Add mistral small3 24B config and recipe by eagle705 · Pull Request #14784 · NVIDIA-NeMo/NeMo
* Support QwenVL for inference API (#14534)
* Support QwenVL for inference engine
* Apply isort and black reformatting
Signed-off-by: meatybobby <meatybobby@users.noreply.github.com>
* Remove comment out
* Reformat
* Skip pylint check
* Add unit tests
* Apply isort and black reformatting
Signed-off-by: meatybobby <meatybobby@users.noreply.github.com>
---------
Signed-off-by: meatybobby <meatybobby@users.noreply.github.com>
Co-authored-by: meatybobby <meatybobby@users.noreply.github.com>
* Hyena: Allow to use unfused RMSNorm + TELinear to restore accuracy and some speed (#14542)
* Fix sequence packing loss calculation (#14437)
* Fix sequence packing loss calculation
Signed-off-by: Rayan Dasoriya <dasoriyarayan@gmail.com>
* Fix nemo2 path
Signed-off-by: Rayan Dasoriya <dasoriyarayan@gmail.com>
* Skip pylint
Signed-off-by: Rayan Dasoriya <dasoriyarayan@gmail.com>
---------
Signed-off-by: Rayan Dasoriya <dasoriyarayan@gmail.com>
Co-authored-by: Dmytro Pykhtar <37850217+dimapihtar@users.noreply.github.com>
* [Audio]: added streaming mode to SpectrogramToAudio (#14524)
* [Audio]: added streaming mode to SpectrogramToAudio
Signed-off-by: Rauf <rnasretdinov@nvidia.com>
* added time buffer
Signed-off-by: Rauf <rnasretdinov@nvidia.com>
* renamed Nf -> num_frames
Signed-off-by: Rauf <rnasretdinov@nvidia.com>
* added AudioToSpectrogram and scale and magnitude power
Signed-off-by: Rauf <rnasretdinov@nvidia.com>
* added multiple chunking support
Signed-off-by: Rauf <rnasretdinov@nvidia.com>
* added properties _stream_initialized, _eps, got rid of _prev_spec_frame
Signed-off-by: Rauf <rnasretdinov@nvidia.com>
* added hanning window
Signed-off-by: Rauf <rnasretdinov@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: nasretdinovr <nasretdinovr@users.noreply.github.com>
* added a docstring regarding streaming istft mode
Signed-off-by: Rauf <rnasretdinov@nvidia.com>
---------
Signed-off-by: Rauf <rnasretdinov@nvidia.com>
Signed-off-by: nasretdinovr <nasretdinovr@users.noreply.github.com>
Co-authored-by: nasretdinovr <nasretdinovr@users.noreply.github.com>
* fix: fix missing rope scaling in exporting llama embedding model (#14523)
Signed-off-by: Zhiyu Li <zhiyul@NVIDIA.com>
* Update evo2 defaults so converted checkpoints have the right parameters (#14514)
* Update evo2 defaults so converted checkpoints have the right parameters
Signed-off-by: John St John <jstjohn@nvidia.com>
* Fix line too long issue
Signed-off-by: John St John <jstjohn@nvidia.com>
* Fix expected changes to configs that are locked into our tests
Signed-off-by: John St John <jstjohn@nvidia.com>
---------
Signed-off-by: John St John <jstjohn@nvidia.com>
* deprecate t0 scripts (#14585)
Signed-off-by: dimapihtar <dpihtar@gmail.com>
* cfg typo correction (#14588)
Signed-off-by: Malay Nagda <malayn@nvidia.com>
* [Perf script] Add use_te_activation_func and activation_func_fp8_input_store flags (#14522)
* Add use te activation func and save act input in fp8 flags
Signed-off-by: Guyue Huang <guyueh@nvidia.com>
* Fix field name
Signed-off-by: Guyue Huang <guyueh@nvidia.com>
* Update scripts/performance/vlm/finetune_qwen25vl_32b.py
Co-authored-by: malay-nagda <malayn@nvidia.com>
Signed-off-by: Guyue Huang <140554423+guyueh1@users.noreply.github.com>
---------
Signed-off-by: Guyue Huang <guyueh@nvidia.com>
Signed-off-by: Guyue Huang <140554423+guyueh1@users.noreply.github.com>
Co-authored-by: malay-nagda <malayn@nvidia.com>
* Modify logging message to signal that RestoreConfig will be used (#14469)
* Bump TE and Mcore (#14568)
* Bump TE and Mcore
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Use Mcore 69b65
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
---------
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Avoid host-device sync in PTL logging (#14489)
* remove sync in logging
Signed-off-by: qiyuw <qiyuw@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: WanZzzzzz <WanZzzzzz@users.noreply.github.com>
* add class and func docstrings in data_sampler.py for pylint
Signed-off-by: qiyuw <qiyuw@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: WanZzzzzz <WanZzzzzz@users.noreply.github.com>
---------
Signed-off-by: qiyuw <qiyuw@nvidia.com>
Signed-off-by: WanZzzzzz <WanZzzzzz@users.noreply.github.com>
Co-authored-by: qiyuw <qiyuw@nvidia.com>
Co-authored-by: WanZzzzzz <WanZzzzzz@users.noreply.github.com>
* Integrate implicit filter kernel with Hyena layer (#14621)
* add 1b arclongcontextconfig
Signed-off-by: Farhad Ramezanghorbani <farhadr@nvidia.com>
* fix device mess
Signed-off-by: Farhad Ramezanghorbani <farhadr@nvidia.com>
* add implicit_filter support
Signed-off-by: Farhad Ramezanghorbani <farhadr@nvidia.com>
* use padded input
Signed-off-by: Farhad Ramezanghorbani <farhadr@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: farhadrgh <farhadrgh@users.noreply.github.com>
* Revert "add 1b arclongcontextconfig"
This reverts commit 029969bae07e5c1651abd519640424d4aaece216.
---------
Signed-off-by: Farhad Ramezanghorbani <farhadr@nvidia.com>
Signed-off-by: farhadrgh <farhadrgh@users.noreply.github.com>
* Fix kv_channels configuration for Gemma2 27b (#14590)
* fix gemma2 27b kv dimension
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
* fix gemma2 27b kv dimension
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
---------
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
* [Flux] small fixes (#14333)
* feat: print expert groups on megatron init (#13874)
Signed-off-by: Alexander Zhipa <azzhipa@amazon.com>
Co-authored-by: Alexander Zhipa <azzhipa@amazon.com>
Signed-off-by: CarlosGomes98 <carlosmiguel.gomes@live.com.pt>
* set a different seed for each dp rank
Signed-off-by: CarlosGomes98 <carlosmiguel.gomes@live.com.pt>
* calculate loss inside autocast
Signed-off-by: CarlosGomes98 <carlosmiguel.gomes@live.com.pt>
* disable per token loss, grad acc fusion
Signed-off-by: CarlosGomes98 <carlosmiguel.gomes@live.com.pt>
* add missing self.seed
Signed-off-by: CarlosGomes98 <carlosmiguel.gomes@live.com.pt>
* black formatting
Signed-off-by: CarlosGomes98 <carlosmiguel.gomes@live.com.pt>
* Apply isort and black reformatting
Signed-off-by: gautham-kollu <gautham-kollu@users.noreply.github.com>
---------
Signed-off-by: Alexander Zhipa <azzhipa@amazon.com>
Signed-off-by: CarlosGomes98 <carlosmiguel.gomes@live.com.pt>
Signed-off-by: gautham-kollu <gautham-kollu@users.noreply.github.com>
Co-authored-by: Alexander Zhipa <alex.zhipa@proton.me>
Co-authored-by: Alexander Zhipa <azzhipa@amazon.com>
Co-authored-by: gautham-kollu <gkollu@nvidia.com>
Co-authored-by: gautham-kollu <gautham-kollu@users.noreply.github.com>
* [Flux] Add MXFP8 Support (#14473)
* [Flux] Add MXFP8 support.
Signed-off-by: Wil Kong <alpha0422@gmail.com>
* [Flux] Add current and block scaling.
Signed-off-by: Wil Kong <alpha0422@gmail.com>
---------
Signed-off-by: Wil Kong <alpha0422@gmail.com>
* use hf hub to download ckpt (#14638)
Signed-off-by: Ao Tang <aot@nvidia.com>
* Fine-tune embedding models (E5-Large-V2 and LLaMA-3.2-1B) on the allnli triplet dataset with NeMo Framework (#14584)
* Create E2E-Embedding-Finetuning
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Update E2E-Embedding-Finetuning
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Delete tutorials/llm/embedding/E2E-Embedding-Finetuning
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Create README.md
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Add files via upload
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Add files via upload
This is a notebook for E2E finetuning a embedding model
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Update README.md
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Update README.md
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Delete tutorials/llm/embedding/E2E-Embedding-Finetuning/download_dataset.py
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Delete tutorials/llm/embedding/E2E-Embedding-Finetuning/finetune_e5.py
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Delete tutorials/llm/embedding/E2E-Embedding-Finetuning/finetune_llama1b.py
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Delete tutorials/llm/embedding/E2E-Embedding-Finetuning/import_e5_large.py
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
* Delete tutorials/llm/embedding/E2E-Embedding-Finetuning/import_llama1b.py
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
---------
Signed-off-by: Hemant Giri <30834697+girihemant19@users.noreply.github.com>
Co-authored-by: Ao Tang <aot@nvidia.com>
* [Perf script] Llama and GPT3 perf script use mlp cast fusion
Signed-off-by: Guyue Huang <guyueh@nvidia.com>
* remove service launch scripts (#14647)
Signed-off-by: dimapihtar <dpihtar@gmail.com>
* warning instead of error with chat template (#14641)
Signed-off-by: jenchen13 <jennifchen@nvidia.com>
* fix notebook (#14643)
Signed-off-by: Chen Cui <chcui@nvidia.com>
* [Audio]: fixed bug in conformet unet (#14626)
Signed-off-by: Rauf <rnasretdinov@nvidia.com>
* Delete tutorials/llm/llama/biomedical-qa directory (#14653)
Signed-off-by: Chen Cui <chcui@nvidia.com>
* Fix code checkout during test (#14658)
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Fix Flux seed as optional Arg (#14652)
* fix flux seed as optional
Signed-off-by: Ao Tang <aot@nvidia.com>
* fix fluxcontrolnet
Signed-off-by: Ao Tang <aot@nvidia.com>
* Fix code checkout during test
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
---------
Signed-off-by: Ao Tang <aot@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Co-authored-by: Charlie Truong <chtruong@nvidia.com>
* remove older TTS tutorials (#14660)
Signed-off-by: Jason <jasoli@nvidia.com>
* Remove PEFT scheme condition from recipe (#14661)
* Remove PEFT scheme condition from recipe
Signed-off-by: Ali Taghibakhshi <71892896+JRD971000@users.noreply.github.com>
* remove unnecessary peft conditioning 12b
---------
Signed-off-by: Ali Taghibakhshi <71892896+JRD971000@users.noreply.github.com>
* Add gpt-oss lora exporter (#14589)
* add gpt-oss lora exporter
Signed-off-by: Chen Cui <chcui@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: cuichenx <cuichenx@users.noreply.github.com>
* update lora exporter for experts
Signed-off-by: Chen Cui <chcui@nvidia.com>
* disallow exporting expert lora since nemo implementation is not equivalent to hf
Signed-off-by: Chen Cui <chcui@nvidia.com>
* linting
Signed-off-by: Chen Cui <chcui@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: cuichenx <cuichenx@users.noreply.github.com>
* address comment
Signed-off-by: Chen Cui <chcui@nvidia.com>
---------
Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: cuichenx <cuichenx@users.noreply.github.com>
Co-authored-by: cuichenx <cuichenx@users.noreply.github.com>
Co-authored-by: Charlie Truong <chtruong@nvidia.com>
* Add NeMo Voice Agent (#14325)
* update streaming ASR
Signed-off-by: stevehuang52 <heh@nvidia.com>
* add voice agent
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update readme
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update websocket
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update readme
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update
Signed-off-by: stevehuang52 <heh@nvidia.com>
* clean up
Signed-off-by: stevehuang52 <heh@nvidia.com>
* clean up
Signed-off-by: stevehuang52 <heh@nvidia.com>
* fix typo
Signed-off-by: stevehuang52 <heh@nvidia.com>
* fix codeQL
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update cfg
Signed-off-by: stevehuang52 <heh@nvidia.com>
* remove unused
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update readme
Signed-off-by: stevehuang52 <heh@nvidia.com>
* change default models
Signed-off-by: stevehuang52 <heh@nvidia.com>
* fix diar diable
Signed-off-by: stevehuang52 <heh@nvidia.com>
* fix diar diable
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update ux
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update tts
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update readme
Signed-off-by: stevehuang52 <heh@nvidia.com>
* fix and update
Signed-off-by: stevehuang52 <heh@nvidia.com>
* fix asr
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update readmme
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update doc and llm dtype
Signed-off-by: stevehuang52 <heh@nvidia.com>
* refactor and add example prompts
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update readme
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update readme
Signed-off-by: stevehuang52 <heh@nvidia.com>
* clean up
Signed-off-by: stevehuang52 <heh@nvidia.com>
* clean up
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update info on streaming sortformer
Signed-off-by: stevehuang52 <heh@nvidia.com>
* move code to 'nemo/agents/voice_agent'
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update doc
Signed-off-by: stevehuang52 <heh@nvidia.com>
* clean up
Signed-off-by: stevehuang52 <heh@nvidia.com>
* refactor
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update doc
Signed-off-by: stevehuang52 <heh@nvidia.com>
* remove the unnecessary streaming state conversion and import it from sortformer_modules, remove PostProcessingParams
Signed-off-by: Weiqing Wang <weiqingw@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: weiqingw4ng <weiqingw4ng@users.noreply.github.com>
* update doc
Signed-off-by: stevehuang52 <heh@nvidia.com>
* clean up
Signed-off-by: stevehuang52 <heh@nvidia.com>
* fix for llama-nemotron template, and refactor
Signed-off-by: stevehuang52 <heh@nvidia.com>
* fix tts separator
Signed-off-by: stevehuang52 <heh@nvidia.com>
* fix for llama-nemotron
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update cfg
Signed-off-by: stevehuang52 <heh@nvidia.com>
* refactor and update doc
Signed-off-by: stevehuang52 <heh@nvidia.com>
* change default llm to qwen
Signed-off-by: stevehuang52 <heh@nvidia.com>
* update doc
Signed-off-by: stevehuang52 <heh@nvidia.com>
---------
Signed-off-by: stevehuang52 <heh@nvidia.com>
Signed-off-by: Weiqing Wang <weiqingw@nvidia.com>
Signed-off-by: weiqingw4ng <weiqingw4ng@users.noreply.github.com>
Co-authored-by: Taejin Park <tango4j@gmail.com>
Co-authored-by: Kunal Dhawan <kunaldhawan97@gmail.com>
Co-authored-by: Weiqing Wang <weiqingw@nvidia.com>
Co-authored-by: weiqingw4ng <weiqingw4ng@users.noreply.github.com>
* Update get_tensor_shapes function whose signature was refactored (#14594)
* Update get_tensor_shapes function whose signature changed and wasn't refactored
Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com>
* Bump Mcore commit to latest on 0.14.0 branch
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Bump Mcore
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Set flux fsdp test to optional
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Fix flux test to skip
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
---------
Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Co-authored-by: Charlie Truong <chtruong@nvidia.com>
* fixing kernel restarting when transcribing (#14665)
* fixing kernel restarting when transcribing
Signed-off-by: Weiqing Wang <weiqingw@nvidia.com>
* fixing the same issue for tutorials/asr/ASR_with_NeMo.ipynb
Signed-off-by: Weiqing Wang <weiqingw@nvidia.com>
* remove the change caused by IDE
Signed-off-by: Weiqing Wang <weiqingw@nvidia.com>
---------
Signed-off-by: Weiqing Wang <weiqingw@nvidia.com>
* Skip trt-llm and vllm install in install test (#14663)
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Canary tutorial fix (#14673)
Signed-off-by: Nune <ntadevosyan@nvidia.com>
* Downgrade "datasets" library version in ASR tutorial to ensure compatibility with HF Datasets used (#14679)
* downgrade dataset in notebooks to ensure comparibility with HF datsets used
Signed-off-by: Kunal Dhawan <kunaldhawan97@gmail.com>
* remove env information from notebook
Signed-off-by: Kunal Dhawan <kunaldhawan97@gmail.com>
---------
Signed-off-by: Kunal Dhawan <kunaldhawan97@gmail.com>
* End_to_End_Diarization_Training.ipynb (#14680)
Signed-off-by: taejinp <tango4j@gmail.com>
* Fix deepseek export dtype (#14307)
* add cast dtype option
Signed-off-by: Chen Cui <chcui@nvidia.com>
* linting
Signed-off-by: Chen Cui <chcui@nvidia.com>
* fix
Signed-off-by: Chen Cui <chcui@nvidia.com>
* add atol option
Signed-off-by: Chen Cui <chcui@nvidia.com>
* Update L2_NeMo_2_Conversion_Test_DeepSeek.sh
Signed-off-by: Chen Cui <chcui@nvidia.com>
* Update state.py
Signed-off-by: Chen Cui <chcui@nvidia.com>
* fix test
Signed-off-by: Chen Cui <chcui@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: cuichenx <cuichenx@users.noreply.github.com>
* fix test
Signed-off-by: Chen Cui <chcui@nvidia.com>
---------
Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: cuichenx <cuichenx@users.noreply.github.com>
Co-authored-by: cuichenx <cuichenx@users.noreply.github.com>
Co-authored-by: Charlie Truong <chtruong@nvidia.com>
* Delete nemo1 notebooks (#14677)
* Delete tutorials/llm/llama/sdg-law-title-generation directory
Signed-off-by: Chen Cui <chcui@nvidia.com>
* Delete tutorials/llm/llama/domain-adaptive-pretraining/code/domain_adaptive_pretraining_nemo1.0.ipynb
Signed-off-by: Chen Cui <chcui@nvidia.com>
---------
Signed-off-by: Chen Cui <chcui@nvidia.com>
* Bump latest Mcore 020abf01 (#14676)
* Bump latest Mcore
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Pin Mcore to 020abf01
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
---------
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* correct shapes (#14425)
Signed-off-by: CarlosGomes98 <carlosmiguel.gomes@live.com.pt>
Co-authored-by: gautham-kollu <gkollu@nvidia.com>
* Fix for "EncDecRNNTBPEModel transcribe() failed with TypeError" (#14698)
* fix decode_ids_to_str for AggregateTokenizer
Signed-off-by: andrusenkoau <andrusenkoau@gmail.com>
* minor fix
Signed-off-by: andrusenkoau <andrusenkoau@gmail.com>
---------
Signed-off-by: andrusenkoau <andrusenkoau@gmail.com>
* Bump modelopt to 0.35.0 and remove `safe_import("modelopt")` in llm collection (#14656)
* Bump modelopt to 0.35.0 and remove safe_import in llm collection
Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>
* Update eagle architecture spec setting
Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com>
* Reduce specdec memory usage
Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com>
---------
Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>
Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com>
Co-authored-by: Charlie Truong <chtruong@nvidia.com>
Co-authored-by: Asha Anoosheh <aanoosheh@nvidia.com>
* Tutorial fix (#14699)
Signed-off-by: Nune <ntadevosyan@nvidia.com>
* Add option for LoRA with Transformer Engine op fuser (#14411)
* Initial implementation of fused LoRA
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Get fused LoRA to run
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Initial work toward tensor-parallel support
Missing all-gather op
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Enable fused LoRA based on model config
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Tweak comments
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Add TE version checks
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Fix linter warning
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: timmoon10 <timmoon10@users.noreply.github.com>
* Use in-place fork/add ops to enable GEMMs with beta=1
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Add ops directly to te.op.Sequential
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Move fused LoRA impl into LoRALinear subclass
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Fix bug where fused impl was always disabled
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: timmoon10 <timmoon10@users.noreply.github.com>
* Support wgrad accumulation fusion
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Add integration test for TE op fuser
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: timmoon10 <timmoon10@users.noreply.github.com>
* Explicitly list module containers that are compatible with list or dict APIs
Mcore subclasses of te.ops.Sequential are iterable, but are not compatible with list API.
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: timmoon10 <timmoon10@users.noreply.github.com>
* Add missing docstring
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: timmoon10 <timmoon10@users.noreply.github.com>
* Update Mcore version
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Update Megatron-LM commit
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Attempt to support forward hooks in fused LoRA
Signed-off-by: Tim Moon <tmoon@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: timmoon10 <timmoon10@users.noreply.github.com>
---------
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Signed-off-by: timmoon10 <timmoon10@users.noreply.github.com>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Signed-off-by: gautham-kollu <gkollu@nvidia.com>
Co-authored-by: timmoon10 <timmoon10@users.noreply.github.com>
Co-authored-by: gautham-kollu <gkollu@nvidia.com>
Co-authored-by: Charlie Truong <chtruong@nvidia.com>
* add load-in-4bit param (#14636)
Signed-off-by: dimapihtar <dpihtar@gmail.com>
* fp4 support (#14625)
Signed-off-by: qiyuw <qiyuw@nvidia.com>
Co-authored-by: qiyuw <qiyuw@nvidia.com>
Co-authored-by: gautham-kollu <gkollu@nvidia.com>
* Update Reasoning-SFT.ipynb (#14716)
Signed-off-by: Chen Cui <chcui@nvidia.com>
* Remove artificial block to vortex fp8 TP (#14684)
* Remove artificial block to vortex fp8 TP
Signed-off-by: John St John <jstjohn@nvidia.com>
* Handle sequence_parallel=True TP>1 case properly where theres an all gather
Signed-off-by: John St John <jstjohn@nvidia.com>
---------
Signed-off-by: John St John <jstjohn@nvidia.com>
* Replace MegatronTokenizer with MegatronLegacyTokenizer (#14721)
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Update ModelCommPGs API from megatron-core (#14578)
* update
Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
* Bump Mcore to b615e73
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Replace ProcessGroupsCollection with ProcessGroupCollection
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Replace pgs_collection with pg_collection
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
---------
Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Co-authored-by: Charlie Truong <chtruong@nvidia.com>
* drop speech_llm example suite (#14683)
Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
* feat: Compatibility modification of megatron-fsdp (#14593)
* nvfsdp_update
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-ptyche01.ptyche.clusters.nvidia.com>
Signed-off-by: jianbinc <shjwudp@gmail.com>
* add megatron-fsdp checkpoint support
Signed-off-by: jianbinc <shjwudp@gmail.com>
* update use_custom_fsdp to use_megatron_fsdp
Signed-off-by: jianbinc <shjwudp@gmail.com>
* revert back pretrain_llama3_8b.py
formt code
Signed-off-by: jianbinc <shjwudp@gmail.com>
* Apply isort and black reformatting
Signed-off-by: shjwudp <shjwudp@users.noreply.github.com>
* keep use_custom_fsdp as backup and notify this will deprecated on m-core 0.14
Signed-off-by: jianbinc <shjwudp@gmail.com>
* Apply isort and black reformatting
Signed-off-by: shjwudp <shjwudp@users.noreply.github.com>
* fix CodeQL check
Signed-off-by: jianbinc <shjwudp@gmail.com>
---------
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-ptyche01.ptyche.clusters.nvidia.com>
Signed-off-by: jianbinc <shjwudp@gmail.com>
Signed-off-by: shjwudp <shjwudp@users.noreply.github.com>
Co-authored-by: Selvaraj Anandaraj <selvaraja@login-ptyche01.ptyche.clusters.nvidia.com>
Co-authored-by: shjwudp <shjwudp@users.noreply.github.com>
* imported get_moe_layer_wise_logging_tracker from megatron core moe_utils (#14694)
* imported get_moe_layer_wise_logging_tracker from megatron core moe_utils
Signed-off-by: Prathamesh Kalamkar <prathamk@thoughtworks.com>
* Apply isort and black reformatting
Signed-off-by: prathamk-tw <prathamk-tw@users.noreply.github.com>
* moved import to the top
* Apply isort and black reformatting
Signed-off-by: prathamk-tw <prathamk-tw@users.noreply.github.com>
---------
Signed-off-by: Prathamesh Kalamkar <prathamk@thoughtworks.com>
Signed-off-by: prathamk-tw <prathamk-tw@users.noreply.github.com>
Co-authored-by: prathamk-tw <prathamk-tw@users.noreply.github.com>
* cast SE weights and activations to fp32 (#14743)
Signed-off-by: Elena Rastorgueva <erastorgueva@nvidia.com>
* remove env var (#14739)
Signed-off-by: Malay Nagda <malayn@nvidia.com>
* detach arg option for run scripts (#14722)
* detach arg option for run scripts
Signed-off-by: Malay Nagda <malayn@nvidia.com>
* int dit opt instances
Signed-off-by: Malay Nagda <malayn@nvidia.com>
---------
Signed-off-by: Malay Nagda <malayn@nvidia.com>
* Use lhotse dataloader for ASR models to support in-manifest channel selection for multichannel recordings (#14586)
* make EncDecCTCModelBPE use lhotse dataloader when transcribing
Signed-off-by: Roman Korostik <rkorostik@nvidia.com>
* make EncDecHybridRNNTCTCBPEModel use lhotse dataloader when transcribing
Signed-off-by: Roman Korostik <rkorostik@nvidia.com>
* make EncDecRNNTBPEModel use lhotse dataloader when transcribing
Signed-off-by: Roman Korostik <rkorostik@nvidia.com>
* clarify some error messages
Signed-off-by: Roman Korostik <rkorostik@nvidia.com>
---------
Signed-off-by: Roman Korostik <rkorostik@nvidia.com>
* Randomized shard slicing for tarred data (#14558)
* Randomized shard slicing for tarred data
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
* Add shuffling shards in untarred sharegpt and multimodal conversation sources
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
* Extend slice_length support to multimodal and sharegpt conversations
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
* Update lhotse requirement version
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
---------
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
* Data prediction objective for flow matching speech enhancement models (#14749)
* flow matching: support x-prediction (data as target for the estimator)
Signed-off-by: Roman Korostik <rkorostik@nvidia.com>
* flow matching: fix model init in x-prediction case
Signed-off-by: Roman Korostik <rkorostik@nvidia.com>
* flow matching: add estimator_target to sampler in example configs
Signed-off-by: Roman Korostik <rkorostik@nvidia.com>
* flow matching: expand tests to include data prediction models
Signed-off-by: Roman Korostik <rkorostik@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: racoiaws <racoiaws@users.noreply.github.com>
---------
Signed-off-by: Roman Korostik <rkorostik@nvidia.com>
Signed-off-by: racoiaws <racoiaws@users.noreply.github.com>
Co-authored-by: racoiaws <racoiaws@users.noreply.github.com>
* Fix Some Failures (#14763)
* Use megatron_fsdp instead of custom_fsdp for Flux tests.
Signed-off-by: Wil Kong <alpha0422@gmail.com>
* Update megatron.core quick_gelu import path.
Signed-off-by: Wil Kong <alpha0422@gmail.com>
---------
Signed-off-by: Wil Kong <alpha0422@gmail.com>
* Support additional Slurm parameters (#14742)
* support additional slurm params and test with nemotron4
* fixed parsing of slurm params
* fix incorrect parsing due to fallback
* add support for all performance scripts
* Apply isort and black reformatting
* remove unused import
---------
Signed-off-by: bdubauski <bdubauski@users.noreply.github.com>
Signed-off-by: Barys Dubauski <bdubauski@nvdia.com>
Co-authored-by: Barys Dubauski <bdubauski@nvdia.com>
Co-authored-by: bdubauski <bdubauski@users.noreply.github.com>
* [Flux] Remove redundant host & device sync. (#14711)
Signed-off-by: Wil Kong <alpha0422@gmail.com>
Co-authored-by: gautham-kollu <gkollu@nvidia.com>
* [Flux] Add cuda_graph_scope and cache images ids for full iteration cuda graph. (#14744)
Signed-off-by: Wil Kong <alpha0422@gmail.com>
Co-authored-by: gautham-kollu <gkollu@nvidia.com>
* Add transducer timestamps without alignments, timestamps to streaming (#14766)
* refactored timestamps, fully identical to previuos
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* removed alignments from rnnt timestamps
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* clean up
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: lilithgrigoryan <lilithgrigoryan@users.noreply.github.com>
* clean up
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* fix tdt confidence without alignments
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* minor fix
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* minor fix
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* minor fix
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* minor fix
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* Add timestamps option to streaming inference script
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Fix config params
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Fix tdt
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* fix tdt durations, clean up
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: lilithgrigoryan <lilithgrigoryan@users.noreply.github.com>
* tests fix, clean up
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: lilithgrigoryan <lilithgrigoryan@users.noreply.github.com>
* remove starting SOS symbols from beam decodings to match timestamps length
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: lilithgrigoryan <lilithgrigoryan@users.noreply.github.com>
---------
Signed-off-by: lilithgrigoryan <lgrigoryan@nvidia.com>
Signed-off-by: lilithgrigoryan <lilithgrigoryan@users.noreply.github.com>
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
Co-authored-by: lilithgrigoryan <lilithgrigoryan@users.noreply.github.com>
Co-authored-by: Vladimir Bataev <vbataev@nvidia.com>
* Adding bf16 Sortformer train and inference (#14627)
* Adding disabled autocast on bce_loss
Signed-off-by: taejinp <tango4j@gmail.com>
* Adding Sortformer BF16 inference
Signed-off-by: taejinp <tango4j@gmail.com>
* Adding BF16 inference and adding a config
Signed-off-by: taejinp <tango4j@gmail.com>
* Apply isort and black reformatting
Signed-off-by: tango4j <tango4j@users.noreply.github.com>
* Adding bf16-mixed option for both training and inference
Signed-off-by: taejinp <tango4j@gmail.com>
* Apply isort and black reformatting
Signed-off-by: tango4j <tango4j@users.noreply.github.com>
* Adding bf16-mixed option for e2e_diarize_speech.py
Signed-off-by: taejinp <tango4j@gmail.com>
* Apply isort and black reformatting
Signed-off-by: tango4j <tango4j@users.noreply.github.com>
---------
Signed-off-by: taejinp <tango4j@gmail.com>
Signed-off-by: tango4j <tango4j@users.noreply.github.com>
Co-authored-by: tango4j <tango4j@users.noreply.github.com>
* Replace texterrors with kaldialign library (#14775)
* replace texterros with kaldialign for f-score computation
Signed-off-by: andrusenkoau <andrusenkoau@gmail.com>
* replace texterros with kaldialign for asr confidence
Signed-off-by: andrusenkoau <andrusenkoau@gmail.com>
* replace texterrors with kaldialign for ASR_Confidence_Estimation.ipynb
Signed-off-by: andrusenkoau <andrusenkoau@gmail.com>
* replace texterrors with kaldialing for ASR_Context_Biasing.ipynb
Signed-off-by: andrusenkoau <andrusenkoau@gmail.com>
* Apply isort and black reformatting
Signed-off-by: andrusenkoau <andrusenkoau@users.noreply.github.com>
* decrease kaldialign version
Signed-off-by: andrusenkoau <andrusenkoau@gmail.com>
---------
Signed-off-by: andrusenkoau <andrusenkoau@gmail.com>
Signed-off-by: andrusenkoau <andrusenkoau@users.noreply.github.com>
Co-authored-by: andrusenkoau <andrusenkoau@users.noreply.github.com>
* Update prune-distill notebooks to Qwen3 + simplify + mmlu eval (#14785)
* Update prune-distill notebooks to Qwen3 + simplify
Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>
* address comments
Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>
* Add readme.rst
Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>
---------
Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>
* ci: Automodel deprecation warning (#14787)
* add deprecation notice
Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
* add deprecation notice
Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
* add deprecation warning
Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
* remove import
Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
* move code
Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
* add more notices
Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: akoumpa <akoumpa@users.noreply.github.com>
* Remove automodel cicd
Signed-off-by: Dong Hyuk Chang <donghyukc@nvidia.com>
* Add deprecation notice for Automodel
Signed-off-by: Dong Hyuk Chang <donghyukc@nvidia.com>
---------
Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
Signed-off-by: akoumpa <akoumpa@users.noreply.github.com>
Signed-off-by: Dong Hyuk Chang <donghyukc@nvidia.com>
Co-authored-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
Co-authored-by: akoumpa <akoumpa@users.noreply.github.com>
* Remove export-deploy, automodel, and eval tutorials (#14790)
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Update gpt_oss.py (#14706)
Signed-off-by: Chen Cui <chcui@nvidia.com>
* MXFP8 must only use E4M3 as dtype (#14793)
Signed-off-by: Aditya Vavre <avavre@nvidia.com>
* fix: Use shutil.copy fallback to handle file metadata permission errors (#14639)
* Add fallback for file copy to handle metadata errors
Signed-off-by: vipnydav <vipinydv@google.com>
* Add robust_copy for resilient file copy
Signed-off-by: vipnydav <vipinydv@google.com>
* Apply isort and black reformatting
Signed-off-by: vipnydav <vipnydav@users.noreply.github.com>
* remove imported Path from test_file.py
Signed-off-by: vipnydav <vipinydv@google.com>
* Move robust_copy method to util file
Signed-off-by: vipnydav <vipinydv@google.com>
* Apply isort and black reformatting
Signed-off-by: vipnydav <vipnydav@users.noreply.github.com>
* Fix lint
Signed-off-by: vipnydav <vipinydv@google.com>
---------
Signed-off-by: vipnydav <vipinydv@google.com>
Signed-off-by: vipnydav <vipnydav@users.noreply.github.com>
Co-authored-by: vipnydav <vipnydav@users.noreply.github.com>
* OneLogger Integration (#13437)
* feat: add callback group definition & callback ABC
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: "Zhengjiang Shao" <zshao@nvidia.com>
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: PytLab <PytLab@users.noreply.github.com>
* feat: insert callback functions of CallbackGroup
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: "Zhengjiang Shao" <zshao@nvidia.com>
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: PytLab <PytLab@users.noreply.github.com>
* chore: PR test for jiashang
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* feat: use __init_subclass__ to cover all ModelPT subclasses
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: "Zhengjiang Shao" <zshao@nvidia.com>
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: PytLab <PytLab@users.noreply.github.com>
* feat: Adding metadata config manager poc
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: "Saju Prasad" <sajup@dc2-container-xterm-023.prd.it.nvidia.com>
Signed-off-by: Saju Prasad <sajup@dc2-container-xterm-023.prd.it.nvidia.com>
* Apply isort and black reformatting
Signed-off-by: sajup-oss <sajup-oss@users.noreply.github.com>
* feat: revert test changes.
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: Updating metadata attributes
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: sajup <sajup@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: sajup-oss <sajup-oss@users.noreply.github.com>
* fix: Adding OneloggerCallback
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: sajup <sajup@nvidia.com>
* fix: Reverting changes in examples/multimodal/speech_llm/modular_audio_gpt_train.py
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: sajup <sajup@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: sajup-oss <sajup-oss@users.noreply.github.com>
* fix: update modular models and megatron GPT models
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* feat: add on_app_start and on_app_end
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: Adding small test example for testing
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: sajup <sajup@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: sajup-oss <sajup-oss@users.noreply.github.com>
* fix: Fixing review comments as discussed with Jiashang
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: "Saju Prasad" <sajup@draco-oci-login-02.cm.cluster>
Signed-off-by: Saju Prasad <sajup@draco-oci-login-02.cm.cluster>
* Apply isort and black reformatting
Signed-off-by: sajup-oss <sajup-oss@users.noreply.github.com>
* fix: updating nemo code to v2
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: sajup <sajup@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: sajup-oss <sajup-oss@users.noreply.github.com>
* fix: updating wandb to get info from env
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: sajup <sajup@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: sajup-oss <sajup-oss@users.noreply.github.com>
* fix: fix som impl issue
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix issue for exp manager.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* feat: remove callback_group
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* feat: fix timingtracker issue
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* feat: fix for startup callbcaks
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* feat: change to adapter
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* feat: use new nv-one-logger
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* feat: add on_app_end
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* feat: make OneLogger configurable
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* feat: remove NeMocallback import
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* feat: fix the enable_onelogger setting.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* feat: clean the code.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* feat: enable onelogger
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* test: Adding few unit tests
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: "Saju Prasad" <sajup@cw-dfw-cs-001-vscode-01.cm.cluster>
Signed-off-by: Saju Prasad <sajup@cw-dfw-cs-001-vscode-01.cm.cluster>
* Apply isort and black reformatting
Signed-off-by: sajup-oss <sajup-oss@users.noreply.github.com>
* feat: tmp fix for functional testing.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: add on_app_end for NeMov2
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* fix: typo.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix the get attributes
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* fix: moving test test_meta_info_manager.py to tests/collections/common/
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: "Saju Prasad" <sajup@cw-dfw-cs-001-vscode-01.cm.cluster>
Signed-off-by: Saju Prasad <sajup@cw-dfw-cs-001-vscode-01.cm.cluster>
* fix: fix format issue.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix lint errors
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* Revert "Apply isort and black reformatting"
This reverts commit de6994d7e6e12e4040a5819cd1375c7a22ee7e0a.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Revert "fix: fix lint errors"
This reverts commit 8e47ecd749a1583597e8b8253f4eee4b231dbdf6.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix linting issues.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix linting issue
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: add copyright info
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: small fix.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix small issues for t5
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix dataloader issue.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* fix: remove dataloader setting.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* feat: update OneLogger.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix hydra runner.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: start using partial config.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix the unused variables
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: change get_one_logger name
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: code clean up.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: import more specific to avoid circular dependency. (#14306)
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: Peiyuan <qipeiyuan@outlook.com>
* fix: use ptl callback from ls
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* feat: fix meta info manager.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix meta data issue.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix the lint issue
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the unit tests.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix minor metadata issue.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix some test issues
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix pytest issue for meta info manager
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix lint issues for optimizers.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix pytest issues.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix CICD issues.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix all pytests
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* chore: fix lint
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* chore: fix unused import issues.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* chore: fix CICD issues.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix the CICD issues.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix the linting issue
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix CICD issues.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the circular import issue.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix some pytests.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: revert some change.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: error handling for init onelogger
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* chore: fix one_logger code.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* chore: remove unused vars.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix CICD for nemo
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* chore: fix NeMo CICD.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* chore: renaming onelogger
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* chore: fix some exception.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* chore: renaming.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* chore: resolve some comments.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* chore: remove duplicate init.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* chore: resolve some github comments.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* chore: fix the linting issue.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* chore(callbacks): restore generic CallbackGroup and route telemetry v… (#14628)
* chore(callbacks): restore generic CallbackGroup and route telemetry via group\n\n- Add BaseCallback and CallbackGroup with update_config and class init hook\n- Register OneLoggerAdapterCallback into group; merge config update into class\n- Replace direct OneLogger API usages with CallbackGroup across code\n- Ensure trainer attaches registered callbacks via group.update_config\n- Add nv-one-logger>=2.0.0 to base requirements\n\nSigned-off-by: Jiashang Hu <jiashangh@nvidia.com>
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* chore: renaming.
* chore: revert the change to install nv-one-logger
* chore: fix the linting issue
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
---------
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Co-authored-by: liquor233 <liquor233@users.noreply.github.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* Add tests for callback group (#14632)
* chore: fix some circular dependency issues.
* chore: move the files to utils.
* chore: add unit tests
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* chore: fix nv-one-logger tests
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* chore: fix lint issue.
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* chore: change the location.
* chore: remaining fix.
* chore: remaining changes.
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* chore: fix the tests
* chore: fix some lint.
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* Revert prompt_encoder.py to c5ef26c (Jason Wang) to undo auto-formatting
* pre-commit: exclude prompt_encoder.py from black/isort formatting
* chore: undo lasst commit.
* fix: fix some part for nemocallback.
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* chore: fix some pytest
* fix: verify the auto-hooked functions are called once
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
---------
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
Co-authored-by: liquor233 <liquor233@users.noreply.github.com>
Co-authored-by: Zhengjiang Shao <zshao@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix the double init issue
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
* fix: fix the push
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Guarantee one logger on_app_end calls (#14691)
* fix: guarantee on_app_end calls can be invoked finally
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
* feat: add context manager creator in CallbackGroup
* Revert "feat: add context manager creator in CallbackGroup"
This reverts commit 381f83de5c914f08707fecb22e4674e7b3f6b104.
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
---------
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
* fix: remove meta info manager (#14689)
* fix: remove meta info manager
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
---------
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Co-authored-by: liquor233 <liquor233@users.noreply.github.com>
* fix: fix some linting issues.
* fix: fix unit tests.
* chore: fix mcore
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the installing problem
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix requirements
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the mcore version.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the mcore version.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the mcore version.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the mcore version.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the mcore version.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the mcore version.
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: use correct global_step for async ckpt success event
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
* fix: fix unit tests
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix requirements
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: refactor the unit tests
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: insert callbacks in CallbackGroup before other PTL callbacks
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
* fix: fix call on app start flag
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* fix: fix unit tests
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: bump nv-one-logger version
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the unit tests
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* fix: fix the cicd issues.
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* fix: fix some lint issues
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix unused import
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: make oneloggernemocallback singleton
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix lint issues
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: make oneloggernemocallback singleton
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* fix: keep the original callbacks order in CallbackGroup when merging with trainer.callbacks
* fix: fix the unit tests
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix unit tests
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* fix: fix lint issues
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix the pickle issue.
* Apply isort and black reformatting
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
* fix: fix issue.
* fix: fix callback
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
* fix: fix callback group
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
---------
Signed-off-by: Zhengjiang Shao <zshao@nvidia.com>
Signed-off-by: PytLab <PytLab@users.noreply.github.com>
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>
Signed-off-by: Saju Prasad <sajup@dc2-container-xterm-023.prd.it.nvidia.com>
Signed-off-by: sajup-oss <sajup-oss@users.noreply.github.com>
Signed-off-by: liquor233 <jiashangh@nvidia.com>
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: sajup <sajup@nvidia.com>
Signed-off-by: sajup <sajup@nvidia.com>
Signed-off-by: Saju Prasad <sajup@draco-oci-login-02.cm.cluster>
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: liquor233 <jiashangh@nvidia.com>
Signed-off-by: Saju Prasad <sajup@cw-dfw-cs-001-vscode-01.cm.cluster>
Signed-off-by: Jiashang Hu <jiashangh@nvidia.com>\nSigned-off-by: Peiyuan <qipeiyuan@outlook.com>
Signed-off-by: liquor233 <liquor233@users.noreply.github.com>
Co-authored-by: PytLab <PytLab@users.noreply.github.com>
Co-authored-by: Jiashang Hu <jiashangh@nvidia.com>
Co-authored-by: Saju Prasad <sajup@dc2-container-xterm-023.prd.it.nvidia.com>
Co-authored-by: sajup-oss <sajup-oss@users.noreply.github.com>
Co-authored-by: sajup <sajup@nvidia.com>
Co-authored-by: liquor233 <liquor233@users.noreply.github.com>
Co-authored-by: Saju Prasad <sajup@draco-oci-login-02.cm.cluster>
Co-authored-by: Saju Prasad <sajup@cw-dfw-cs-001-vscode-01.cm.cluster>
Co-authored-by: Peiyuan <qipeiyuan@outlook.com>
Co-authored-by: Peiyuan Qi <bqi@nvidia.com>
* Disable blank Issues (#14788)
Signed-off-by: Pablo Garay <pagaray@nvidia.com>
* Add community label bot (#14796)
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Add mistral small3 24B config and recipe (#14784)
* Add mistral small3 24B config and recipe
Signed-off-by: Joosung Yoon <joosungy@nvidia.com>
---------
Signed-off-by: Joosung Yoon <joosungy@nvidia.com>
Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
Signed-off-by: Alexandros Koumparoulis <153118171+akoumpa@users.noreply.github.com>
Co-authored-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
Co-authored-by: Alexandros Koumparoulis <153118171+akoumpa@users.noreply.github.com>
* Update changelog for `r2.3.0` (#14812)
* beep boop: Update changelog
Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* Update changelog for 2.3.3
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
* Fix changelog for 2.3.3
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
---------
Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: Charlie Truong <chtruong@nvidia.com>
* QWEN2.5-VL 7B FP8 Recipe (#14801)
* QWEN2.5-VL FP8 Recipe
Signed-off-by: Lifu Zhang <tomzhanglf@gmail.com>
* Apply isort and black reformatting
Signed-off-by: tomlifu <tomlifu@users.noreply.github.com>
* add model configs
Signed-off-by: Lifu Zhang <tomzhanglf@gmail.com>
---------
Signed-off-by: Lifu Zhang <tomzhanglf@gmail.com>
Signed-off-by: tomlifu <tomlifu@users.noreply.github.com>
Co-authored-by: tomlifu <tomlifu@users.noreply.github.com>
* disk space management: nemo install test (#14822)
* Add Customization Capabilities to Cache-Aware Models (#14757)
* Add Customization Capabilities to Cache-Aware Models
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Unify params with other transcription scripts
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Fix usage with manifests containing relative paths
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Fix decoding config setup
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Return back output_path
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Raise not implemented error if batched beam search performed with partial hypotheses
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Raise not implemented error if batched beam search in transducer performed with partial hypotheses
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Fix after merge
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Fix att_context_size param
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Use optional for left_chunks
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Apply isort and black reformatting
Signed-off-by: artbataev <artbataev@users.noreply.github.com>
* Unify parameters with transcribe_speech
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Fix docstring
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Unify dtype selection
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Fix unused variables
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
* Enhance inline documentation. Set compute_dtype=float32 by default.
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
---------
Signed-off-by: Vladimir Bataev <vbataev@nvidia.com>
Signed-off-by: artbataev <artbataev@users.noreply.github.com>
Co-authored-by: artbataev <artbataev@users.noreply.github.com>
* Evo2 address rare over-masking in 1m context dataset (#14821)
* Address problems where sometimes in 1m dataset there are very large masked segments
Signed-off-by: John St John <jstjohn@nvidia.com>
* only flip the tag extra if the segment length is too long
Signed-off-by: John St John <jstjohn@nvidia.com>
* Undo the change to the pre commit config
Signed-off-by: John St John <jstjohn@nvidia.com>
* Add clarifying comments about the state flipping logic
Signed-off-by: John St John <jstjohn@nvidia.com>
---------
Signed-off-by: John St John <jstjohn@nvidia.com>
* Update cherry-pick workflow to use version 0.63.0 (#14832)
* Update cherry-pick workflow to use version 0.63.0
Signed-off-by: Pablo Garay <palenq@gmail.com>
* Update cherry-pick workflow version tag
Signed-off-by: Pablo Garay <palenq@gmail.com>
---------
Signed-off-by: Pablo Garay <palenq@gmail.com>
* docs: Removing automodel items (#14840)
Signed-off-by: Andrew Schilling <aschilling@nvidia.com>
* update docs per guidance (#14841)
* Update changelog for `v2.4.1` (#14828)
* beep boop: Update changelog
Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* Fix changelog for 2.4.1
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
---------
Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: Charlie Truong <chtruong@nvidia.com>
* Fi…