Timur Abishev

Add performance tests for flatMap operation in OperationContainer

9

This PR adds tests for `flatMap` / `flatMap` composition in context of `OperationContainer`. Tests are implemented using JMH. Tests are not ideal, but they give us some estimation of costs...

We added `summingbird-core-test/src/main/scala/com/twitter/summingbird/ArbitraryWorkaround.scala` file to workaround scala check issue. It should be removed as soon as scalacheck release will be out.

In case of two parallel `sumByKey`s with different parameters we enable partial aggregation on `flatMap` with parameters of one of them, which is wrong.

Storm platform fails in case of `sumByKey` in one branch and `flatMap` in another with `flatMap`ed source fails in runtime

1

Producer ``` val source = spoutSource.flatMap(e => List(e)) source.sumByKey(store1).also( source.flatMap(branchFlatMap).sumByKey(store2) ) ``` fails with `java.lang.ClassCastException: java.lang.Integer cannot be cast to scala.Tuple2` exception in runtime.

Support INT4 Weights, FP8 Activations in TRT-LLM

7

### System Info H100, Llama 70B ### Who can help? _No response_ ### Information - [ ] The official example scripts - [ ] My own modified scripts ### Tasks...

Support Gemma 1.1 model

5

### System Info Model: https://huggingface.co/google/gemma-1.1-2b-it ### Who can help? @byshiue ### Information - [ ] The official example scripts - [ ] My own modified scripts ### Tasks - [...

In order to support Gemma 1.1 definition from huggingface (https://huggingface.co/google/gemma-1.1-2b-it/blob/main/config.json) `gelu_pytorch_tanh` activation function should be recognized.

### System Info - 1x H100 SXM - tensorrt-llm 0.12.0.dev2024080600 ### Who can help? @Tracin _No response_ ### Information - [ ] The official example scripts - [ ] My...