FIX OOM converting large models on 24Gb cards#29
Open
FearL0rd wants to merge 3 commits intosilveroxides:mainfrom
Open
FIX OOM converting large models on 24Gb cards#29FearL0rd wants to merge 3 commits intosilveroxides:mainfrom
FearL0rd wants to merge 3 commits intosilveroxides:mainfrom
Conversation
fix OOM GPU Memory Error During Model Quantization
Optimizes the quantization pipeline to handle 1B+ element layers (e.g., Gemma 3 embeddings) on 24GB hardware. Resolves consecutive CUDA OOMs by dynamically switching processing devices based on layer volume. Introduced mem_threshold to trigger full CPU optimization for massive layers. Implemented lazy dequantization in _compute_loss_and_grad to avoid GPU temporary tensor overhead. Migrated optimization loops to use in-place arithmetic (div_, add_, mul_). Added proactive CUDA cache flushing and state offloading to CPU.
Fixed a lifecycle bug in _optimize_original where W_float32 was accessed after deletion during subspace optimization. Reordered operations to ensure the initial rounded state is captured before VRAM liberation. Streamlined _optimize_original to calculate P_orig and W_q_refined consecutively before deleting original FP32 weights. Optimized subspace updates using torch.addmm_ for zero-overhead matrix modifications. Fixed variable scoping for is_massive logic to ensure safe execution on large embedding layers.
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Optimizes the quantization pipeline to handle 1B+ element layers (e.g., Gemma 3
embeddings) on 24GB hardware. Resolves consecutive CUDA OOMs by
dynamically switching processing devices based on layer volume.
Introduced mem_threshold to trigger full CPU optimization for massive layers.
Implemented lazy dequantization in _compute_loss_and_grad to avoid GPU
temporary tensor overhead.
Migrated optimization loops to use in-place arithmetic (div_, add_, mul_).
Added proactive CUDA cache flushing and state offloading to CPU.