-
Hello all,
I'm using my own data for training a transformer model for machine translation. I am using the standard pipeline with t2t-datagen and t2t-trainer and it's fine to train the model. In som…
-
**Describe the bug**
I think there might be something wrong with the current LISA implementation. There is no difference in training loss, no matter how many layers are active.
Not using LMFlow bu…
-
I am getting an `pickle.UnpicklingError` when trying to train again on a previously trained checkpoint with open_clip `v2.27.0+`.
This is similar to https://github.com/mlfoundations/open_clip/issue…
-
**Motivation**
Modern analytical engines use relational operator properties to find optimal plan. Property is a value associated with the operator that doesn't change operator's equivalence and t…
-
```
What steps will reproduce the problem?
1. Wrap some code that contains WRITE statements with f2py
2. Try to capture this output on Python stdout / stderr descriptors
3. Watch yourself fail
What i…
-
Since our DistributedOptimizer takes another optimizer and turns it into ZeRO-1, can we make it do a fused optimizer like this? It should take an optimizer and turn it into a fused ZeRO-1 in a generic…
-
## 🐞Describe the bug
- Using TensorFlow/keras with mixed precision training fails to export
- Keras converter issue
NotImplementedError: Cast: Provided destination type fp16 not supported.
## …
-
Currently, those passes only use the main thread, but those passes are mostly stateless and we need to move logic to minifier anyway, because of side effect check.
I profiled it, and those tw…
-
From my very brief go-over of the code, it looks like it's only looking for some of the framework-specific trace messages on traced scripts. For more accurate metric, we can add some kind of small mar…
-
Because the synchronous Compile API has a 4kb limit in real-world implementations and the async/streaming APIs use a Promise, there's currently no way to implement a wasm equivalent of production JITs…