-
`TracingIterator` wraps `Iterator` returned from kafka, and starts scopes in `next()` and closes them in `hasNext()` for each record in the iterator.
This is a clever way to put the context in scop…
-
**Describe the bug**
Current code is not atomic and is not batch safe. Also no roll-back.
If a documents chunks span a batch the upsert will remove docs and reinsert in one batch and then redo it…
-
### Description
The below snippet causes drastically different results if the loss function is JIT'd vs not compiled. This results in a bad optimization where the version with the JIT'd loss functi…
-
### Contact Details
_No response_
### Description
**Description:**
When I select "Add a Sketch," the planes are displayed, but it is impossible to select one. There is no hover effect, and clickin…
-
### System Info
transformers version - 4.33.2
I'm using the trainer api as such, so it pushes the latest checkpoint to huggingface each epoch:
```
from transformers import TrainingArguments, T…
-
Hi there,
There are some variants which are not singletons which are called singletons from [create_singleton_dump.py](https://github.com/weinstockj/passenger_count_variant_calling/blob/master/crea…
-
### 🐛 Describe the bug
I found that `torch.nn.InstanceNorm2d` layer, which by default does not track running mean and var, is failing at runtime. It successfully goes through the lowering process wit…
-
Dear all,
I know this topic has been addressed but I cannot make it right.. I need to merge to custom made a3m files of an heterodimer to run colabfold_batch in multimer mode.
What I tried is a…
-
Empty line in batch breaks batch pruning in beam search. We remove finished batch entries during beam-search. This fails when an entry in the batch is empty.
-
### Misc discussion on performance
Checking #6164, _prepare_model_input_tensors has been refactored for the purpose of performance improvement. I investigated the performance of _prepare_model_input_…