You Jiacheng
You Jiacheng
### Feature request `torch.nn.Linear,Conv2d...` will call `self.reset_parameters()` inside their `__init__`. I'd like to make `reset_parameters` be a no-op inside `no_init_weights` context manager. ### Motivation `no_init_weights` is used in `from_pretrained` to...
I would appreciate it if anyone could let me know which part of jittor source code do the optimization. Thanks in advance!💕
I want to use jittor as a backend of JAX(use custom interpreter of JAX), thus it will be great to have a static graph construct(and run) API.
After https://github.com/google/jax/commit/2bd010ae889461edb95f5cd102e74278f888e7cc many variants of XLA translation rules are consolidated into one unified new style, `xla.backend_specific_translations` is only backward compatibility shim now. FWIW, most of `jax.lax` primitive has been migrated...
I'd like that model(with `Optimizer`) and `DataLoader` use different `device_placement` value to be possible. i.e., I want to let accelerate deal with device placement for model, but not for dataloader....
I want to check the integrity of downloaded data, and I tried: ```python from hashlib import md5 import mmap from pathlib import Path from urllib.parse import urlparse import boto3 s3...
From timstamp 532s to 1063s. Video player(including https://visualize.ego4d-data.org/011ee98a-afc2-4088-b4d1-e0f7d84f3611 web player) and nvidia-dali will stuck. PyAV will decode black frames.
It seems to cause performance issues since I use NFS which may be backed by blob storage and append will trigger full rewrite.
My Python is slow when import packages(e.g. via network), it can easily exceed the 30 seconds timeout in `_wait_for_ports` since the server setup is slow. Then I got following errors:...
If UI and business logic share same data and/or data are lists, imperative UI model is really verbose. If I use declarative UI model, following code is enough. But if...