-
### Issue type
Feature Request
### Have you reproduced the bug with TensorFlow Nightly?
Yes
### Source
binary
### TensorFlow version
tf 2.15
### Custom code
No
### OS pla…
-
This is an API specification for:
- Threadpools and other executors using the Task Parallelism and fork-join model
- Channels
- Some extra definitions so that we have a common vocabulary.
The RF…
-
So our DQN training is exceptionally slow. Currently my projection for 10,000 episodes is 550 days, and thats only testing for a sample size of 50 episodes. Later in the training we should be reaching…
-
# Log compaction for hybrid state machines
## Overview
A hybrid state machine is a state machine that keeps some of its state in memory
and some of its state, typically payload data, in the Raf…
-
I tried multiple sets of experiments, but found that ZB is better than 1F1B. Interleaved 1F1B seems to be slightly faster than ZB_V, slightly slower than ZB_2P but saves a lot of GPU memory.
mach…
zhj96 updated
6 months ago
-
At the recent Kokkos bootcamp in Oakland, I spoke with @crtrott and learned that hierarchical parallelism is not currently supported when a WorkGraphPolicy is used. I would like to request this featur…
-
### 🐛 Describe the bug
I understand that this error came out of flash attention software stack, but it seems there is no related issue except for #https://github.com/Dao-AILab/flash-attention/issues/…
-
**Is your feature request related to a problem? Please describe.**
The current Spark shuffle has two modes. For tasks with less than 200 partitions, configurable. It will write the shuffle data out t…
-
## 🚀 Feature
## Motivation
Let's suppose we have a model (e.g., `torchvision.models.resnet18()`. The user wants to extract a set of intermediate activations of this model (e.g., to get the out…
-
-- Process 0 terminated with the following error:
Traceback (most recent call last):
File "C:\Users\11500\.conda\envs\GPTSoVits\lib\site-packages\torch\multiprocessing\spawn.py", line 76, in _wrap…