interweave: unified GPU matmul offload server (replaces ~20 prototypes)#30
Merged
Conversation
One canonical server for the Interweave heterogeneous-offload path, replacing the ~20 matmul_*.py prototypes. Dtype dispatch registry (F32/F16/Q8_0/Q4_K/ MXFP4) with correct ggml block sizes (144/34/17, fixes the 148-byte Q4_K drift), wire contract pinned to the real POWER8 client (row-major, F32 default, magic-discriminated single and batch), input validation with row-width block alignment and F32 working-set caps, per-recv socket timeout, GPU serialization lock, little-endian guard, localhost-bound health and metrics sidecar, and graceful CPU/NumPy fallback. Ships a CPU-runnable pytest suite (golden dequant oracles, all 5 dtypes end to end, persistent multi-op, DoS rejection, health HTTP) plus a GPU/CPU parity test. All 14 tests pass on the C4130 deployment box. Old prototype files on the deployment box stay in place until a client-confirmed cutover. Signed-off-by: Scott <scottbphone12@gmail.com>
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
One canonical matmul offload server for the Interweave path. This is the code that has been running the POWER8 to C4130 offload, cleaned up into a single file with a real test suite instead of the ~20 matmul_*.py prototypes that grew on the deployment box.
What it does:
Testing: 14 tests, CPU-runnable (golden dequant oracles, all 5 dtypes end to end, persistent multi-op, DoS rejection, health endpoint) plus a GPU/CPU parity test. All pass on the C4130 (V100 + M40) as of today.