Conversation
Add _ATATEmbedderMulti wrapper and run_export_multi() that emit all three aggregation variants (token, mean, full) as named outputs in a single file. Add benchmark_multi.py comparing the single-file approach to running three separate ONNX files. Results on CPU (batch=4): - File size: 4.6 MB vs 13.8 MB for three files (3.0x smaller) - Latency: 104 ms vs 252 ms sequential (2.4x faster) - Outputs are numerically identical (max diff = 0)
Compare each single-output file against requesting only that output from the multi-output file. onnxruntime prunes unused outputs, so per-variant latency is essentially the same (within ±5% noise), while the multi-output file is 3x smaller on disk.
Exports the Informer photo encoder from AstroMLCore/AstroM3-CLIP-photo as a single multi-output ONNX file (mean, max, sequence at d_model=128). ProbSparse attention is replaced with standard SDPA for ONNX compatibility, following the same approach used for ATCAT. Weights are loaded from the standalone HuggingFace checkpoint (classification head silently ignored via strict=False). Test data is sourced from AstroMLCore/AstroM3Processed (full_0 split).
- Remove max pooling (not motivated by Informer design — no training signal for max to be meaningful) - Add validate.py comparing patched PyTorch vs ONNX (max abs diff ~2e-6) - Note CC-BY-4.0 weights license in README and model card frontmatter - Move _CLASS_NAMES to test_data.py (only used there) - Add onnxruntime to project deps (needed by validate)
…x weight/dataset description
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fix #1