Artem Bolgar
Artem Bolgar
Just recently hit that issue: different browsers may report different number of buttons (and theoretically - axes) for the same controllers (even though the order of those, obviously, must be...
It makes sense to remove WebVR once WebXR is there, but AFAICS the WebVR is removed w/o adding WebXR. The origin trial is expiring in November, AFAIK. What is the...
Current A-Frame here is 0.7.0 which is old and doesn't have many fixes. For example, it works purely with Oculus Browser on Go / Quest. Updating A-Frame is not simple...
AFAIK, there is no way to control which color space is used for rendering. I am not sure we even define the default color space which is used by the...
### Description Experimental/Whisper notebook (speedup.ipynb) is not working. When run the unmodified notebook on RTX 4080/4090 (i.e., it is using the large-v2 model), it takes a lot of time to...
### Description Torch.dynamo is not working on H100 due to obsolete triton & pytorch ### Steps to reproduce Easily reproducible on H100 by running 'pytest -k benchmark' ### Expected Behavior...
Adding id_to_piece, piece_to_id and is_special_token functionality to TokenizerInterface and the corresponding implementations. Thus, the interface can be used by user's code to encode/decode single tokens. These new functions are not...
Int4 quantization requires CUDA device, however, in current impl --device param was overridden with 'cpu' unconditionally.
According to Mistral's paper the block size for Mistral-7B should be 8192 (ref: https://arxiv.org/pdf/2310.06825.pdf, https://huggingface.co/docs/transformers/en/model_doc/mistral). But currently it is set to the default value (2048).
Seems like requirements.txt is a bit obsolete. Also adding .venv and .vscode dirs to gitignore