models

models copied to clipboard

[WIP] Add horovod for distributed training

1

WIP. Moved from #783 because CI was stuck.

edknv

Retrieval models definition using new RetrievalModelV2, InputBlockV2 and ModelOutput classes

16

Fixes #611 Fixes #805 Uses a workaround in the integration test to avoid the issue #144 ### Goals :soccer: - The objective of this PR is to implement a new...

sararb

Ensure metrics passed to `model.compile()` are always reset

2

Fixes #144 ### Goals :soccer: - Debugging the source of issue #144, I noticed that Keras `.evaluate()` method does not reset metrics defined in the `compile()` methods. - While this...

sararb

[Task] Profiling the training&evaluation pipeline for retrieval and ranking models

4

# 🚀 Feature request Profiling the training&evaluation pipeline for retrieval and ranking models and finding bottlenecks that should be solved for better performance and GPU utilization. For performing this task,...

gabrielspmoreira

### Bug description I get the following error when I set `post='sequence_mean'` arg. ``` AttributeError Traceback (most recent call last) Cell In [40], line 39 22 model = mm.Model( 23...

rnyak

Dataset/booking

2

Fixes # (issue) ### Goals :soccer: ### Implementation Details :construction: ### Testing Details :mag:

marcromeyn

[BUG] Evaluation scores of topk_encoder.evaluate(...) are inconsistents

1

### Bug description I observed inconsistent evaluation metrics when running the integration tests with the new API: The first call of topk_encoder.evaluate() has some internal states in metrics that make...

sararb

[BUG] InputBlock gives incorrect shape for continuous sequence features

1

### Bug description When using InputBlock to take in continouse sequence features, expected shape would be `[batch_size, seq_length, num_cont_features]`, but instead if became `[batch_size, num_cont_features*seq_length]`. And when concatenated with categorical...

zhiruiwang

### Goals :soccer: - Improve reliability of tests using top-k metrics ### Implementation Details :construction: - Adds `seed` parameter to TopKMetric classes to ensure results are consistent when evaluation run...

oliverholworthy

Incremental training with layer freezing exp

19

This PR cleans and organizes the example [09-Incremental-Training-with-Different-Learning-Rates-and-Layer-Freezing.ipynb](https://github.com/NVIDIA-Merlin/models/pull/667/files/1165d07b6765471dc7cb697918068f93184db873#diff-cb7137ba6d70ba60f4e5bfa06cf5d5121667447eb7ca41f339c1329e02203af2) in this PR https://github.com/NVIDIA-Merlin/models/pull/667. Missing: Will update the logo.

rnyak