Comments (3)
Hi @muellerzr, thanks for the response!
For the experiments above, I have already disabled the learning rate scheduler.
In addition, I have tried adjust the learning rate according to learning_rate *= accelerator.num_processes
given in the official performance guideline. I still see a significant difference in the training performance.
FYI, here is the result after using learning_rate *= 4
when training with 4 GPUs:
(shadow) bxiao@ip-10-45-101-134:/sensei-fs/users/bxiao/test_multiGPUs$ accelerate launch --config_file config.yaml ./cv_example.py --data_dir ./images
The following values were not passed to `accelerate launch` and had defaults used instead:
`--dynamo_backend` was set to a value of `'no'`
To avoid this warning pass in values for each of the problematic parameters or run `accelerate config`.
0.17.1
0.17.1
0.17.1
0.17.1
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1478/1478 [00:35<00:00, 41.07it/s]
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 370/370 [00:10<00:00, 35.43it/s]
epoch 0: 75.24
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1478/1478 [00:34<00:00, 42.35it/s]
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 370/370 [00:10<00:00, 36.49it/s]
epoch 1: 76.52
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1478/1478 [00:34<00:00, 42.67it/s]
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 370/370 [00:10<00:00, 35.66it/s]
epoch 2: 77.33
from accelerate.
Have you also tried scaling the learning rate according to the multiple GPUs? (What I mean by this is in multi-GPU the scheduler is stepped N times, which could account for some of this)
from accelerate.
Thanks, let me try running this today and see what happens
from accelerate.
Related Issues (20)
- Multiple processes on CPUs
- 4-bit quantization cannot load weights to meta device for bias terms of the linear layer: NotImplementedError: Cannot copy out of meta tensor; no data!
- SageMaker config stopped working in Accelerate version 0.29+ HOT 1
- Does Accelerate support tensor parallelism?
- can't set `enable_cpu_affinity` through accelerate config in version 0.31.0.dev0
- PicklingError: Can't pickle <function Embedding.forward at XXXXXXX> it's not the same object as torch.nn.modules.sparse.Embedding.forward HOT 5
- Duplicate elements in `split_between_processes` HOT 4
- How to specify the backend of Trainer
- Cuda Out of memory while loading PEFT weights using accelerate on multi gpu
- Accelerate 0.30.0 Breaks FSDP QLora HOT 5
- Accelerate FSDP RuntimeError: Tensors of the same index must be on the same device and the same dtype
- minor typo in Accelerator docs
- Multi GPU process stuck in the dataloader loop, single GPU works fine. HOT 7
- accelerate test does not work when using 3 GPUs
- pydantic.errors.PydanticUserError: A non-annotated attribute was detected: `is_crew_class = True` HOT 1
- unexpected lr scheduler behavior when using accelerate HOT 8
- Error while trying to fine-tune using FSDP on a TPU HOT 2
- Unable to load mistralai/Mixtral-8x7B-Instruct-v0.1 using mps HOT 5
- Unable to launch DeepSpeed multinode training with a heterogenous mix of # devices per node. HOT 5
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 accelerate.