qbcompiler.configs.presets¶
Auto-generated preset definitions from config_schema.yaml.
Attributes¶
| Name | Description |
|---|---|
PRESETS |
PRESETS¶
PRESETS: Dict[str, dict] = {
"classification": {
"description": "Image classification models (ResNet, EfficientNet, ViT, etc.)",
"config": {"calibration": {"mode": 1, "output": 0}},
},
"detection": {
"description": "Object detection models (YOLO, SSD, DETR, etc.)",
"config": {"calibration": {"mode": 1, "output": 1}},
},
"classification_torchvision": {
"description": "Torchvision classification models with standard preprocessing",
"extends": "classification",
"config": {
"uint8Input": {"apply": True, "inputs": []},
"imageChannels": 3,
"preprocessing": {
"apply": True,
"autoConvertFormat": True,
"pipeline": [
{"op": "resize", "size": 256, "mode": "bilinear", "backend": "pil"},
{"op": "centerCrop", "height": 224, "width": 224},
{
"op": "normalize",
"mean": [0.485, 0.456, 0.406],
"std": [0.229, 0.224, 0.225],
"scaleToUint8": True,
"fuseIntoFirstLayer": True,
},
],
"inputConfigs": {},
},
},
},
"yolo_640": {
"description": "YOLO detection models with 640x640 letterbox preprocessing",
"extends": "detection",
"config": {
"uint8Input": {"apply": True, "inputs": []},
"imageChannels": 3,
"preprocessing": {
"apply": True,
"autoConvertFormat": True,
"pipeline": [
{
"op": "letterbox",
"height": 640,
"width": 640,
"padValue": 114,
"backend": "opencv",
}
],
"inputConfigs": {},
},
},
},
"yolo_1280": {
"description": "YOLO detection models with 1280x1280 letterbox preprocessing",
"extends": "detection",
"config": {
"uint8Input": {"apply": True, "inputs": []},
"imageChannels": 3,
"preprocessing": {
"apply": True,
"autoConvertFormat": True,
"pipeline": [
{
"op": "letterbox",
"height": 1280,
"width": 1280,
"padValue": 114,
"backend": "opencv",
}
],
"inputConfigs": {},
},
},
},
"llm": {
"description": "Large Language Models (LLaMA, Qwen, Gemma, etc.)",
"config": {
"equivalentTransformation": {
"QK": {"apply": True},
"UD": {"apply": True},
"VO": {"apply": True},
"SpinR1": {"apply": True},
"SpinR2": {"apply": True},
"OptimizeFFN": {"apply": True},
},
"llm": {
"apply": True,
"attributes": {
"maxSequenceLength": 4096,
"maxCacheLength": 4096,
"calibration": {"useFullSeqLength": True},
},
},
"calibration": {"mode": 0, "output": 0},
},
},
"llm_fast": {
"description": "LLM with faster compilation (less accuracy optimization)",
"extends": "llm",
"config": {
"equivalentTransformation": {
"QK": {"apply": False},
"UD": {"apply": False},
"VO": {"apply": False},
"SpinR1": {"apply": False},
"SpinR2": {"apply": False},
"OptimizeFFN": {"apply": False},
},
"llm": {
"apply": True,
"attributes": {"calibration": {"useFullSeqLength": False}},
},
},
},
"vision_transformer": {
"description": "Vision Transformer models (ViT, DeiT, Swin, etc.)",
"config": {
"calibration": {"method": 1, "mode": 0},
"bit": {"transformer": {"activation": {"output": 16, "ffn": 16}}},
},
},
"multimodal": {
"description": "Multimodal models (CLIP, BLIP, LLaVA, etc.)",
"config": {"calibration": {"method": 3}, "llm": {"apply": True}},
},
}
Preset¶
Preset utility class.
Methods¶
| Name | Description |
|---|---|
list |
List available preset names. |
get |
Get a preset by name. |
describe |
Get preset description. |
list¶
List available preset names.
get¶
Get a preset by name.
describe¶
Get preset description.
Functions¶
| Name | Description |
|---|---|
list_presets |
List available preset names. |
get_preset |
Get a CompileConfig from a preset name. |
list_presets¶
List available preset names.
get_preset¶
Get a CompileConfig from a preset name.