Comments (2)
loadams
commented on May 29, 2024
Hi @yiakwy-xpu-ml-framework-team - can you share the error that you hit with this? This isn't just logging that is inside the if statement here.
from megatron-deepspeed.
yiakwy-xpu-ml-framework-team
commented on May 29, 2024
Hi @loadams we believe hat bf16 has been already supported (NVIDIA/apex#1178) and no need to converted to float without being detached from main graph.
The function is for stat purpose hence the cast op (bf16 -> fp32) extends unnecessary memory liveness and add addition memory copy of the tensor, which increase memory usage.
The minimal reproducer could be launching a mini GPT trainer and verify wether memory drops for a certain portion.
from megatron-deepspeed.
Related Issues (20)
- [QUESTION] How can I finetune a model without zero_pp_rank_*_mp_rank_00_optim_states.pt
- Error message in checkpoint_{loader, saver}_megatron.py is usually a red herring
- load_args_from_checkpoint fails if arguments are missing
- Some issues in the Twin-Flow feature provided by Zero-offload ++ HOT 1
- Support for distributed finetuning
- Question about order of transpose operation.
- Support TransformerEngine HOT 1
- The pile and its BookCorpus subset are not available HOT 2
- Will offline++be forcibly uninstalled after specifying a ratio? HOT 1
- ImportError: /mnt/afs/hzy/Megatron-DeepSpeed/megatron/fused_kernels/build/scaled_upper_triang_masked_softmax_cuda.so HOT 2
- error in generate_text.sh HOT 1
- bos_token_id is assigned to be eos_token_id in _HFTokenizer HOT 1
- how to convert deepspeed model to megatron, when pp=2, tp=2, nnode=2 HOT 1
- Doubts about GPU memory
- [QUESTION]how to use' nsight compute' to profile 'pretrain_llama2_distributed.sh' in the 'examples_deepspeed' folder ?
- it seems there is a version problem please help me
- [QUESTION] Does the dev team have a plan to merge Mega-LM 0.4?
- Error set num-experts>1 when running the generate_test.sh
- 2nodes, 4 gpu, tp=2,pp=2, timeout
- The link (https://the-eye.eu/public/AI/pile_neox/data/BookCorpusDataset_text_document.bin ) has expired. HOT 1
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 megatron-deepspeed.