From DeepSpeed to FSDP and Back Again with Hugging Face Accelerate
•
44
Stas Bekman
stas
AI & ML interests
Toolmaker. Software creator, optimizer and harmonizer.
Makes things work and fly at Contextual.AI
Training LLM/RAG/Generative AI/Machine Learning/Scalability
Recent Activity
updated
a model
about 1 month ago
stas/ml-engineering-book
Articles
Organizations
stas's activity
updated
a
model
about 1 month ago
Space isn't working because there is a runtime error
1
#9 opened 2 months ago
by
stas
stas
upvoted
an
article
3 months ago
Article
Mixture of Experts Explained
•
197
Fix FileNotFoundError
3
#2 opened 3 months ago
by
lhoestq
lhoestq
updated
a
model
4 months ago
Casting Issue?
4
#40 opened 5 months ago
by
FelixLabelle
FelixLabelle
authored
a
paper
5 months ago
posted
an
update
5 months ago
Post
1082
The Universal Checkpointing paper is out! https://arxiv.org/abs/2406.18820
If you remember the Bigscience BLOOM-176B training, Tunji Ruwase and I co-invented this technology for Megatron-Deepspeed in order to enable to quickly scale up and down node topology while continuing training.
Since then the DeepSpeed team continued improving on that and it has now been fully integrated into Deepspeed.
The blog post is here: https://github.com/microsoft/DeepSpeed/blob/master/blogs/deepspeed-ucp/README.md
If you remember the Bigscience BLOOM-176B training, Tunji Ruwase and I co-invented this technology for Megatron-Deepspeed in order to enable to quickly scale up and down node topology while continuing training.
Since then the DeepSpeed team continued improving on that and it has now been fully integrated into Deepspeed.
The blog post is here: https://github.com/microsoft/DeepSpeed/blob/master/blogs/deepspeed-ucp/README.md
upvoted
an
article
5 months ago
Article
From DeepSpeed to FSDP and Back Again with Hugging Face Accelerate
•
44
upvoted
an
article
7 months ago
Article
Introducing Idefics2: A Powerful 8B Vision-Language Model for the community
•
166
Upload book cover
1
#1 opened 8 months ago
by
julien-c
julien-c
metadata: set license
1
#2 opened 8 months ago
by
julien-c
julien-c
Reacted to
their
post with 🤗
8 months ago
Post
A combined effort from the IBM + Pytorch teams achieved an incredible training performance with ZeRO/FSDP on par with 3D parallelism on H100s, while having just 800Gbps inter-node connection.
This is because they got an almost full overlap between comms and compute and have introduced a novel selective activation recomputation method which recalculates only large but inexpensive activations.
Check out their post here: https://pytorch.org/blog/maximizing-training/
This is because they got an almost full overlap between comms and compute and have introduced a novel selective activation recomputation method which recalculates only large but inexpensive activations.
Check out their post here: https://pytorch.org/blog/maximizing-training/
posted
an
update
9 months ago
Post
A combined effort from the IBM + Pytorch teams achieved an incredible training performance with ZeRO/FSDP on par with 3D parallelism on H100s, while having just 800Gbps inter-node connection.
This is because they got an almost full overlap between comms and compute and have introduced a novel selective activation recomputation method which recalculates only large but inexpensive activations.
Check out their post here: https://pytorch.org/blog/maximizing-training/
This is because they got an almost full overlap between comms and compute and have introduced a novel selective activation recomputation method which recalculates only large but inexpensive activations.
Check out their post here: https://pytorch.org/blog/maximizing-training/
posted
an
update
9 months ago
Post
If you're trying to run MoE Mixtral-8x7b under DeepSpeed w/ HF Transformers it's likely to hang on the first forward.
The solution is here https://github.com/microsoft/DeepSpeed/pull/4966?_x_tr_sl=auto&_x_tr_tl=en&_x_tr_hl=en-US#issuecomment-1989671378
and you need deepspeed>=0.13.0
Thanks to Masahiro Tanaka for the fix.
The solution is here https://github.com/microsoft/DeepSpeed/pull/4966?_x_tr_sl=auto&_x_tr_tl=en&_x_tr_hl=en-US#issuecomment-1989671378
and you need deepspeed>=0.13.0
Thanks to Masahiro Tanaka for the fix.
replied to
their
post
10 months ago
I pinged Elio to see if he wants to join.
posted
an
update
10 months ago
Post
Hear, hear, AMD MI300Xs have started to emerge much sooner than expected.
Here is a 2-part benchmarks report on performing BLOOM-176B inference using @MSFTDeepSpeed optimized for AMD MI300X.
1. https://www.evp.cloud/post/diving-deeper-insights-from-our-llm-inference-testing
2. https://www.evp.cloud/post/diving-deeper-insights-from-our-llm-inference-testing-part-2
This was published in response to our BLOOM-176B super-fast inference blog post https://huggingface.co/blog/bloom-inference-pytorch-scripts
Note that these have 192GB of HBM!
The NVIDIA monopoly is strong, but it'll have to start sharing the pie and hopefully drive the costs down at least somewhat.
Thanks to https://www.linkedin.com/in/eliovp for sharing this writeup with me.
p.s. at the PyTorch conference in the fall, the AMD representative said we will see MI300X available to us mortals in Q4-2024/Q1-2025.
Here is a 2-part benchmarks report on performing BLOOM-176B inference using @MSFTDeepSpeed optimized for AMD MI300X.
1. https://www.evp.cloud/post/diving-deeper-insights-from-our-llm-inference-testing
2. https://www.evp.cloud/post/diving-deeper-insights-from-our-llm-inference-testing-part-2
This was published in response to our BLOOM-176B super-fast inference blog post https://huggingface.co/blog/bloom-inference-pytorch-scripts
Note that these have 192GB of HBM!
The NVIDIA monopoly is strong, but it'll have to start sharing the pie and hopefully drive the costs down at least somewhat.
Thanks to https://www.linkedin.com/in/eliovp for sharing this writeup with me.
p.s. at the PyTorch conference in the fall, the AMD representative said we will see MI300X available to us mortals in Q4-2024/Q1-2025.
-
- 5 replies
authored
a
paper
10 months ago
replied to
their
post
10 months ago
Thank you for the kind words, Jeff!
We are still waiting for BLOOM v2.0 from HF!
posted
an
update
10 months ago
Post
"The Case for Co-Designing Model Architectures with Hardware"
This is a long overdue paper that we have started discussing back when training BLOOM-176.
Basically this paper tells you how to design your model's dimensions for an optimal training throughput.
Fantastic!
Yours truly contributed the SwiGLU section ;)
https://twitter.com/QuentinAnthon15/status/1752393989813375119
https://arxiv.org/abs/2401.14489
This is a long overdue paper that we have started discussing back when training BLOOM-176.
Basically this paper tells you how to design your model's dimensions for an optimal training throughput.
Fantastic!
Yours truly contributed the SwiGLU section ;)
https://twitter.com/QuentinAnthon15/status/1752393989813375119
https://arxiv.org/abs/2401.14489
-
- 2 replies