[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7F9V3NFYKF2Z6REGPSDBCFS":3,"skill-same:01M7F9V3NFYKF2Z6REGPSDBCFS":63},{"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":28,"commitSha":29,"skillDirectory":30,"forksCount":31,"repositoryHtmlUrl":32,"repositoryHomepageUrl":6,"ownerHtmlUrl":33,"ownerBlogUrl":6,"ownerDescription":34,"updatedAt":35,"install":36,"resources":48},"01M7F9V3NFYKF2Z6REGPSDBCFS","model-architecture",null,"Inspect and configure HRM ACT v1 model internals, losses, FlashAttention layers, and dynamic model identifiers.","lza6\u002FArmor-piercing-system-prompts","lza6","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F128880206?v=4","01M7EB3K7WPX5T28EJ2NVHE6EG","Chinese-tingfeng","01M7EB3K97E1KQ16QWKJMFNGX8",11,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Fhrm\u002Fsub-skills\u002Fmodel-architecture\u002FSKILL.md",[],[],"2026-11-16 10:05:34",false,"---\nname: model-architecture\ndescription: \"Inspect and configure HRM ACT v1 model internals, losses,\n  FlashAttention layers, and dynamic model identifiers.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: Apache 2.0\n---\n\n# HRM Model Architecture\n\nUse this sub-skill when the task is to understand, configure, inspect, or debug\nthe Hierarchical Reasoning Model (HRM) ACT v1 implementation rather than to run\nfull training.\n\n## When to use\n\n- The user asks about HRM high-level\u002Flow-level recurrent modules, ACT halting,\n  Q heads, carry state, sparse puzzle embeddings, StableMax loss, or\n  FlashAttention-backed attention blocks.\n- A config mentions `arch.name=hrm.hrm_act_v1@HierarchicalReasoningModel_ACTV1`\n  or `arch.loss.name=losses@ACTLossHead`.\n- A model import fails because `flash_attn`, `flash_attn_interface`,\n  `adam_atan2_backend`, CUDA, or dynamic `module@class` identifiers are wrong.\n- The user needs to alter hidden size, heads, cycles, layers, positional\n  encodings, halt steps, or loss type safely.\n\n## Route map\n\n1. Read [references\u002Farchitecture.md](references\u002Farchitecture.md) for the HRM\n   ACT v1 object graph, carry objects, input\u002Foutput tensors, and halting\n   mechanics.\n2. Read [references\u002Fapi-reference.md](references\u002Fapi-reference.md) for verified\n   config fields, function\u002Fclass signatures, losses, and dynamic identifier\n   rules.\n3. Read [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md) for\n   import\u002Fbackend\u002Fconfig errors.\n4. Run [scripts\u002Finspect_model_config.py](scripts\u002Finspect_model_config.py) to\n   inspect identifiers and optionally run CUDA dependency smokes from an HRM\n   checkout.\n\n## Critical facts\n\n- The default architecture config uses\n  `hrm.hrm_act_v1@HierarchicalReasoningModel_ACTV1` wrapped by\n  `losses@ACTLossHead` with `loss_type=stablemax_cross_entropy`.\n- `models\u002Flayers.py` imports FlashAttention at module import time. If neither\n  `flash_attn_interface` nor `flash_attn` is importable, HRM model imports fail.\n- `pretrain.py` imports `AdamATan2` from `adam_atan2`; real training requires\n  the compiled `adam_atan2_backend` CUDA extension.\n- Real model construction and forward execution are CUDA-oriented. CPU imports\n  and source inspection are useful for config facts but do not verify the\n  required training\u002Fevaluation backend.\n- The repository's tiny model smoke under current dependency versions exposed a\n  runtime issue where FlashAttention output may be non-contiguous and\n  `models\u002Flayers.py` calls `.view(...)`; use this as troubleshooting evidence\n  if a forward pass fails with a stride\u002Fview error.\n\n## Boundaries\n\n- This sub-skill owns model\u002Floss\u002Fconfig\u002FAPI understanding and environment\n  smokes for model dependencies.\n- Use `data-preparation` for dataset schema, builders, and visualization.\n- Use `training-evaluation` for Hydra training commands, checkpoint evaluation,\n  W&B, and ARC post-processing.\n- Do not present a CPU import as proof that HRM can train or evaluate. The\n  backend gate is CUDA + FlashAttention + adam-atan2 backend.\n\n## Bounded inspection\n\nFrom an HRM checkout, run:\n\n```bash\npython \u003Cskill>\u002Fsub-skills\u002Fmodel-architecture\u002Fscripts\u002Finspect_model_config.py \\\n  --repo-root \u002Fpath\u002Fto\u002FHRM --json\n```\n\nIf a CUDA environment is prepared and the user wants dependency readiness:\n\n```bash\npython \u003Cskill>\u002Fsub-skills\u002Fmodel-architecture\u002Fscripts\u002Finspect_model_config.py \\\n  --repo-root \u002Fpath\u002Fto\u002FHRM --cuda-smoke\n```\n\nThe helper verifies imports and dependency smokes. It intentionally avoids long\ntraining and does not require real datasets or checkpoints.\n","# HRM Model Architecture\n\nUse this sub-skill when the task is to understand, configure, inspect, or debug\nthe Hierarchical Reasoning Model (HRM) ACT v1 implementation rather than to run\nfull training.\n\n## When to use\n\n- The user asks about HRM high-level\u002Flow-level recurrent modules, ACT halting,\n  Q heads, carry state, sparse puzzle embeddings, StableMax loss, or\n  FlashAttention-backed attention blocks.\n- A config mentions `arch.name=hrm.hrm_act_v1@HierarchicalReasoningModel_ACTV1`\n  or `arch.loss.name=losses@ACTLossHead`.\n- A model import fails because `flash_attn`, `flash_attn_interface`,\n  `adam_atan2_backend`, CUDA, or dynamic `module@class` identifiers are wrong.\n- The user needs to alter hidden size, heads, cycles, layers, positional\n  encodings, halt steps, or loss type safely.\n\n## Route map\n\n1. Read [references\u002Farchitecture.md](references\u002Farchitecture.md) for the HRM\n   ACT v1 object graph, carry objects, input\u002Foutput tensors, and halting\n   mechanics.\n2. Read [references\u002Fapi-reference.md](references\u002Fapi-reference.md) for verified\n   config fields, function\u002Fclass signatures, losses, and dynamic identifier\n   rules.\n3. Read [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md) for\n   import\u002Fbackend\u002Fconfig errors.\n4. Run [scripts\u002Finspect_model_config.py](scripts\u002Finspect_model_config.py) to\n   inspect identifiers and optionally run CUDA dependency smokes from an HRM\n   checkout.\n\n## Critical facts\n\n- The default architecture config uses\n  `hrm.hrm_act_v1@HierarchicalReasoningModel_ACTV1` wrapped by\n  `losses@ACTLossHead` with `loss_type=stablemax_cross_entropy`.\n- `models\u002Flayers.py` imports FlashAttention at module import time. If neither\n  `flash_attn_interface` nor `flash_attn` is importable, HRM model imports fail.\n- `pretrain.py` imports `AdamATan2` from `adam_atan2`; real training requires\n  the compiled `adam_atan2_backend` CUDA extension.\n- Real model construction and forward execution are CUDA-oriented. CPU imports\n  and source inspection are useful for config facts but do not verify the\n  required training\u002Fevaluation backend.\n- The repository's tiny model smoke under current dependency versions exposed a\n  runtime issue where FlashAttention output may be non-contiguous and\n  `models\u002Flayers.py` calls `.view(...)`; use this as troubleshooting evidence\n  if a forward pass fails with a stride\u002Fview error.\n\n## Boundaries\n\n- This sub-skill owns model\u002Floss\u002Fconfig\u002FAPI understanding and environment\n  smokes for model dependencies.\n- Use `data-preparation` for dataset schema, builders, and visualization.\n- Use `training-evaluation` for Hydra training commands, checkpoint evaluation,\n  W&B, and ARC post-processing.\n- Do not present a CPU import as proof that HRM can train or evaluate. The\n  backend gate is CUDA + FlashAttention + adam-atan2 backend.\n\n## Bounded inspection\n\nFrom an HRM checkout, run:\n\n```bash\npython \u003Cskill>\u002Fsub-skills\u002Fmodel-architecture\u002Fscripts\u002Finspect_model_config.py \\\n  --repo-root \u002Fpath\u002Fto\u002FHRM --json\n```\n\nIf a CUDA environment is prepared and the user wants dependency readiness:\n\n```bash\npython \u003Cskill>\u002Fsub-skills\u002Fmodel-architecture\u002Fscripts\u002Finspect_model_config.py \\\n  --repo-root \u002Fpath\u002Fto\u002FHRM --cuda-smoke\n```\n\nThe helper verifies imports and dependency smokes. It intentionally avoids long\ntraining and does not require real datasets or checkpoints.\n",{"name":5,"license":24,"metadata":25,"description":7,"disable-model-invocation":27},"Apache 2.0",{"disco-role":26},"operating",true,"main","d69096e239f2a89aabc2cd078c47664a9d3b1e11","06-前端与设计\u002Fhrm\u002Fsub-skills\u002Fmodel-architecture",3,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts","https:\u002F\u002Fgithub.com\u002Flza6","个人vx：Tf00798\r\n个人✈：@TFGY999","2026-10-08 14:29:22",{"source":8,"skillName":5,"commands":37,"prompt":47,"zipAvailable":27},[38,41,44],{"tool":39,"command":40},"npx","npx skills add lza6\u002FArmor-piercing-system-prompts --skill model-architecture",{"tool":42,"command":43},"bunx","bunx skills add lza6\u002FArmor-piercing-system-prompts --skill model-architecture",{"tool":45,"command":46},"pnpm","pnpm dlx skills add lza6\u002FArmor-piercing-system-prompts --skill model-architecture","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Fhrm\u002Fsub-skills\u002Fmodel-architecture\u002FSKILL.md",[49,55,59],{"path":50,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":53,"sourceHtmlUrl":54},"references\u002Fapi-reference.md","REFERENCE","参考文档",3574,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Fhrm\u002Fsub-skills\u002Fmodel-architecture\u002Freferences\u002Fapi-reference.md",{"path":56,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":57,"sourceHtmlUrl":58},"references\u002Farchitecture.md",3305,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Fhrm\u002Fsub-skills\u002Fmodel-architecture\u002Freferences\u002Farchitecture.md",{"path":60,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":61,"sourceHtmlUrl":62},"references\u002Ftroubleshooting.md",3320,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Fhrm\u002Fsub-skills\u002Fmodel-architecture\u002Freferences\u002Ftroubleshooting.md",[64,72,77,84,91,98],{"skillNo":65,"name":66,"nameZh":6,"description":67,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":68,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":19},"01M7FFG485KY83TD6HSC4YPCMC","obsidian-clipper-template-creator","Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.","MIT","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator\u002FSKILL.md",[],[],{"skillNo":73,"name":66,"nameZh":6,"description":67,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":68,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M7FFG0T6AZ0STEJPSB3CX84K","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator-antigravity-awesome-skills-main\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":68,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":19},"01M7FFFY0QEM14SKWPDYGMXM2P","python-api","Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP\u002FmAP semantics, and source-style import diagnostics.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Fpython-api\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":68,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M7FFFVZQZ2GJ0C0TGP7JHD20","file-evaluation","Folder-based PASCAL VOC AP\u002FmAP evaluation from ground-truth and detection text files.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Ffile-evaluation\u002FSKILL.md",[],[],{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":68,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":19},"01M7FFFSFNYSE77SJ6PE1CK2DA","object-detection-metrics","Guides legacy Object-Detection-Metrics PASCAL VOC AP\u002FmAP evaluation from detection text folders and direct Python API classes.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":68,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":19},"01M7FFFP0GCK1RVXBCPQ8APBMT","nx-workspace-patterns","Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fnx-workspace-patterns\u002FSKILL.md",[],[]]