Add ONNX export support for PatchTST
#2101
Open
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
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
xenova
commented
Nov 28, 2024
michaelbenayoun
approved these changes
Dec 12, 2024
echarlaix
approved these changes
Dec 13, 2024
Comment on lines
+260
to
+261
| "patchtst": "ibm/test-patchtst", | ||
| "patchtsmixer": "ibm/test-patchtsmixer", |
There was a problem hiding this comment.
already included in PYTORCH_EXPORT_MODELS_TINY so can be removed
Suggested change
| "patchtst": "ibm/test-patchtst", | |
| "patchtsmixer": "ibm/test-patchtsmixer", |
Comment on lines
+317
to
+320
| ("pt", "patchtsmixer", "feature-extraction"): ("transformers", "PatchTSMixerModel"), | ||
| ("pt", "patchtsmixer", "time-series-forecasting"): ("transformers", "PatchTSMixerForPrediction"), | ||
| ("pt", "patchtst", "feature-extraction"): ("transformers", "PatchTSTModel"), | ||
| ("pt", "patchtst", "time-series-forecasting"): ("transformers", "PatchTSTForPrediction"), |
There was a problem hiding this comment.
shouldn't AutoModel load the models as expected https://huggingface.co/datasets/huggingface/transformers-metadata/blob/main/pipeline_tags.json#L666-L667 ?
Suggested change
| ("pt", "patchtsmixer", "feature-extraction"): ("transformers", "PatchTSMixerModel"), | |
| ("pt", "patchtsmixer", "time-series-forecasting"): ("transformers", "PatchTSMixerForPrediction"), | |
| ("pt", "patchtst", "feature-extraction"): ("transformers", "PatchTSTModel"), | |
| ("pt", "patchtst", "time-series-forecasting"): ("transformers", "PatchTSTForPrediction"), | |
| ("pt", "patchtsmixer", "time-series-forecasting"): ("transformers", "PatchTSMixerForPrediction"), | |
| ("pt", "patchtst", "time-series-forecasting"): ("transformers", "PatchTSTForPrediction"), |
| @@ -177,6 +177,7 @@ class OnnxConfig(ExportConfig, ABC): | |||
| "text2text-generation": OrderedDict({"logits": {0: "batch_size", 1: "decoder_sequence_length"}}), | |||
| "text-classification": OrderedDict({"logits": {0: "batch_size"}}), | |||
| "text-generation": OrderedDict({"logits": {0: "batch_size", 1: "sequence_length"}}), | |||
| "time-series-forecasting": OrderedDict({"prediction_outputs": {0: "batch_size"}}), | |||
There was a problem hiding this comment.
I can't find "time-series-forecasting" in https://huggingface.co/datasets/huggingface/transformers-metadata/blob/main/pipeline_tags.json, do you know if this specific to PatchTST models ? (-> PatchTSTXxxForPrediction)
| @@ -2445,3 +2446,51 @@ class EncoderDecoderOnnxConfig(EncoderDecoderBaseOnnxConfig): | |||
| NORMALIZED_CONFIG_CLASS = NormalizedEncoderDecoderConfig | |||
|
|
|||
| DEFAULT_ONNX_OPSET = 14 # uses SDPA in Transformers, hence opset>=14. | |||
|
|
|||
|
|
|||
| class PatchTSTDummyInputGenerator(DummyInputGenerator): | |||
There was a problem hiding this comment.
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.
What does this PR do?
This PR adds ONNX export support for PatchTST models, as well as fixes other models that contain the
torch.unfoldoperator, which is not supported bytorch.onnx.export. This is done by applying a set of default patching specs.Fixes huggingface/transformers.js#1047
Before submitting
Who can review?
@michaelbenayoun, @echarlaix