Skip to content

recipe(multilingual-e5-small): add CPU fp32/fp16 feature-extraction and sentence-similarity recipes #1182

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
Qiong Wu (qiowu) (DingmaomaoBJTU) wants to merge 2 commits into
base: main
Choose a base branch
from
Draft

recipe(multilingual-e5-small): add CPU fp32/fp16 feature-extraction and sentence-similarity recipes #1182

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5f1e504
recipe(multilingual-e5-small): add CPU fp32/fp16 feature-extraction a…
DingmaomaoBJTU Jul 23, 2026
8e7c621
fix(inference): synthesize required token type ids
ssss141414 Aug 13, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
91 changes: 91 additions & 0 deletions examples/recipes/intfloat_multilingual-e5-small/cpu/cpu/feature-extraction_fp16_config.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,91 @@
{
"export": {
"opset_version": 17,
"batch_size": 1,
"export_params": true,
"do_constant_folding": true,
"verbose": false,
"dynamo": false,
"enable_hierarchy_tags": true,
"clean_onnx": false,
"hierarchy_tag_format": "full",
"input_tensors": [
{
"name": "input_ids",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
250037
]
},
{
"name": "attention_mask",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
2
]
},
{
"name": "token_type_ids",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
2
]
}
],
"output_tensors": [
{
"name": "last_hidden_state"
}
],
"compatibility": {
"transformers_attention": "eager"
}
},
"optim": {
"clamp_constant_values": true
},
"quant": {
"mode": "fp16",
"samples": 10,
"calibration_method": "minmax",
"weight_type": "uint8",
"activation_type": "uint8",
"per_channel": false,
"symmetric": false,
"weight_symmetric": null,
"activation_symmetric": null,
"save_calibration": false,
"distribution": "uniform",
"seed": null,
"calibration_load_path": null,
"calibration_save_path": null,
"op_types_to_quantize": null,
"nodes_to_exclude": null,
"task": "feature-extraction",
"model_id": "intfloat/multilingual-e5-small",
"model_type": "bert",
"fp16_keep_io_types": true,
"fp16_op_block_list": null
},
"compile": null,
"loader": {
"task": "feature-extraction",
"model_class": "AutoModel",
"model_type": "bert"
}
}
69 changes: 69 additions & 0 deletions examples/recipes/intfloat_multilingual-e5-small/cpu/cpu/feature-extraction_fp32_config.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,69 @@
{
"export": {
"opset_version": 17,
"batch_size": 1,
"export_params": true,
"do_constant_folding": true,
"verbose": false,
"dynamo": false,
"enable_hierarchy_tags": true,
"clean_onnx": false,
"hierarchy_tag_format": "full",
"input_tensors": [
{
"name": "input_ids",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
250037
]
},
{
"name": "attention_mask",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
2
]
},
{
"name": "token_type_ids",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
2
]
}
],
"output_tensors": [
{
"name": "last_hidden_state"
}
],
"compatibility": {
"transformers_attention": "eager"
}
},
"optim": {
"clamp_constant_values": true
},
"quant": null,
"compile": null,
"loader": {
"task": "feature-extraction",
"model_class": "AutoModel",
"model_type": "bert"
}
}
91 changes: 91 additions & 0 deletions examples/recipes/intfloat_multilingual-e5-small/cpu/cpu/sentence-similarity_fp16_config.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,91 @@
{
"export": {
"opset_version": 17,
"batch_size": 1,
"export_params": true,
"do_constant_folding": true,
"verbose": false,
"dynamo": false,
"enable_hierarchy_tags": true,
"clean_onnx": false,
"hierarchy_tag_format": "full",
"input_tensors": [
{
"name": "input_ids",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
250037
]
},
{
"name": "attention_mask",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
2
]
},
{
"name": "token_type_ids",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
2
]
}
],
"output_tensors": [
{
"name": "last_hidden_state"
}
],
"compatibility": {
"transformers_attention": "eager"
}
},
"optim": {
"clamp_constant_values": true
},
"quant": {
"mode": "fp16",
"samples": 10,
"calibration_method": "minmax",
"weight_type": "uint8",
"activation_type": "uint8",
"per_channel": false,
"symmetric": false,
"weight_symmetric": null,
"activation_symmetric": null,
"save_calibration": false,
"distribution": "uniform",
"seed": null,
"calibration_load_path": null,
"calibration_save_path": null,
"op_types_to_quantize": null,
"nodes_to_exclude": null,
"task": "sentence-similarity",
"model_id": "intfloat/multilingual-e5-small",
"model_type": "bert",
"fp16_keep_io_types": true,
"fp16_op_block_list": null
},
"compile": null,
"loader": {
"task": "sentence-similarity",
"model_class": "AutoModel",
"model_type": "bert"
}
}
69 changes: 69 additions & 0 deletions examples/recipes/intfloat_multilingual-e5-small/cpu/cpu/sentence-similarity_fp32_config.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,69 @@
{
"export": {
"opset_version": 17,
"batch_size": 1,
"export_params": true,
"do_constant_folding": true,
"verbose": false,
"dynamo": false,
"enable_hierarchy_tags": true,
"clean_onnx": false,
"hierarchy_tag_format": "full",
"input_tensors": [
{
"name": "input_ids",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
250037
]
},
{
"name": "attention_mask",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
2
]
},
{
"name": "token_type_ids",
"dtype": "int32",
"shape": [
1,
512
],
"value_range": [
0,
2
]
}
],
"output_tensors": [
{
"name": "last_hidden_state"
}
],
"compatibility": {
"transformers_attention": "eager"
}
},
"optim": {
"clamp_constant_values": true
},
"quant": null,
"compile": null,
"loader": {
"task": "sentence-similarity",
"model_class": "AutoModel",
"model_type": "bert"
}
}
27 changes: 26 additions & 1 deletion src/winml/modelkit/models/winml/feature_extraction.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@
from collections import OrderedDict
from typing import Any

import numpy as np
from transformers.utils.generic import ModelOutput

from .base import WinMLPreTrainedModel
Expand Down Expand Up @@ -58,7 +59,31 @@ def forward(self, **kwargs: Any) -> ModelOutput:
order. Tensors keep their native rank (no unsqueeze); downstream
pooling handles 1-D and 2-D after raw[0].
"""
outputs = self._run_inference(self._format_inputs(**kwargs))
inputs = self._format_inputs(**kwargs)
if "token_type_ids" not in inputs and "input_ids" in inputs:
input_names = self.io_config.get("input_names", [])
if "token_type_ids" in input_names:
input_index = input_names.index("token_type_ids")
input_types = self.io_config.get("input_types", [])
input_shapes = self.io_config.get("input_shapes", [])
if input_index < len(input_types) and input_index < len(input_shapes):
required_shape = input_shapes[input_index]
actual_shape = inputs["input_ids"].shape
shape_matches = len(required_shape) == len(actual_shape) and all(
not isinstance(dimension, int)
or dimension <= 0
or dimension == actual_dimension
for dimension, actual_dimension in zip(
required_shape, actual_shape, strict=True
)
)
if shape_matches:
inputs["token_type_ids"] = np.zeros(
actual_shape,
dtype=np.dtype(input_types[input_index]),
)

outputs = self._run_inference(inputs)
# WinMLEncoderDecoderModel expects its encoder sub-component to expose
# hidden states as "last_hidden_state". Alias the primary output when an
# encoder ONNX graph named it otherwise (e.g. "encoder_hidden_states").
Expand Down
Loading
Loading