUINE_INTEL’ was not declared in this scope"

Issue - State: closed - Opened by Jeronymous almost 2 years ago - 2 comments

#1119 - Does batch inference introduce performance gains?

Issue - State: closed - Opened by ItakeLs almost 2 years ago - 3 comments
Labels: question

#1113 - Get whisper's model result contain token score

Issue - State: closed - Opened by shumeirao almost 2 years ago - 1 comment

#1104 - Int4 Support

Issue - State: open - Opened by fmac2000 almost 2 years ago - 6 comments
Labels: enhancement

#1077 - Feature request: Word level scores

Issue - State: open - Opened by ItakeLs about 2 years ago - 11 comments
Labels: enhancement

#1075 - Will this work with stable-ts?

Issue - State: closed - Opened by bakermanbrian about 2 years ago - 5 comments

#1074 - Strange output of Flan-T5-Large when using float16 converted model

Issue - State: open - Opened by Matthieu-Tinycoaching about 2 years ago - 25 comments
Labels: enhancement

#1072 - Feature request: AMD GPU support with oneDNN AMD support

Issue - State: open - Opened by santhoshtr about 2 years ago - 51 comments
Labels: enhancement, help wanted

#1064 - Use of Flan-t5 for answer generation

Issue - State: closed - Opened by Matthieu-Tinycoaching about 2 years ago - 5 comments

#1052 - Splitting LLM layers across multiple GPUs

Issue - State: closed - Opened by JOHW85 about 2 years ago - 6 comments
Labels: enhancement

#1024 - Score: How to get translation `score`

Issue - State: closed - Opened by tahirahmad2030 about 2 years ago - 3 comments
Labels: question

#1008 - Support encoder-only models like BERT

Issue - State: closed - Opened by guillaumekln about 2 years ago - 9 comments
Labels: enhancement

#932 - TypeError: cannot pickle 'ctranslate2.translator.Translator' object

Issue - State: closed - Opened by Matthieu-Tinycoaching over 2 years ago - 5 comments

#896 - Implement end_sequence for text generation

Issue - State: closed - Opened by juliensalinas over 2 years ago - 4 comments
Labels: enhancement

#835 - support different pre_norm and layernorm_embedding in TransformerSpec for EncoderDecoderModel huggingface

Issue - State: closed - Opened by nlpcat over 2 years ago - 4 comments
Labels: enhancement

#832 - Transformers Marian produces translation that ends in gibberish for multilingual models

Issue - State: closed - Opened by raphaelmerx over 2 years ago - 5 comments
Labels: bug

#796 - Support for RNN based decoder units

Issue - State: open - Opened by harishankar-gopalan almost 3 years ago - 9 comments
Labels: enhancement

#693 - Can batch translation on CPU result in different output?

Issue - State: open - Opened by robertBrnnn about 3 years ago - 9 comments
Labels: question, cpu

#578 - Support speech-to-text Transformer model

Issue - State: closed - Opened by gfiameni over 3 years ago - 6 comments
Labels: enhancement

#559 - Can Tensor2Tensor Export to Ctranslate2 Format?

Issue - State: closed - Opened by scotfang over 3 years ago - 5 comments
Labels: enhancement, help wanted

#455 - I am trying to cross compile ctranslate2 with openblas but I am getting this error -> libctranslate2.so: undefined reference to `cblas_sgemm'

Issue - State: closed - Opened by WilliamVJacob almost 4 years ago - 3 comments

#375 - Vectorize GEMM output dequantization and fuse bias + ReLU

Issue - State: closed - Opened by guillaumekln about 4 years ago
Labels: enhancement, cpu

#355 - Support ARM 64-bit architecture

Pull Request - State: closed - Opened by keichi about 4 years ago - 24 comments

#307 - Update README for next version PyPI

Pull Request - State: closed - Opened by guillaumekln over 4 years ago

#182 - Experimental integration of MKL's packed GEMM

Pull Request - State: closed - Opened by guillaumekln almost 5 years ago - 1 comment

#178 - Implement GPU TopK without TensorRT

Issue - State: closed - Opened by guillaumekln almost 5 years ago - 8 comments
Labels: enhancement, help wanted, gpu

#5 - Support ONNX graphs

Issue - State: closed - Opened by guillaumekln over 5 years ago - 10 comments
Labels: enhancement

GitHub / OpenNMT/CTranslate2 issues and pull requests