Tom Fogal

Implement clone()

4

Fixes #342. - [x] Was this discussed/approved via a Github issue? (no need for typos and docs improvements) - [x] Did you read the [contributor guideline](https://github.com/Lightning-AI/pytorch-lightning/blob/main/.github/CONTRIBUTING.md), Pull Request section? -...

Our nightly (but not 2.3) CI jobs are spitting out warnings: ``` thunder/tests/distributed/test_ddp.py::CompileDDPTest::test_ddp_model_as_argument PASSED /usr/local/lib/python3.10/dist-packages/torch/autograd/profiler.py:215: UserWarning: The attribute `use_cuda` will be deprecated soon, please use ``use_device = 'cuda'`` instead. warn(...

Enable cuDNN executor by default

3

## 🚀 Feature Now that cudnn-fe is publicly installable, we should add it into the default executors. ### Motivation cudnn often gives great performance :-) ### Pitch https://github.com/Lightning-AI/lightning-thunder/blob/73ca5dc79179773c542d71b3f81d02ae0a220329/thunder/__init__.py#L162 this logic...

Support NeMo MegatronImagen network

1

## 🚀 Feature NeMo's "Imagen" network implements text-to-image generation. Initial `examine`: ``` Found 55 distinct operations, of which 46 (83.6%) are supported ``` ### Motivation ### Pitch ### Work items...

Call teardown even when something fails.

2

# What does this PR do ? Avoid tons of terminal spam such as: ``` [rank0]:[W528 18:14:19.086678488 ProcessGroupNCCL.cpp:1122] WARNING: process group has NOT been destroyed before it is being destructed....

Unknown attribute _base inside Megatron core

12

## 🚀 Model / language coverage Running Megatron GPT from NeMo, we seem to have issues with [this line from Megatron core](https://github.com/NVIDIA/Megatron-LM/blob/c7a1f82d761577e6ca0338d3521eac82f2aa0904/megatron/core/utils.py#L147). [Some context from the caller in this particular...

enter/exit_autocast of torch.amp.autocast_mode

5

## 🚀 Model / language coverage I'm trying to get a fuller picture of what we need to support NeVA. As such I'm using: ```python def thunder_backend(gm, args): gm.real_recompile() from...

Add nvtx around some important components.

3

## What does this PR do? Starts to add nvtx around (some) of the larger or at least more variable components of our compilation and execution. These help with identifying...

Support Phi3-mini-128k

3

## 🚀 Model / language coverage Support the https://huggingface.co/microsoft/Phi-3-mini-128k-instruct model. This is a tracking issue. Dynamo is splitting this into 13 subgraphs. The good news is that `examine` doesn't find...

## 🚀 Model / language coverage Support the [Qwn2-VL-2B-Instruct model](https://huggingface.co/Qwen/Qwen2-VL-2B-Instruct). ### Pitch This is an ask from internal NVIDIA colleagues. ### Minimal Repro TBD. cc @tfogal