Comments (4)
Hi there, I have met the exactly same question, do you find a way to solve it? Now I have to copy the *.npy files manually twice from node 0 to node 1 when I run a new model.
from megatron-lm.
I think the system assumes the cluster is running with a shared file system like lustre.
from megatron-lm.
Hi there, I have met the exactly same question, do you find a way to solve it? Now I have to copy the *.npy files manually twice from node 0 to node 1 when I run a new model.
Did you ever find answer to this problem? Thanks
from megatron-lm.
Hi there, I have met the exactly same question, do you find a way to solve it? Now I have to copy the *.npy files manually twice from node 0 to node 1 when I run a new model.
Did you ever find answer to this problem? Thanks
If you don't plan to use a file sharing system like lustre, you'd have to manually copy and paste the dataset cache to all your nodes and comment out the self.unique_description_hash in megatron_dataset.py, otherwise it will try to load a new set of .npy files every time. Hope it helps!
from megatron-lm.
Related Issues (20)
- [BUG] Permormance drop while training with MoE HOT 6
- Is this assertion correct? HOT 5
- why Megatron's ParallelAttention currently only supports SelfAttention with a 'causal' MaskType?
- [BUG] Inference for multiple nodes not working. HOT 1
- [BUG] "referenced before assignment" when training MOE+PP HOT 3
- [BUG] Sequence Parallel is not All-gathered when using Flash Attention? HOT 1
- [QUESTION]why f and g must conjucates each other?
- [QUESTION] Why `--overlap-param-gather` is supported only with mcore models? HOT 2
- checkpoint_util.py HOT 1
- [BUG] NCCL TIMEOUT ( maybe ALLREDUCE ? ) HOT 1
- [QUESTION] why the time of one iter in nsys longer than that in the ouput log?
- [BUG] cross-entropy loss not computed correctly when label_smoothing is enabled
- [ENHANCEMENT] How to install this package with editable mode? HOT 1
- [QUESTION] Why take too much time to sync up barrier information between ranks
- [BUG]
- [QUESTION] In RotaryEmbedding, the datatype of inv_freq and the corresponding sin/cos computations should be maintained as torch.float32?
- [QUESTION]Why does Megatron-LM using gloo backend when Creating Parrallel Group ? HOT 1
- [BUG] ModuleNotFoundError: No module named 'scaled_softmax_cuda' HOT 4
- [QUESTION] why pipeline-model-parallel size should be greater than 2 with interleaved schedule ? HOT 4
- [QUESTION] attention_mask is None in middle pipeline stage? HOT 2
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-lm.