-
in your code, tf.placeholder is not compatible with eager execution
but when I turn of eager mode, there is a new error "numpy() is only available when eager execution is enabled"
how can I resolve …
-
Does LEAP do eager loading at all?
We've just been profiling our application and LEAP is making a extreme number of select statements.
Is there anyway to get LEAP to autoload all the relations when…
-
**Describe the bug**
I am trying to train a NeRF with my image data (that I ran through colmap), following the instructions in [this page](https://docs.nerf.studio/quickstart/custom_dataset.html). Ru…
-
Description : How linkedList is implemented internally to have better understanding of data structure and know its implementation with the help of classes in java.
Please assign me these problem un…
-
Tracking bug for implementing lazy expansion of symbolic macros (the eager version of which was implemented in #19922).
Tentatively targeting Bazel 8.1.
@brandjon @susinmotion FYI
-
### 🐛 Describe the bug
Category | Model | Accuracy
-- | -- | --
torchbench_amp_bf16_training | basic_gnn_edgecnn | eager_two_runs_differ
torchbench_amp_bf16_training | …
-
### 🐛 Describe the bug
When using torch.compile on a model that includes the backward pass of torch.cumprod, the compiled version is significantly slower than the eager execution, especially for la…
-
Filing this eagerly, because I figure it can't hurt. :)
-
### System Info
- TensorRT-LLM version: 0.10.0.dev2024050700
(I doubt any other information is relevant)
### Who can help?
@kaiyux
### Information
- [ ] The official example scripts
- […
-
v100和npu下不支持flash attention,于是尝试了另外两种attention实现。
pytorch2.1.1,尝试了sdpa和eager,输出都比较异常:
sdpa,输出N个(N>=1):
enerated_ids_trimmed [tensor([151644, 151644, 151644, 151644, 151644, 151644, 151644, 151644…