Allocate managed memory if device memory runs out #709
+31
−5
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
karpathy
reviewed
Aug 16, 2024
| @@ -393,13 +393,13 @@ void gpt2_allocate_state(GPT2 *model, int B, int T) { | |||
| printf0("allocating %zu MiB for AdamW optimizer state v\n", (shard_num_parameters * sizeof(float)) >> 20); | |||
| assert(model->m_memory == nullptr); | |||
| assert(model->v_memory == nullptr); | |||
| cudaCheck(cudaMalloc((void**)&model->m_memory, shard_num_parameters * sizeof(float))); | |||
| cudaCheck(cudaMalloc((void**)&model->v_memory, shard_num_parameters * sizeof(float))); | |||
| cudaMallocConditionallyManaged((void**)&model->m_memory, shard_num_parameters * sizeof(float)); | |||
There was a problem hiding this comment.
imo we should try to update the import statements to show which file any function (e.g. the new Managed manaloc) comes from (here cuda_utils)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Use cudaMallocManaged to allocate optimizer states if we run out of device memory, so we can still train (slowly) even if we cannot fit the optimizer state
This is based on #694 , which should be merged first