Fix hgq device bug - #80
Open
MartinaJoergensen wants to merge 3 commits into
Open
Conversation
merge to main for 0.1.0
HGQQuantizer.build() read device = self._k.device, a stale CPU placeholder set at __init__ time, instead of the input tensor's real device. Since the module is built lazily on the first forward pass -- after the model's own .to(device) call already ran -- new bit-width parameters/buffers always landed on CPU, causing: RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! Fix: thread the real device through build() from x.device at the forward() call site instead. Applied across hgq_quantizer.py, activations.py, and layers.py, plus matching test updates. Verified working via profile_hgq_memory.py and a full 3-epoch end-to-end training run (pquant-hgq-crashtest-accum-cwfix). See cern-nextgen#78.
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
3 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
HGQQuantizer.build()readdevice = self._k.device, a stale CPU placeholder set at__init__time, instead of the input tensor's real device. Since the module is built on the first forward pass, after the model's own.to(device)call already ran, new bit-width parameters/buffers always landed on CPU, causing:RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!Fix: thread the real device through
build()fromx.deviceat theforward()call site instead. Applied acrosshgq_quantizer.py,activations.py, andlayers.py, plus matching test updates.Verified working via
profile_hgq_memory.pyand a full 3-epoch end-to-end training runSee #78.