Comments (6)
@kshitij12345 - From our discussion, I remember you were looking into this. I believe this is what is causing the memory operations but needs further investigation.
from lightning-thunder.
cc @eqy re: fragmentation lunch discussion
from lightning-thunder.
Does TORCH_NCCL_AVOID_RECORD_STREAMS=1
help?
from lightning-thunder.
@eqy Yes, it does. Either of the two env variables give the same performance benefit. Is this fair to call this a memory fragmentation issue or is this something else you think?
from lightning-thunder.
As per offline discussion with @ptrblck , we should enable TORCH_NCCL_AVOID_RECORD_STREAMS=1
by default in thunder
.
from lightning-thunder.
cc - @IvanYashchuk @mruberry Can we enable this env var by default in Thunder or should we rely on nvidia containers do enable this?
from lightning-thunder.
Related Issues (20)
- Recursion error in transformer module with NeMo Stable Diffusion HOT 3
- Hang using thunder.jit with tokenizer in NeMo Stable Diffusion HOT 5
- Constraints to insert static numbers
- CI: Re-Enable torchrun call in Zero to Thunder notebook
- dtype inconsistencies when dividing/rounding tensors
- thunder.jit of AutoEncoder in NeMo Stable Diffusion slower than eager HOT 4
- Dynamic shape needs to be modeled in trace
- OOM errors for Gemma-7, pythia-12b, Llama-2-13b-hf and Nous-Hermes-13b with FSDP zero3 and 2x8 H100 HOT 1
- Refine recording of source locations HOT 5
- Nous-Hermes-13b on 1x8 H100 FSDP zero2 with thunder_cudnn is 23% slower than with inductor
- fsdp(jit(...)) transform can use more memory compared to jit(fsdp(...)) HOT 1
- nvfuserex has problems taking getitem. HOT 3
- load/save_state_dict hooks for early transforms
- Training Llama-2-13b-hf on 2x8 H100 with Thunder inductor is 47% slower than with Inductor
- FP8 Linear and conv with cudnn HOT 1
- Support RN50 BatchNorm fusions with cudnn
- CI : PyTorch nightly CI failing with `FutureWarning: is_compiling is deprecated. Use torch.compiler.is_compiling() instead.`
- Distill API for module transformations from distributed / quantization uses of ThunderModule attributes
- TransformerEngine API changed and caused test failure `AttributeError: 'TELinear' object has no attribute 'fp8_weight_shapes'`
- FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from lightning-thunder.