-
## What
I propose to provide a consistent reliable way of handling concurrent state updates.
## Why
At the moment some concurrent updates result into obsolete `fields`. That causes various unexpe…
-
My actor model and critic model are both 7b in size. When I run the step3, there will be memory overflow, which consumes about 225G of memory space. Is there any solution to reduce memory consumption …
-
I need to wait on multiple workflows' successful completion.
If I don't misunderstand this action, the only way to achieve this when using the `wait-on-check-action` is to repeat the action as many…
-
When utilizing Axolotl, the training loss reduces to 0 following the gradient accumulation steps. Is this expected behaviour?
With Torchrun, the training loss consistently remains NaN.
Thank…
-
By: anonymous
- Specification file
- Unclear definition and terminology
- Clients, servers, senders, recipients -- this led to confusion
over message semantics
-…
-
https://github.com/MaterializeInc/materialize/pull/20995 introduced a consolidation at every `Union` that has a `Negate`ed input, to quickly resolve the serious performance issue resulting from the ac…
-
Thanks for the open source of Video-ChatGPT, I really like this work very much.
I am now trying to train Video-ChatGPT now.
However, I only have a single node server with 8 4090 GPUs.
I would lik…
-
**Note from the teaching team:** This bug was reported during the _Part II (Evaluating Documents)_ stage of the PE. **You may reject this bug if it is not related to the quality of documentation.**
![…
-
**Source / repo**
https://github.com/huggingface/pytorch-image-models
**Model description**
[DESCRIPTION]
**Dataset**
[DATASET]
**Literature benchmark source**
[URL]
**Literature benchmark per…
-
When using Monster Blocks, every time I open a token's sheet that is present on the map, all of the Active Effects with math operators fire and are resolved. This causes the values to be incorrect fro…