Alexandre Strube

Working demo?

2

Hey, I just got here via the TLDR Newsletter (https://tldr.tech) This project looks great! However, some screenshots or even a demo would be a nice showcase of what's expected to...

Generating shell scripts instead of functions for wrappers without specifying on the container.yaml

14

As the wrapper_scripts option is now enabled by default, shpc will only create scripts in cases the container.yaml file has the docker_scripts/singularity_scripts tag. My suggestion would be that, instead of...

feature request: container URIs that require auth

9

https://github.com/singularityhub/singularity-hpc/blob/main/registry/nvcr.io/nvidia/hpc-benchmarks/container.yaml It seems like there was an error during the automated creation of this file, or am I misinterpreting the `latest` and `tags`?

MPI_COMM_TYPE_SHARED returns NUMA domains, resulting in wrong hvd.local_rank

4

**Environment:** 1. Framework: Tensorflow/PyTorch 2. Framework version: all 3. Horovod version: all 4. MPI version: all 5. CUDA version: all 6. NCCL version: all 7. Python version: 3.* 8. Spark...

lmod.load from within jupyter does not behave the same as the jupyter-lmod load [aka export modules]

1

The jupyter-lmod interface has a button which shows the following: > Export modules >Add this in a notebook to load the same modules : > ```import lmod > await lmod.purge(force=True)...

--auto-devices goes to a single gpu

2

### Describe the bug When running the server in default, it maps to a couple gpus, but not so well - the last ones are always underused. Llama-65b-hf, for example,...

Fix for `ValueError: Tokenizer class GPTNeoXTokenizer does not exist or is not currently imported.`

1

According to the discussion on transformers, there's a fix for FastChat: https://github.com/huggingface/transformers/issues/17756#issuecomment-1573319214 ```patch diff --git a/fastchat/model/model_adapter.py b/fastchat/model/model_adapter.py index facfbee..c1b6d35 100644 --- a/fastchat/model/model_adapter.py +++ b/fastchat/model/model_adapter.py @@ -43,7 +43,7 @@ class BaseAdapter:...

Different worker with different models don't update the web interface

1

It seems like when I have different workers with different models, I still only see one of them. Like here, I have a worker on port 21002 and one worker...

Seems like the num_gpus counts from 0. ``` CUDA_VISIBLE_DEVICES="3,4" python3 -m fastchat.serve.model_worker --model-path ../text-generation-webui/models/vicuna-13b/ --port 310001 --worker http://localhost:31001 --num-gpus 2 2023-06-16 10:07:22 | INFO | model_worker | args: Namespace(host='localhost', port=310001,...