[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78AY02DVFF3SP42YF7QB0QG":3,"skill-same:01M78AY02DVFF3SP42YF7QB0QG":118},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":31,"commitSha":32,"skillDirectory":33,"forksCount":14,"repositoryHtmlUrl":34,"repositoryHomepageUrl":35,"ownerHtmlUrl":36,"ownerBlogUrl":37,"ownerDescription":38,"updatedAt":39,"install":40,"resources":53},"01M78AY02DVFF3SP42YF7QB0QG","tao-train-oneformer",null,"OneFormer for universal image segmentation. Unifies panoptic, instance, and semantic segmentation with a single architecture using task-conditioned queries. Use when training, evaluating, exporting, quantizing, or running inference for a TAO OneFormer model. Trigger phrases include \"train OneFormer\", \"universal segmentation\", \"task-conditioned segmentation\", \"panoptic \u002F instance \u002F semantic in one model\".","Solizardking\u002Fskillhub-main","Solizardking","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F108096810?v=4","01M786KDB10AYV042N0F9BD5B5","Clawd","01M786NYYXSZK1D05WM8JQPC48",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: tao-train-oneformer\ndescription: OneFormer for universal image segmentation. Unifies panoptic, instance, and semantic segmentation with a\n  single architecture using task-conditioned queries. Use when training, evaluating, exporting, quantizing, or running\n  inference for a TAO OneFormer model. Trigger phrases include \"train OneFormer\", \"universal segmentation\",\n  \"task-conditioned segmentation\", \"panoptic \u002F instance \u002F semantic in one model\".\nlicense: Apache-2.0\ncompatibility: Requires docker + nvidia-container-toolkit.\nmetadata:\n  version: \"0.1.0\"\n  author: NVIDIA Corporation\nallowed-tools: Read Bash\ntags:\n- segmentation\n---\n\n# OneFormer\n\nOneFormer for universal image segmentation. Unifies panoptic, instance, and semantic segmentation with a single architecture using task-conditioned queries.\n\nSet train.pretrained_backbone and\u002For train.pretrained_model.\n\nFor TAO Deploy TensorRT actions (`gen_trt_engine`, TensorRT `evaluate`, and TensorRT `inference`), read `references\u002Ftao-deploy-oneformer.md` first. Deploy spec templates live in this skill's `references\u002F` folder with the `spec_template_deploy_*.yaml` prefix.\n\n## Dataclass Schemas\n\nGenerated TAO Core schemas are packaged in `schemas\u002F\u003Caction>.schema.json`, with `schemas\u002Fmanifest.json` listing available actions. Each generated schema also emits `references\u002Fspec_template_\u003Caction>.yaml` from the schema top-level `default` field. AutoML enablement is declared at the model layer in `references\u002Fskill_info.yaml` via `automl_enabled`. Runnable AutoML still requires `schemas\u002Ftrain.schema.json` and `references\u002Fspec_template_train.yaml` to exist and parse. Use the packaged train schema for `automl_default_parameters`, `automl_disabled_parameters`, defaults, min\u002Fmax bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect `~\u002Ftao-core` at runtime; maintainers regenerate schemas\u002Ftemplates before packaging the skill bank.\n\n## Train Action Policy\n\nThis model is AutoML-enabled at the model layer. Before handling any train-stage request, read `references\u002Fskill_info.yaml` and resolve the run override from either an explicit `automl_policy` value or the user's workflow request. Use `automl_policy: on` by default and only expose `on` \u002F `off` in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as `automl_policy: off` for this run only. When `automl_policy: on`, `automl_enabled: true`, and both `schemas\u002Ftrain.schema.json` and `references\u002Fspec_template_train.yaml` are packaged, route the train action through `tao-skill-bank:tao-run-automl` by default with this model's `skill_dir`. Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and `automl_policy`. Use direct model training only when `automl_policy: off` or the packaged train schema\u002Ftemplate is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.\n\nNon-train actions such as `evaluate`, `inference`, `export`, and deploy flows stay in this model skill. The per-run `automl_policy` override does not change model metadata.\n\n## Training Requirements\n\n- **Dataset type:** segmentation\n- **Formats:** coco_panoptic, coco\n- **Monitoring metric:** mIoU\n\n### Per-Action Dataset Requirements\n\n| Action | Spec Key | Source | Files | List? |\n|---|---|---|---|---|\n| evaluate | dataset.train.images | train_datasets | images.tar.gz | No |\n| evaluate | dataset.label_map | train_datasets | label_map.json | No |\n| evaluate | dataset.train.annotations | train_datasets | annotations.json | No |\n| evaluate | dataset.train.panoptic | train_datasets | images_panoptic.tar.gz | No |\n| evaluate | dataset.val.images | eval_dataset | images.tar.gz | No |\n| evaluate | dataset.val.annotations | eval_dataset | annotations.json | No |\n| evaluate | dataset.val.panoptic | eval_dataset | images_panoptic.tar.gz | No |\n| evaluate | dataset.test.images | eval_dataset | images.tar.gz | No |\n| evaluate | dataset.test.annotations | eval_dataset | annotations.json | No |\n| evaluate | dataset.test.panoptic | eval_dataset | images_panoptic.tar.gz | No |\n| inference | dataset.train.images | train_datasets | images.tar.gz | No |\n| inference | dataset.label_map | train_datasets | label_map.json | No |\n| inference | dataset.train.annotations | train_datasets | annotations.json | No |\n| inference | dataset.train.panoptic | train_datasets | images_panoptic.tar.gz | No |\n| inference | dataset.val.images | eval_dataset | images.tar.gz | No |\n| inference | dataset.val.annotations | eval_dataset | annotations.json | No |\n| inference | dataset.val.panoptic | eval_dataset | images_panoptic.tar.gz | No |\n| inference | dataset.test.images | inference_dataset | images.tar.gz | No |\n| quantize | dataset.train.images | train_datasets | images.tar.gz | No |\n| quantize | dataset.train.annotations | train_datasets | annotations.json | No |\n| quantize | dataset.label_map | train_datasets | label_map.json | No |\n| quantize | dataset.train.panoptic | train_datasets | images_panoptic.tar.gz | No |\n| quantize | dataset.val.images | eval_dataset | images.tar.gz | No |\n| quantize | dataset.val.annotations | eval_dataset | annotations.json | No |\n| quantize | dataset.val.panoptic | eval_dataset | images_panoptic.tar.gz | No |\n| quantize | dataset.test.images | eval_dataset | images.tar.gz | No |\n| quantize | dataset.quant_calibration_dataset.images_dir | calibration_dataset | images.tar.gz | No |\n| train | dataset.train.images | train_datasets | images.tar.gz | No |\n| train | dataset.train.annotations | train_datasets | annotations.json | No |\n| train | dataset.label_map | train_datasets | label_map.json | No |\n| train | dataset.train.panoptic | train_datasets | images_panoptic.tar.gz | No |\n| train | dataset.val.images | eval_dataset | images.tar.gz | No |\n| train | dataset.val.annotations | eval_dataset | annotations.json | No |\n| train | dataset.val.panoptic | eval_dataset | images_panoptic.tar.gz | No |\n| train | dataset.test.images | eval_dataset | images.tar.gz | No |\n\n### Typical Spec Overrides\n\nData source overrides are **mandatory for every action** — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in `spec_overrides`.\n\n```python\nS3_TRAIN = \"s3:\u002F\u002Fbucket\u002Fdata\u002Ftrain\"\nS3_EVAL = \"s3:\u002F\u002Fbucket\u002Fdata\u002Feval\"\nS3_INFERENCE = \"s3:\u002F\u002Fbucket\u002Fdata\u002Finference\"\nS3_CALIBRATION = \"s3:\u002F\u002Fbucket\u002Fdata\u002Fcalibration\"\n```\n\n**train (mandatory data sources):**\n```python\n{\n    \"train.num_gpus\": 1,\n    \"train.num_epochs\": 10,\n    \"train.checkpoint_interval\": 10,\n    \"train.validation_interval\": 10,\n    \"model.sem_seg_head.num_classes\": 133,\n    \"dataset.contiguous_id\": True,\n    \"train.precision\": \"32\",\n    \"dataset.train.images\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.train.annotations\": f\"{S3_TRAIN}\u002Fannotations.json\",\n    \"dataset.label_map\": f\"{S3_TRAIN}\u002Flabel_map.json\",\n    \"dataset.train.panoptic\": f\"{S3_TRAIN}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.val.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val.annotations\": f\"{S3_EVAL}\u002Fannotations.json\",\n    \"dataset.val.panoptic\": f\"{S3_EVAL}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.test.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n}\n```\n\n**evaluate (mandatory data sources):**\n```python\n{\n    \"evaluate.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"model.sem_seg_head.num_classes\": 133,\n    \"dataset.contiguous_id\": True,\n    \"dataset.train.images\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.label_map\": f\"{S3_TRAIN}\u002Flabel_map.json\",\n    \"dataset.train.annotations\": f\"{S3_TRAIN}\u002Fannotations.json\",\n    \"dataset.train.panoptic\": f\"{S3_TRAIN}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.val.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val.annotations\": f\"{S3_EVAL}\u002Fannotations.json\",\n    \"dataset.val.panoptic\": f\"{S3_EVAL}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.test.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.test.annotations\": f\"{S3_EVAL}\u002Fannotations.json\",\n    \"dataset.test.panoptic\": f\"{S3_EVAL}\u002Fimages_panoptic.tar.gz\",\n}\n```\n\n**export:**\n```python\n{\n    \"export.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"model.sem_seg_head.num_classes\": 133,\n    \"model.export\": True,\n    \"export.onnx_file\": \"\u002Fresults\u002Foneformer_export_640.onnx\",\n}\n```\n\n**inference (mandatory data sources):**\n```python\n{\n    \"inference.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"dataset.train.images\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.label_map\": f\"{S3_TRAIN}\u002Flabel_map.json\",\n    \"dataset.train.annotations\": f\"{S3_TRAIN}\u002Fannotations.json\",\n    \"dataset.train.panoptic\": f\"{S3_TRAIN}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.val.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val.annotations\": f\"{S3_EVAL}\u002Fannotations.json\",\n    \"dataset.val.panoptic\": f\"{S3_EVAL}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.test.images\": f\"{S3_INFERENCE}\u002Fimages.tar.gz\",\n    \"inference.images_dir\": f\"{S3_INFERENCE}\u002Fimages.tar.gz\",\n}\n```\n\n**quantize (mandatory data sources):**\n```python\n{\n    \"quantize.model_path\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"dataset.train.images\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.train.annotations\": f\"{S3_TRAIN}\u002Fannotations.json\",\n    \"dataset.label_map\": f\"{S3_TRAIN}\u002Flabel_map.json\",\n    \"dataset.train.panoptic\": f\"{S3_TRAIN}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.val.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val.annotations\": f\"{S3_EVAL}\u002Fannotations.json\",\n    \"dataset.val.panoptic\": f\"{S3_EVAL}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.test.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.quant_calibration_dataset.images_dir\": f\"{S3_CALIBRATION}\u002Fimages.tar.gz\",\n}\n```\n\n## Checkpoint Selection\n\nOneFormer training writes epoch-step checkpoints such as\n`model_epoch_000_step_00017.pth` and may also write a\n`oneformer_model_latest.pth` symlink. For checkpoint-dependent actions, use the\nmodel-skill or SDK parent-model resolver and pass the exact selected checkpoint\npath into `evaluate.checkpoint`, `inference.checkpoint`, `export.checkpoint`,\n`quantize.model_path`, or `train.resume_training_checkpoint_path`. Do not pick\nthe `oneformer_model_latest.pth` symlink by name unless the user explicitly asks\nfor latest checkpoint behavior. If the resolver reports a best checkpoint, use\nthat best checkpoint for evaluation\u002Fexport\u002Finference; if the user asks for a\nspecific epoch or step, use the matching epoch-step checkpoint.\n## Eval Dataset\n\nOptional. Val data configured alongside train in the dataset config.\n\n## Important Parameters\n\n- **model.sem_seg_head.num_classes**: Number of segmentation class indices available to the head. Default 133 for COCO panoptic data when `dataset.contiguous_id: True` remaps raw category ids through the label map. Do not shrink this to a global workflow class count unless the label map and annotations have actually been reduced to that class set.\n- **model.one_former.hidden_dim**: Keep at 256 for local smoke runs unless\n  the text encoder width is changed in lock-step. Reducing hidden_dim alone\n  causes a text feature\u002Fcontext dimension mismatch during training.\n- **model.backbone.name**: Default D2SwinTransformer (Swin-based). embed_dim=192, depths=[2,2,18,2] by default.\n- **train.num_epochs**: Default 50 — significantly higher than most TAO models. OneFormer needs more epochs for convergence.\n- **train.optim.lr**: Learning rate. Default 1e-5. Lower than Mask2Former's 2e-4.\n- **model.task_toggling**: Enable\u002Fdisable specific tasks: semantic_on, instance_on, panoptic_on.\n- **export.task**: Export task mode. Options: semantic, instance, panoptic. Default semantic. Export input defaults to 640x640.\n- **inference.mode**: Inference mode. Options: semantic, instance, panoptic. Default semantic. image_size defaults to [1024, 1024].\n- **evaluate.iou_per_class**: Report per-class IoU in evaluation. Default True.\n\n## Multi-GPU \u002F Multi-Node\n\n**Launch method:** Lightning-managed (single `python` process, Lightning spawns workers).\n\n| Spec Key | Description | Default |\n|----------|-------------|---------|\n| `train.num_gpus` | Number of GPUs | 1 |\n| `train.gpu_ids` | GPU device indices | [0] |\n| `train.num_nodes` | Number of nodes | 1 |\n\n- Uses explicit `DDPStrategy` with `find_unused_parameters=True`, `gradient_as_bucket_view=True`, `process_group_backend=\"nccl\"`\n- `sync_batchnorm` is always enabled\n- No fsdp support — DDP only\n\n**Multi-node env vars** (set by orchestrator): `WORLD_SIZE`, `NODE_RANK`, `MASTER_ADDR`, `MASTER_PORT`, `NUM_GPU_PER_NODE`.\n\n## Hardware\n\nMinimum 2 GPU(s), recommended 4 GPU(s). 24GB+ (A100 recommended) VRAM per GPU. OneFormer is memory-intensive like Mask2Former. batch_size=1 is the default. Multi-GPU needed for reasonable training speed, especially with 50 epochs.\n\n## Error Patterns\n\n**CUDA out of memory**: batch_size is already 1. Reduce image resolution or use a smaller Swin configuration.\n\n**Extracted S3 tarball points one level too high**: For local Docker runs,\n`images.tar.gz` and `images_panoptic.tar.gz` may extract wrapper directories\nsuch as `images\u002F` and `images_panoptic\u002F`. Set `dataset.*.images`,\n`dataset.*.panoptic`, `inference.images_dir`, and quantization calibration\npaths to the actual folder containing image or panoptic files, not the wrapper\ndirectory. A one-level-too-high path fails with `FileNotFoundError` for the\nfirst annotation image even though recursive file counts look correct.\n\n**default_specs missing results_dir**: The CLI `default_specs` subtask ignores\n`-e` experiment specs for `results_dir`; pass a Hydra-style override instead:\n`oneformer default_specs results_dir=\u002Fpath\u002Fto\u002Fdefault_specs`.\n\n**Invalid Lightning precision `fp32`**: Use `train.precision: \"32\"` in\ntrain\u002FAutoML\u002Fevaluate\u002Finference specs. The current Lightning stack rejects the\nlegacy `fp32` string.\n\n**PyTorch 2.6 checkpoint load failure on downstream actions**: Current\nOneFormer checkpoints include OmegaConf objects. For checkpoints produced by\nthe same trusted TAO train\u002FAutoML workflow, set\n`TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1` in downstream evaluate, inference, export,\nquantize, or resume job env vars so Lightning can load the full checkpoint.\nDo not use this env var for untrusted checkpoints.\n\n**CUDA device-side assert in matcher\u002Fclass cost**: If training fails in\n`oneformer\u002Futils\u002Fmatcher.py` while indexing `out_prob[:, tgt_ids]`, compare\nthe effective target ids with `model.sem_seg_head.num_classes`. The packaged\nCOCO panoptic sample has 133 compact classes after `dataset.contiguous_id:\nTrue` remapping, so use `model.sem_seg_head.num_classes: 133` even when a\nbroader validation workflow passes a smaller generic `num_classes` value.\nOnly use a smaller class count when the label map and annotations are reduced\nto that exact contiguous class set.\n\n**Inference returns PASS with no predictions**: OneFormer prediction reads\n`inference.images_dir`, not `dataset.test.images`. Declare and populate\n`inference.images_dir` with the image folder or tarball for every inference\nrun. `dataset.test.images` may still be useful for shared dataset context, but\nit does not drive the PyTorch predict dataloader.\n\n**Export output path pre-created as a directory**: Do not declare\n`export.onnx_file` as a file output. The OneFormer exporter asserts that the\nONNX path does not already exist, while the local runner pre-creates declared\noutput paths. Set `export.onnx_file` explicitly in the spec to a non-existing\nfile path under the mounted results tree. Keep the default 640x640 export\nshape for smoke validation; very small export shapes can trigger PyTorch ONNX\nshape-inference failures.\n\n**Quantize cannot find the training label map from an AutoML checkpoint**:\nOneFormer Lightning checkpoints retain train-time absolute dataset paths in\ntheir saved hparams. When running downstream actions from an AutoML child\ncheckpoint, keep the parent AutoML job directory accessible at its original\n`\u002Fresults\u002F\u003Cjob_id>` path inside the action container in addition to passing the\nresolved checkpoint path. Otherwise quantize can fail while loading checkpoint\nhparams even when the current spec includes a valid `dataset.label_map`.\n\n**Slow training**: 50 default epochs with batch_size=1 is slow on single GPU. Use multi-GPU distributed training.\n\n## Spec Param \u002F Parent Model Inference\n\nModel-specific inference mappings belong in this MD file, not in `config.json`. Generated runners should read this section and apply the mappings with SDK helpers before `create_job()`. This mirrors the old microservices `infer_params.py` flow.\n\nInference mappings from TAO Core `oneformer.config.json`:\n\n| Action | Spec Field | Inference Function | Meaning |\n|---|---|---|---|\n| evaluate | `encryption_key` | `key` | encryption key |\n| evaluate | `evaluate.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| evaluate | `evaluate.trt_engine` | `parent_model` | model file inferred from the parent job results folder |\n| evaluate | `results_dir` | `output_dir` | current job results directory |\n| export | `encryption_key` | `key` | encryption key |\n| export | `export.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| export | `export.onnx_file` | `create_onnx_file` | output ONNX path |\n| export | `results_dir` | `output_dir` | current job results directory |\n| gen_trt_engine | `encryption_key` | `key` | encryption key |\n| gen_trt_engine | `gen_trt_engine.onnx_file` | `parent_model` | model file inferred from the parent job results folder |\n| gen_trt_engine | `gen_trt_engine.trt_engine` | `create_engine_file` | output TensorRT engine path |\n| gen_trt_engine | `results_dir` | `output_dir` | current job results directory |\n| inference | `encryption_key` | `key` | encryption key |\n| inference | `inference.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| inference | `inference.trt_engine` | `parent_model` | model file inferred from the parent job results folder |\n| inference | `results_dir` | `output_dir` | current job results directory |\n| quantize | `encryption_key` | `key` | encryption key |\n| quantize | `quantize.model_path` | `parent_model` | model file inferred from the parent job results folder |\n| quantize | `results_dir` | `output_dir` | current job results directory |\n| train | `encryption_key` | `key` | encryption key |\n| train | `results_dir` | `output_dir` | current job results directory |\n| train | `train.pretrained_backbone` | `{'link': 'https:\u002F\u002Fgithub.com\u002FSwinTransformer\u002Fstorage\u002Freleases\u002Fdownload\u002Fv1.0.8\u002Fswin_tiny_patch4_window7_224_22k.pth', 'destination_path': '\u002Fptm\u002Fmask2former\u002Fswin_tiny_patch4_window7_224_22k\u002Fswin_tiny_patch4_window7_224_22k.pth'}` | {'link': 'https:\u002F\u002Fgithub.com\u002FSwinTransformer\u002Fstorage\u002Freleases\u002Fdownload\u002Fv1.0.8\u002Fswin_tiny_patch4_window7_224_22k.pth', 'destination_path': '\u002Fptm\u002Fmask2former\u002Fswin_tiny_patch4_window7_224_22k\u002Fswin_tiny_patch4_window7_224_22k.pth'} |\n| train | `train.pretrained_model` | `ptm_if_no_resume_model` | PTM when no resume checkpoint exists |\n| train | `train.resume_training_checkpoint_path` | `resume_model` | model file inferred from the current job results folder |\n\nFor `parent_model` or `parent_model_folder`, pass the upstream train\u002Fexport\u002FAutoML child job id as `parent_job_id`. The SDK lists the parent result folder, filters checkpoint artifacts, and returns the selected model file or folder. Do not add these mappings back to `config.json` and do not patch generated runner scripts to guess checkpoint paths.\n\n## Deployment\n\n- [tao-deploy-oneformer](references\u002Ftao-deploy-oneformer.md)\n","# OneFormer\n\nOneFormer for universal image segmentation. Unifies panoptic, instance, and semantic segmentation with a single architecture using task-conditioned queries.\n\nSet train.pretrained_backbone and\u002For train.pretrained_model.\n\nFor TAO Deploy TensorRT actions (`gen_trt_engine`, TensorRT `evaluate`, and TensorRT `inference`), read `references\u002Ftao-deploy-oneformer.md` first. Deploy spec templates live in this skill's `references\u002F` folder with the `spec_template_deploy_*.yaml` prefix.\n\n## Dataclass Schemas\n\nGenerated TAO Core schemas are packaged in `schemas\u002F\u003Caction>.schema.json`, with `schemas\u002Fmanifest.json` listing available actions. Each generated schema also emits `references\u002Fspec_template_\u003Caction>.yaml` from the schema top-level `default` field. AutoML enablement is declared at the model layer in `references\u002Fskill_info.yaml` via `automl_enabled`. Runnable AutoML still requires `schemas\u002Ftrain.schema.json` and `references\u002Fspec_template_train.yaml` to exist and parse. Use the packaged train schema for `automl_default_parameters`, `automl_disabled_parameters`, defaults, min\u002Fmax bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect `~\u002Ftao-core` at runtime; maintainers regenerate schemas\u002Ftemplates before packaging the skill bank.\n\n## Train Action Policy\n\nThis model is AutoML-enabled at the model layer. Before handling any train-stage request, read `references\u002Fskill_info.yaml` and resolve the run override from either an explicit `automl_policy` value or the user's workflow request. Use `automl_policy: on` by default and only expose `on` \u002F `off` in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as `automl_policy: off` for this run only. When `automl_policy: on`, `automl_enabled: true`, and both `schemas\u002Ftrain.schema.json` and `references\u002Fspec_template_train.yaml` are packaged, route the train action through `tao-skill-bank:tao-run-automl` by default with this model's `skill_dir`. Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and `automl_policy`. Use direct model training only when `automl_policy: off` or the packaged train schema\u002Ftemplate is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.\n\nNon-train actions such as `evaluate`, `inference`, `export`, and deploy flows stay in this model skill. The per-run `automl_policy` override does not change model metadata.\n\n## Training Requirements\n\n- **Dataset type:** segmentation\n- **Formats:** coco_panoptic, coco\n- **Monitoring metric:** mIoU\n\n### Per-Action Dataset Requirements\n\n| Action | Spec Key | Source | Files | List? |\n|---|---|---|---|---|\n| evaluate | dataset.train.images | train_datasets | images.tar.gz | No |\n| evaluate | dataset.label_map | train_datasets | label_map.json | No |\n| evaluate | dataset.train.annotations | train_datasets | annotations.json | No |\n| evaluate | dataset.train.panoptic | train_datasets | images_panoptic.tar.gz | No |\n| evaluate | dataset.val.images | eval_dataset | images.tar.gz | No |\n| evaluate | dataset.val.annotations | eval_dataset | annotations.json | No |\n| evaluate | dataset.val.panoptic | eval_dataset | images_panoptic.tar.gz | No |\n| evaluate | dataset.test.images | eval_dataset | images.tar.gz | No |\n| evaluate | dataset.test.annotations | eval_dataset | annotations.json | No |\n| evaluate | dataset.test.panoptic | eval_dataset | images_panoptic.tar.gz | No |\n| inference | dataset.train.images | train_datasets | images.tar.gz | No |\n| inference | dataset.label_map | train_datasets | label_map.json | No |\n| inference | dataset.train.annotations | train_datasets | annotations.json | No |\n| inference | dataset.train.panoptic | train_datasets | images_panoptic.tar.gz | No |\n| inference | dataset.val.images | eval_dataset | images.tar.gz | No |\n| inference | dataset.val.annotations | eval_dataset | annotations.json | No |\n| inference | dataset.val.panoptic | eval_dataset | images_panoptic.tar.gz | No |\n| inference | dataset.test.images | inference_dataset | images.tar.gz | No |\n| quantize | dataset.train.images | train_datasets | images.tar.gz | No |\n| quantize | dataset.train.annotations | train_datasets | annotations.json | No |\n| quantize | dataset.label_map | train_datasets | label_map.json | No |\n| quantize | dataset.train.panoptic | train_datasets | images_panoptic.tar.gz | No |\n| quantize | dataset.val.images | eval_dataset | images.tar.gz | No |\n| quantize | dataset.val.annotations | eval_dataset | annotations.json | No |\n| quantize | dataset.val.panoptic | eval_dataset | images_panoptic.tar.gz | No |\n| quantize | dataset.test.images | eval_dataset | images.tar.gz | No |\n| quantize | dataset.quant_calibration_dataset.images_dir | calibration_dataset | images.tar.gz | No |\n| train | dataset.train.images | train_datasets | images.tar.gz | No |\n| train | dataset.train.annotations | train_datasets | annotations.json | No |\n| train | dataset.label_map | train_datasets | label_map.json | No |\n| train | dataset.train.panoptic | train_datasets | images_panoptic.tar.gz | No |\n| train | dataset.val.images | eval_dataset | images.tar.gz | No |\n| train | dataset.val.annotations | eval_dataset | annotations.json | No |\n| train | dataset.val.panoptic | eval_dataset | images_panoptic.tar.gz | No |\n| train | dataset.test.images | eval_dataset | images.tar.gz | No |\n\n### Typical Spec Overrides\n\nData source overrides are **mandatory for every action** — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in `spec_overrides`.\n\n```python\nS3_TRAIN = \"s3:\u002F\u002Fbucket\u002Fdata\u002Ftrain\"\nS3_EVAL = \"s3:\u002F\u002Fbucket\u002Fdata\u002Feval\"\nS3_INFERENCE = \"s3:\u002F\u002Fbucket\u002Fdata\u002Finference\"\nS3_CALIBRATION = \"s3:\u002F\u002Fbucket\u002Fdata\u002Fcalibration\"\n```\n\n**train (mandatory data sources):**\n```python\n{\n    \"train.num_gpus\": 1,\n    \"train.num_epochs\": 10,\n    \"train.checkpoint_interval\": 10,\n    \"train.validation_interval\": 10,\n    \"model.sem_seg_head.num_classes\": 133,\n    \"dataset.contiguous_id\": True,\n    \"train.precision\": \"32\",\n    \"dataset.train.images\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.train.annotations\": f\"{S3_TRAIN}\u002Fannotations.json\",\n    \"dataset.label_map\": f\"{S3_TRAIN}\u002Flabel_map.json\",\n    \"dataset.train.panoptic\": f\"{S3_TRAIN}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.val.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val.annotations\": f\"{S3_EVAL}\u002Fannotations.json\",\n    \"dataset.val.panoptic\": f\"{S3_EVAL}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.test.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n}\n```\n\n**evaluate (mandatory data sources):**\n```python\n{\n    \"evaluate.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"model.sem_seg_head.num_classes\": 133,\n    \"dataset.contiguous_id\": True,\n    \"dataset.train.images\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.label_map\": f\"{S3_TRAIN}\u002Flabel_map.json\",\n    \"dataset.train.annotations\": f\"{S3_TRAIN}\u002Fannotations.json\",\n    \"dataset.train.panoptic\": f\"{S3_TRAIN}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.val.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val.annotations\": f\"{S3_EVAL}\u002Fannotations.json\",\n    \"dataset.val.panoptic\": f\"{S3_EVAL}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.test.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.test.annotations\": f\"{S3_EVAL}\u002Fannotations.json\",\n    \"dataset.test.panoptic\": f\"{S3_EVAL}\u002Fimages_panoptic.tar.gz\",\n}\n```\n\n**export:**\n```python\n{\n    \"export.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"model.sem_seg_head.num_classes\": 133,\n    \"model.export\": True,\n    \"export.onnx_file\": \"\u002Fresults\u002Foneformer_export_640.onnx\",\n}\n```\n\n**inference (mandatory data sources):**\n```python\n{\n    \"inference.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"dataset.train.images\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.label_map\": f\"{S3_TRAIN}\u002Flabel_map.json\",\n    \"dataset.train.annotations\": f\"{S3_TRAIN}\u002Fannotations.json\",\n    \"dataset.train.panoptic\": f\"{S3_TRAIN}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.val.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val.annotations\": f\"{S3_EVAL}\u002Fannotations.json\",\n    \"dataset.val.panoptic\": f\"{S3_EVAL}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.test.images\": f\"{S3_INFERENCE}\u002Fimages.tar.gz\",\n    \"inference.images_dir\": f\"{S3_INFERENCE}\u002Fimages.tar.gz\",\n}\n```\n\n**quantize (mandatory data sources):**\n```python\n{\n    \"quantize.model_path\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"dataset.train.images\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.train.annotations\": f\"{S3_TRAIN}\u002Fannotations.json\",\n    \"dataset.label_map\": f\"{S3_TRAIN}\u002Flabel_map.json\",\n    \"dataset.train.panoptic\": f\"{S3_TRAIN}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.val.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val.annotations\": f\"{S3_EVAL}\u002Fannotations.json\",\n    \"dataset.val.panoptic\": f\"{S3_EVAL}\u002Fimages_panoptic.tar.gz\",\n    \"dataset.test.images\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.quant_calibration_dataset.images_dir\": f\"{S3_CALIBRATION}\u002Fimages.tar.gz\",\n}\n```\n\n## Checkpoint Selection\n\nOneFormer training writes epoch-step checkpoints such as\n`model_epoch_000_step_00017.pth` and may also write a\n`oneformer_model_latest.pth` symlink. For checkpoint-dependent actions, use the\nmodel-skill or SDK parent-model resolver and pass the exact selected checkpoint\npath into `evaluate.checkpoint`, `inference.checkpoint`, `export.checkpoint`,\n`quantize.model_path`, or `train.resume_training_checkpoint_path`. Do not pick\nthe `oneformer_model_latest.pth` symlink by name unless the user explicitly asks\nfor latest checkpoint behavior. If the resolver reports a best checkpoint, use\nthat best checkpoint for evaluation\u002Fexport\u002Finference; if the user asks for a\nspecific epoch or step, use the matching epoch-step checkpoint.\n## Eval Dataset\n\nOptional. Val data configured alongside train in the dataset config.\n\n## Important Parameters\n\n- **model.sem_seg_head.num_classes**: Number of segmentation class indices available to the head. Default 133 for COCO panoptic data when `dataset.contiguous_id: True` remaps raw category ids through the label map. Do not shrink this to a global workflow class count unless the label map and annotations have actually been reduced to that class set.\n- **model.one_former.hidden_dim**: Keep at 256 for local smoke runs unless\n  the text encoder width is changed in lock-step. Reducing hidden_dim alone\n  causes a text feature\u002Fcontext dimension mismatch during training.\n- **model.backbone.name**: Default D2SwinTransformer (Swin-based). embed_dim=192, depths=[2,2,18,2] by default.\n- **train.num_epochs**: Default 50 — significantly higher than most TAO models. OneFormer needs more epochs for convergence.\n- **train.optim.lr**: Learning rate. Default 1e-5. Lower than Mask2Former's 2e-4.\n- **model.task_toggling**: Enable\u002Fdisable specific tasks: semantic_on, instance_on, panoptic_on.\n- **export.task**: Export task mode. Options: semantic, instance, panoptic. Default semantic. Export input defaults to 640x640.\n- **inference.mode**: Inference mode. Options: semantic, instance, panoptic. Default semantic. image_size defaults to [1024, 1024].\n- **evaluate.iou_per_class**: Report per-class IoU in evaluation. Default True.\n\n## Multi-GPU \u002F Multi-Node\n\n**Launch method:** Lightning-managed (single `python` process, Lightning spawns workers).\n\n| Spec Key | Description | Default |\n|----------|-------------|---------|\n| `train.num_gpus` | Number of GPUs | 1 |\n| `train.gpu_ids` | GPU device indices | [0] |\n| `train.num_nodes` | Number of nodes | 1 |\n\n- Uses explicit `DDPStrategy` with `find_unused_parameters=True`, `gradient_as_bucket_view=True`, `process_group_backend=\"nccl\"`\n- `sync_batchnorm` is always enabled\n- No fsdp support — DDP only\n\n**Multi-node env vars** (set by orchestrator): `WORLD_SIZE`, `NODE_RANK`, `MASTER_ADDR`, `MASTER_PORT`, `NUM_GPU_PER_NODE`.\n\n## Hardware\n\nMinimum 2 GPU(s), recommended 4 GPU(s). 24GB+ (A100 recommended) VRAM per GPU. OneFormer is memory-intensive like Mask2Former. batch_size=1 is the default. Multi-GPU needed for reasonable training speed, especially with 50 epochs.\n\n## Error Patterns\n\n**CUDA out of memory**: batch_size is already 1. Reduce image resolution or use a smaller Swin configuration.\n\n**Extracted S3 tarball points one level too high**: For local Docker runs,\n`images.tar.gz` and `images_panoptic.tar.gz` may extract wrapper directories\nsuch as `images\u002F` and `images_panoptic\u002F`. Set `dataset.*.images`,\n`dataset.*.panoptic`, `inference.images_dir`, and quantization calibration\npaths to the actual folder containing image or panoptic files, not the wrapper\ndirectory. A one-level-too-high path fails with `FileNotFoundError` for the\nfirst annotation image even though recursive file counts look correct.\n\n**default_specs missing results_dir**: The CLI `default_specs` subtask ignores\n`-e` experiment specs for `results_dir`; pass a Hydra-style override instead:\n`oneformer default_specs results_dir=\u002Fpath\u002Fto\u002Fdefault_specs`.\n\n**Invalid Lightning precision `fp32`**: Use `train.precision: \"32\"` in\ntrain\u002FAutoML\u002Fevaluate\u002Finference specs. The current Lightning stack rejects the\nlegacy `fp32` string.\n\n**PyTorch 2.6 checkpoint load failure on downstream actions**: Current\nOneFormer checkpoints include OmegaConf objects. For checkpoints produced by\nthe same trusted TAO train\u002FAutoML workflow, set\n`TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1` in downstream evaluate, inference, export,\nquantize, or resume job env vars so Lightning can load the full checkpoint.\nDo not use this env var for untrusted checkpoints.\n\n**CUDA device-side assert in matcher\u002Fclass cost**: If training fails in\n`oneformer\u002Futils\u002Fmatcher.py` while indexing `out_prob[:, tgt_ids]`, compare\nthe effective target ids with `model.sem_seg_head.num_classes`. The packaged\nCOCO panoptic sample has 133 compact classes after `dataset.contiguous_id:\nTrue` remapping, so use `model.sem_seg_head.num_classes: 133` even when a\nbroader validation workflow passes a smaller generic `num_classes` value.\nOnly use a smaller class count when the label map and annotations are reduced\nto that exact contiguous class set.\n\n**Inference returns PASS with no predictions**: OneFormer prediction reads\n`inference.images_dir`, not `dataset.test.images`. Declare and populate\n`inference.images_dir` with the image folder or tarball for every inference\nrun. `dataset.test.images` may still be useful for shared dataset context, but\nit does not drive the PyTorch predict dataloader.\n\n**Export output path pre-created as a directory**: Do not declare\n`export.onnx_file` as a file output. The OneFormer exporter asserts that the\nONNX path does not already exist, while the local runner pre-creates declared\noutput paths. Set `export.onnx_file` explicitly in the spec to a non-existing\nfile path under the mounted results tree. Keep the default 640x640 export\nshape for smoke validation; very small export shapes can trigger PyTorch ONNX\nshape-inference failures.\n\n**Quantize cannot find the training label map from an AutoML checkpoint**:\nOneFormer Lightning checkpoints retain train-time absolute dataset paths in\ntheir saved hparams. When running downstream actions from an AutoML child\ncheckpoint, keep the parent AutoML job directory accessible at its original\n`\u002Fresults\u002F\u003Cjob_id>` path inside the action container in addition to passing the\nresolved checkpoint path. Otherwise quantize can fail while loading checkpoint\nhparams even when the current spec includes a valid `dataset.label_map`.\n\n**Slow training**: 50 default epochs with batch_size=1 is slow on single GPU. Use multi-GPU distributed training.\n\n## Spec Param \u002F Parent Model Inference\n\nModel-specific inference mappings belong in this MD file, not in `config.json`. Generated runners should read this section and apply the mappings with SDK helpers before `create_job()`. This mirrors the old microservices `infer_params.py` flow.\n\nInference mappings from TAO Core `oneformer.config.json`:\n\n| Action | Spec Field | Inference Function | Meaning |\n|---|---|---|---|\n| evaluate | `encryption_key` | `key` | encryption key |\n| evaluate | `evaluate.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| evaluate | `evaluate.trt_engine` | `parent_model` | model file inferred from the parent job results folder |\n| evaluate | `results_dir` | `output_dir` | current job results directory |\n| export | `encryption_key` | `key` | encryption key |\n| export | `export.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| export | `export.onnx_file` | `create_onnx_file` | output ONNX path |\n| export | `results_dir` | `output_dir` | current job results directory |\n| gen_trt_engine | `encryption_key` | `key` | encryption key |\n| gen_trt_engine | `gen_trt_engine.onnx_file` | `parent_model` | model file inferred from the parent job results folder |\n| gen_trt_engine | `gen_trt_engine.trt_engine` | `create_engine_file` | output TensorRT engine path |\n| gen_trt_engine | `results_dir` | `output_dir` | current job results directory |\n| inference | `encryption_key` | `key` | encryption key |\n| inference | `inference.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| inference | `inference.trt_engine` | `parent_model` | model file inferred from the parent job results folder |\n| inference | `results_dir` | `output_dir` | current job results directory |\n| quantize | `encryption_key` | `key` | encryption key |\n| quantize | `quantize.model_path` | `parent_model` | model file inferred from the parent job results folder |\n| quantize | `results_dir` | `output_dir` | current job results directory |\n| train | `encryption_key` | `key` | encryption key |\n| train | `results_dir` | `output_dir` | current job results directory |\n| train | `train.pretrained_backbone` | `{'link': 'https:\u002F\u002Fgithub.com\u002FSwinTransformer\u002Fstorage\u002Freleases\u002Fdownload\u002Fv1.0.8\u002Fswin_tiny_patch4_window7_224_22k.pth', 'destination_path': '\u002Fptm\u002Fmask2former\u002Fswin_tiny_patch4_window7_224_22k\u002Fswin_tiny_patch4_window7_224_22k.pth'}` | {'link': 'https:\u002F\u002Fgithub.com\u002FSwinTransformer\u002Fstorage\u002Freleases\u002Fdownload\u002Fv1.0.8\u002Fswin_tiny_patch4_window7_224_22k.pth', 'destination_path': '\u002Fptm\u002Fmask2former\u002Fswin_tiny_patch4_window7_224_22k\u002Fswin_tiny_patch4_window7_224_22k.pth'} |\n| train | `train.pretrained_model` | `ptm_if_no_resume_model` | PTM when no resume checkpoint exists |\n| train | `train.resume_training_checkpoint_path` | `resume_model` | model file inferred from the current job results folder |\n\nFor `parent_model` or `parent_model_folder`, pass the upstream train\u002Fexport\u002FAutoML child job id as `parent_job_id`. The SDK lists the parent result folder, filters checkpoint artifacts, and returns the selected model file or folder. Do not add these mappings back to `config.json` and do not patch generated runner scripts to guess checkpoint paths.\n\n## Deployment\n\n- [tao-deploy-oneformer](references\u002Ftao-deploy-oneformer.md)\n",{"name":5,"tags":24,"license":15,"metadata":26,"description":7,"allowed-tools":29,"compatibility":30},[25],"segmentation",{"author":27,"version":28},"NVIDIA Corporation","0.1.0","Read Bash","Requires docker + nvidia-container-toolkit.","main","be3de0244dc1a57e323d8160787ebb40fb64930e","public\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main","https:\u002F\u002Fskills.x402.wtf","https:\u002F\u002Fgithub.com\u002FSolizardking","https:\u002F\u002Fmusebook.trade","Got a mile high and never came down","2026-10-08 23:20:13",{"source":8,"skillName":5,"commands":41,"prompt":51,"zipAvailable":52},[42,45,48],{"tool":43,"command":44},"npx","npx skills add Solizardking\u002Fskillhub-main --skill tao-train-oneformer",{"tool":46,"command":47},"bunx","bunx skills add Solizardking\u002Fskillhub-main --skill tao-train-oneformer",{"tool":49,"command":50},"pnpm","pnpm dlx skills add Solizardking\u002Fskillhub-main --skill tao-train-oneformer","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002FSKILL.md",true,[54,60,64,68,72,76,79,83,87,91,95,99,102,106,110,114],{"path":55,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":58,"sourceHtmlUrl":59},"BENCHMARK.md","REFERENCE","参考文档",4072,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002FBENCHMARK.md",{"path":61,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":62,"sourceHtmlUrl":63},"metadata.json",622,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Fmetadata.json",{"path":65,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":66,"sourceHtmlUrl":67},"references\u002Fskill_info.yaml",9661,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Fskill_info.yaml",{"path":69,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":70,"sourceHtmlUrl":71},"references\u002Fspec_template_deploy_evaluate.yaml",521,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Fspec_template_deploy_evaluate.yaml",{"path":73,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":74,"sourceHtmlUrl":75},"references\u002Fspec_template_deploy_gen_trt_engine.yaml",277,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Fspec_template_deploy_gen_trt_engine.yaml",{"path":77,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":70,"sourceHtmlUrl":78},"references\u002Fspec_template_deploy_inference.yaml","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Fspec_template_deploy_inference.yaml",{"path":80,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":81,"sourceHtmlUrl":82},"references\u002Fspec_template_evaluate.yaml",4751,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Fspec_template_evaluate.yaml",{"path":84,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":85,"sourceHtmlUrl":86},"references\u002Fspec_template_export.yaml",4820,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Fspec_template_export.yaml",{"path":88,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":89,"sourceHtmlUrl":90},"references\u002Fspec_template_gen_trt_engine.yaml",4883,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Fspec_template_gen_trt_engine.yaml",{"path":92,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":93,"sourceHtmlUrl":94},"references\u002Fspec_template_inference.yaml",4796,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Fspec_template_inference.yaml",{"path":96,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":97,"sourceHtmlUrl":98},"references\u002Fspec_template_quantize.yaml",4604,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Fspec_template_quantize.yaml",{"path":100,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":97,"sourceHtmlUrl":101},"references\u002Fspec_template_train.yaml","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Fspec_template_train.yaml",{"path":103,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":104,"sourceHtmlUrl":105},"references\u002Ftao-deploy-oneformer.md",6134,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Ftao-deploy-oneformer.md",{"path":107,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":108,"sourceHtmlUrl":109},"references\u002Ftao-deploy-oneformer.skill_info.yaml",3759,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Freferences\u002Ftao-deploy-oneformer.skill_info.yaml",{"path":111,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":112,"sourceHtmlUrl":113},"skill-card.md",4002,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Fskill-card.md",{"path":115,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":116,"sourceHtmlUrl":117},"verification.json",3790,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-train-oneformer\u002Fverification.json",[119,127,134,141,148,156],{"skillNo":120,"name":121,"nameZh":6,"description":122,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":123,"licenseFileUrl":6,"sourceHtmlUrl":124,"categories":125,"keywords":126,"publishedAt":19},"01M78E0TGCAMNPS207BA3G07S7","zkrouter","Self-hosted OpenAI-compatible LLM router — smart tier-based routing, ZK-stamped routing decisions, mode overrides, free OpenRouter at install via the Birth bot. Save 60-80% on inference costs by routing to the cheapest capable model across 12+ OpenRouter models.","MIT","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fzkrouter\u002FSKILL.md",[],[],{"skillNo":128,"name":129,"nameZh":6,"description":130,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":123,"licenseFileUrl":6,"sourceHtmlUrl":131,"categories":132,"keywords":133,"publishedAt":19},"01M78E0R1Z8BDT6EDTHY17MTNB","zk","For custom ZK Solana programs and privacy-preserving applications to prevent double spending. Guide to integrate rent-free nullifier PDAs for double-spend prevention.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fzk\u002FSKILL.md",[],[],{"skillNo":135,"name":136,"nameZh":6,"description":137,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":123,"licenseFileUrl":6,"sourceHtmlUrl":138,"categories":139,"keywords":140,"publishedAt":19},"01M78E0P3ZTZ1YWF8MD5MM23QE","zk-omni-messaging","Zero-knowledge omnichain messaging between Robinhood Chain and Solana using CheshireZkOmniMessenger (msgType 4) with nullifier anti-replay and the zk-omni-relayer service.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fzk-omni-messaging\u002FSKILL.md",[],[],{"skillNo":142,"name":143,"nameZh":6,"description":144,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":123,"licenseFileUrl":6,"sourceHtmlUrl":145,"categories":146,"keywords":147,"publishedAt":19},"01M78E0JXZ2PNB1896VXAGDBWD","zerion-wallet-data","Use when the user asks to check a wallet's portfolio, token balances, DeFi positions, PnL, transaction history, or NFTs. Provides interpreted on-chain data across 40+ EVM chains and Solana via x402 pay-per-request or API key — no raw RPC decoding needed.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fzerion-wallet-data\u002FSKILL.md",[],[],{"skillNo":149,"name":150,"nameZh":6,"description":151,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":123,"licenseFileUrl":152,"sourceHtmlUrl":153,"categories":154,"keywords":155,"publishedAt":19},"01M78E0G14FH9GRRTQYMKD5E59","youtube-clipper","YouTube 视频智能剪辑工具。下载视频和字幕，AI 分析生成精细章节（几分钟级别）， 用户选择片段后自动剪辑、翻译字幕为中英双语、烧录字幕到视频，并生成总结文案。 使用场景：当用户需要剪辑 YouTube 视频、生成短视频片段、制作双语字幕版本时。 关键词：视频剪辑、YouTube、字幕翻译、双语字幕、视频下载、clip video","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fyoutube-clipper\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fyoutube-clipper\u002FSKILL.md",[],[],{"skillNo":157,"name":158,"nameZh":6,"description":159,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":123,"licenseFileUrl":6,"sourceHtmlUrl":160,"categories":161,"keywords":162,"publishedAt":19},"01M78E0C2ZYA99CFFDNB7ZMK5S","yield-optimization","Multi-chain yield optimization via yield.xyz (StakeKit). Discover 2,988+ yield opportunities across 75+ blockchains, build deposit\u002Fwithdrawal transactions with shell scripts, and sign them with a MoonPay wallet. Use when the user wants to earn yield, find the best lending or staking rate, enter\u002Fexit a position, or check portfolio balances.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Fskillhub-main\u002Fblob\u002Fmain\u002Fskills\u002Fyield-optimization\u002FSKILL.md",[],[]]