Migrate existing python code from Pytorch DDP to using Huggingface Accelerate
Budget: $10 – $30 AUD
I'm in need of someone to convert my existing Pytorch DDP code to Huggingface Accelerate framework.
Key Project Details:
- The primary goal of this migration is to move from existing Pytorch DDP, to Huggingface Accelerate architecture.
- The code should be able to run on CPU-only mode, on a single GPU machine, and also on multiple-GPU machines. The GPU's are all Nvidia.
- The code will need to be tested on my machines (CPU-only machine; machine with single-GPU; and machine with multiple GPUs) and execute smoothly without error using my datasets. If the code does not execute out of the box, you will need to log in to my computer to set it up to ensure smooth running. Pytorch environment under PyCharm/Ubuntu is already set up.
These two links should give you an idea of what Huggingface Accelerate is:
https://huggingface.co/blog/pytorch-ddp-accelerate-transformers
https://wandb.ai/wandb_fc/pytorch-image-models/reports/An-Introduction-to-HuggingFace-s-Accelerate-Library--Vmlldzo2MzgzNzA
Ideal Candidate:
- Proficient in Pytorch and Huggingface Accelerate framework.
- Previous experience in migrating distributed training code is highly desirable.
- Understanding of model and data parallelism concepts.
- Excellent problem-solving skills and attention to detail.
Start now.
Time frame: 1 week.
Budget: $25
Key Project Details:
- The primary goal of this migration is to move from existing Pytorch DDP, to Huggingface Accelerate architecture.
- The code should be able to run on CPU-only mode, on a single GPU machine, and also on multiple-GPU machines. The GPU's are all Nvidia.
- The code will need to be tested on my machines (CPU-only machine; machine with single-GPU; and machine with multiple GPUs) and execute smoothly without error using my datasets. If the code does not execute out of the box, you will need to log in to my computer to set it up to ensure smooth running. Pytorch environment under PyCharm/Ubuntu is already set up.
These two links should give you an idea of what Huggingface Accelerate is:
https://huggingface.co/blog/pytorch-ddp-accelerate-transformers
https://wandb.ai/wandb_fc/pytorch-image-models/reports/An-Introduction-to-HuggingFace-s-Accelerate-Library--Vmlldzo2MzgzNzA
Ideal Candidate:
- Proficient in Pytorch and Huggingface Accelerate framework.
- Previous experience in migrating distributed training code is highly desirable.
- Understanding of model and data parallelism concepts.
- Excellent problem-solving skills and attention to detail.
Start now.
Time frame: 1 week.
Budget: $25