Jae-Won Chung
Jae-Won Chung
Hi, I forked this repository and read it through, upon which I found numerous less-pythonic/inefficient pieces of code, less-informative object names, and unused arguments and variables. Examples include things like...
### Describe the bug I have completions (via the `autocompletion` keyword argument in Arguments and Options) that include a colon (`:`) in the string. In zsh, these completions get split...
Hi, I'm seeing if it's possible to integrate `adaptdl` with PyTorch Lightning (specifically the [Deepspeech2 open source repo](https://github.com/SeanNaren/deepspeech.pytorch)). Potential problems I see are: 1. Would the `adaptdl` specific the dataloader...
Hi! I'm trying out the pipeline parallelism example in `microsoft/DeepSpeedExamples` [here](https://github.com/microsoft/DeepSpeedExamples/tree/master/pipeline_parallelism) on DeepSpeed commit ebed51df787579b8b3f836bb155ce0bf97f4ab66. I turned on `wall_clock_breakdown` in `ds_config.json` and have four pipeline stages across four A40 GPUs....
Happy new year! I built a docker image and pushed to docker hub for armv7l with torch 1.2.0 and torchvision 0.4.0. Hope this helps for those who are struggling with...
I believe that `TrainSchedule`, which implements the synchronous 1F1B pipeline training schedule, uses one more buffer than what is actually needed in every pipeline stage except for the last. https://github.com/microsoft/DeepSpeed/blob/d323abd80f62bebb9924bb85feb72b57c25af50d/deepspeed/runtime/pipe/schedule.py#L243-L247...
I've added support for the `revision` parameter in `load_model` and `load_compress_model`. It explicitly defaults to `"main"`, which is also the default in Huggingface `from_pretrained` methods. I believe all of the...
Hi, Amazing framework. Thanks a lot. I was comparing the inference speed (seconds) of `python -m fastchat.serve.cli` with and without `--load-8bit`. Both with the Vicuna 7B and 13B models, `--load-8bit`...
Added: - Zeus [framework](https://ml.energy/zeus) and [research paper](https://www.usenix.org/conference/nsdi23/presentation/you) -- ML energy optimization based on measurement - [ML.ENERGY Leaderboard](https://ml.energy/leaderboard) -- LLM inference energy measurement - [Chase paper](https://www.climatechange.ai/papers/iclr2023/29) - Carbon optimization for DNN...
I'm experiencing frequent segfaults on MacOS 12.6 when my Neovim's Vimtex triggers automatic LaTeX compilation on file save. I've attached the MacOS crash report (the window that asks whether or...