Skip to content

fix(build): honor --no-optimize and add multilingual similarity recipes #1370

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
ssss141414 wants to merge 2 commits into
base: main
Choose a base branch
from
Draft

fix(build): honor --no-optimize and add multilingual similarity recipes #1370

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
a4ad838
fix(build): honor no-optimize in CLI pipelines
ssss141414 Aug 26, 2026
3cd2226
test(build): prove compile dataflow with no optimize
ssss141414 Aug 28, 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
79 changes: 79 additions & 0 deletions ...dell-research-harvard_lt-wikidata-comp-multi/cpu/cpu/sentence-similarity_fp16_config.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,79 @@
{
"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,
250002
]
},
{
"name": "attention_mask",
"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": "dell-research-harvard/lt-wikidata-comp-multi",
"model_type": "xlm-roberta",
"fp16_keep_io_types": true,
"fp16_op_block_list": null
},
"compile": null,
"loader": {
"task": "sentence-similarity",
"model_class": "AutoModel",
"model_type": "xlm-roberta"
}
}
57 changes: 57 additions & 0 deletions ...dell-research-harvard_lt-wikidata-comp-multi/cpu/cpu/sentence-similarity_fp32_config.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,57 @@
{
"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,
250002
]
},
{
"name": "attention_mask",
"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": "xlm-roberta"
}
}
54 changes: 29 additions & 25 deletions src/winml/modelkit/commands/build.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2068,6 +2068,7 @@ def _build_hf_pipeline(

max_iters: int = extra_kwargs.pop("hack_max_optim_iterations", 3)
allow_unsupported_nodes: bool = extra_kwargs.pop("allow_unsupported_nodes", False)
skip_optimize = extra_kwargs.pop("skip_optimize", False) or config.skip_optimize
model_label = model_id or "random-init"

# ── Validate + setup ─────────────────────────────────────────
Expand Down Expand Up @@ -2142,18 +2143,19 @@ def _name(base: str) -> str:
stage_timings.append(("Export", _export_elapsed))

# ── Optimize stage ───────────────────────────────────────────
current_path, _ = _run_optimize_stage(
config=config,
model_path=current_path,
optimized_path=optimized_path,
ep=ep,
device=device,
max_iters=max_iters,
stage_timings=stage_timings,
show_io_first=False,
analyze_output_path=analyze_result_path,
allow_unsupported_nodes=allow_unsupported_nodes,
)
if not skip_optimize:
current_path, _ = _run_optimize_stage(
config=config,
model_path=current_path,
optimized_path=optimized_path,
ep=ep,
device=device,
max_iters=max_iters,
stage_timings=stage_timings,
show_io_first=False,
analyze_output_path=analyze_result_path,
allow_unsupported_nodes=allow_unsupported_nodes,
)

# Persist config after autoconf
config_path.write_text(json.dumps(config.to_dict(), indent=2))
Expand Down Expand Up @@ -2209,6 +2211,7 @@ def _build_onnx_pipeline(

max_iters: int = extra_kwargs.pop("hack_max_optim_iterations", 3)
allow_unsupported_nodes: bool = extra_kwargs.pop("allow_unsupported_nodes", False)
skip_optimize = extra_kwargs.pop("skip_optimize", False)

# ── Validate + setup ─────────────────────────────────────────
if not onnx_path.exists():
Expand Down Expand Up @@ -2251,21 +2254,22 @@ def _build_onnx_pipeline(
# config before any stage reads it, otherwise the optimize stage will still
# run on integer ops and the quantize stage may try to re-quantize.
ensure_pre_quantized_stamped(config, current_path)
skip_optimize = skip_optimize or config.skip_optimize

# ── Optimize stage (first stage for ONNX — show I/O here) ────
current_path, _ = _run_optimize_stage(
config=config,
model_path=current_path,
optimized_path=optimized_path,
ep=ep,
device=device,
max_iters=max_iters,
stage_timings=stage_timings,
show_io_first=True,
analyze_output_path=analyze_result_path,
allow_unsupported_nodes=allow_unsupported_nodes,
skip_optimize=config.skip_optimize,
)
if not skip_optimize:
current_path, _ = _run_optimize_stage(
config=config,
model_path=current_path,
optimized_path=optimized_path,
ep=ep,
device=device,
max_iters=max_iters,
stage_timings=stage_timings,
show_io_first=True,
analyze_output_path=analyze_result_path,
allow_unsupported_nodes=allow_unsupported_nodes,
)

config_path.write_text(json.dumps(config.to_dict(), indent=2))

Expand Down
Loading
Loading