[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6KD54C3KP6E6EPASRGQ3GNS":3,"skill-same:01M6KD54C3KP6E6EPASRGQ3GNS":53},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":26,"linkOnly":27,"markdownContent":28,"markdownBody":29,"frontMatter":30,"refName":31,"commitSha":32,"skillDirectory":33,"forksCount":34,"repositoryHtmlUrl":35,"repositoryHomepageUrl":36,"ownerHtmlUrl":37,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":38,"install":39,"resources":52},"01M6KD54C3KP6E6EPASRGQ3GNS","pytorch-loss-implementation",null,"Implement loss functions in PyTorch with proper tensor operations.","cxcscmu\u002FSkillLearnBench","cxcscmu","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F142838272?v=4","01M6KCQDFQWPDSN0MKZY96MSQP","01M6KCQDM7G70TKJ28NTSQC0RZ",93,"MIT","https:\u002F\u002Fgithub.com\u002Fcxcscmu\u002FSkillLearnBench\u002Fblob\u002Fmain\u002Fskills\u002Fb1-one-shot-claude-haiku-4-5\u002Fnlp-paper-reproduction\u002Fpytorch-loss-implementation\u002FSKILL.md",[],[18,19,20,21,22,23,24,25],"pytorch","implementation","pytorch loss implementation","implement","tensor","operations","functions","with","2026-11-10 10:12:30",false,"---\nname: pytorch-loss-implementation\ndescription: Implement loss functions in PyTorch with proper tensor operations.\n---\n\n# PyTorch Loss Implementation\n\n## Key Concepts\n\n### 1. Tensor Operations\n```python\nimport torch\n\n# Sigmoid function\nsigmoid_output = torch.sigmoid(input_tensor)\n\n# Log function\nlog_output = torch.log(input_tensor)\n\n# Mean reduction\nmean_loss = loss.mean()\n\n# Sum reduction\nsum_loss = loss.sum()\n```\n\n### 2. Batch Processing\n```python\n# Batch dimension handling\nbatch_size = tensor.shape[0]\nx = tensor[:batch_size\u002F\u002F2]  # First half\ny = tensor[batch_size\u002F\u002F2:]  # Second half\n\n# Ensure same device and dtype\ntensor = tensor.to(device=model.device, dtype=torch.float32)\n```\n\n### 3. Numerical Stability\n\n**Log-Sigmoid Stability**\n```python\n# Avoid: log(sigmoid(x)) can cause numerical issues\n# Instead use:\nstable_loss = torch.nn.functional.logsigmoid(x)\n# Or manually:\nloss = -torch.log(torch.sigmoid(x) + 1e-10)\n```\n\n**Handling Small Values**\n```python\n# Add epsilon to avoid log(0)\nsafe_log = torch.log(value + 1e-8)\n\n# Clamp to valid range\nclamped = torch.clamp(value, min=1e-10, max=1.0)\n```\n\n### 4. Device Handling\n```python\n# Ensure all tensors on same device\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\ntensor = tensor.to(device)\n\n# Or get from model\ndevice = next(model.parameters()).device\ntensor = tensor.to(device)\n```\n\n## Loss Function Pattern\n\n```python\ndef compute_loss(logits, labels, temperature=1.0, margin=0.5):\n    # 1. Normalize\u002Fcompute rewards\n    rewards = logits \u002F (sequence_length + 1e-8)\n\n    # 2. Compute differences\n    diff = temperature * rewards[:n\u002F\u002F2] - temperature * rewards[n\u002F\u002F2:] - margin\n\n    # 3. Apply objective\n    loss_per_pair = -torch.log(torch.sigmoid(diff) + 1e-10)\n\n    # 4. Reduce\n    loss = loss_per_pair.mean()\n\n    return loss\n```\n\n## Debugging Tips\n\n1. **Check tensor shapes** at each step\n2. **Use .detach()** for inspecting values without affecting gradients\n3. **Verify numerical stability** with small inputs\n4. **Test gradients** with `loss.backward()`\n5. **Print intermediate values** for debugging\n\n## Performance Tips\n\n- Use in-place operations where safe: `tensor.log_()`\n- Avoid unnecessary cloning\u002Fcopying\n- Batch operations are faster than loops\n- Use PyTorch functions over custom loops\n","# PyTorch Loss Implementation\n\n## Key Concepts\n\n### 1. Tensor Operations\n```python\nimport torch\n\n# Sigmoid function\nsigmoid_output = torch.sigmoid(input_tensor)\n\n# Log function\nlog_output = torch.log(input_tensor)\n\n# Mean reduction\nmean_loss = loss.mean()\n\n# Sum reduction\nsum_loss = loss.sum()\n```\n\n### 2. Batch Processing\n```python\n# Batch dimension handling\nbatch_size = tensor.shape[0]\nx = tensor[:batch_size\u002F\u002F2]  # First half\ny = tensor[batch_size\u002F\u002F2:]  # Second half\n\n# Ensure same device and dtype\ntensor = tensor.to(device=model.device, dtype=torch.float32)\n```\n\n### 3. Numerical Stability\n\n**Log-Sigmoid Stability**\n```python\n# Avoid: log(sigmoid(x)) can cause numerical issues\n# Instead use:\nstable_loss = torch.nn.functional.logsigmoid(x)\n# Or manually:\nloss = -torch.log(torch.sigmoid(x) + 1e-10)\n```\n\n**Handling Small Values**\n```python\n# Add epsilon to avoid log(0)\nsafe_log = torch.log(value + 1e-8)\n\n# Clamp to valid range\nclamped = torch.clamp(value, min=1e-10, max=1.0)\n```\n\n### 4. Device Handling\n```python\n# Ensure all tensors on same device\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\ntensor = tensor.to(device)\n\n# Or get from model\ndevice = next(model.parameters()).device\ntensor = tensor.to(device)\n```\n\n## Loss Function Pattern\n\n```python\ndef compute_loss(logits, labels, temperature=1.0, margin=0.5):\n    # 1. Normalize\u002Fcompute rewards\n    rewards = logits \u002F (sequence_length + 1e-8)\n\n    # 2. Compute differences\n    diff = temperature * rewards[:n\u002F\u002F2] - temperature * rewards[n\u002F\u002F2:] - margin\n\n    # 3. Apply objective\n    loss_per_pair = -torch.log(torch.sigmoid(diff) + 1e-10)\n\n    # 4. Reduce\n    loss = loss_per_pair.mean()\n\n    return loss\n```\n\n## Debugging Tips\n\n1. **Check tensor shapes** at each step\n2. **Use .detach()** for inspecting values without affecting gradients\n3. **Verify numerical stability** with small inputs\n4. **Test gradients** with `loss.backward()`\n5. **Print intermediate values** for debugging\n\n## Performance Tips\n\n- Use in-place operations where safe: `tensor.log_()`\n- Avoid unnecessary cloning\u002Fcopying\n- Batch operations are faster than loops\n- Use PyTorch functions over custom loops\n",{"name":5,"description":7},"main","726e1f79b96b0cb3d729d0e55a1c847e0b5353ae","skills\u002Fb1-one-shot-claude-haiku-4-5\u002Fnlp-paper-reproduction\u002Fpytorch-loss-implementation",8,"https:\u002F\u002Fgithub.com\u002Fcxcscmu\u002FSkillLearnBench","https:\u002F\u002Fcxcscmu.github.io\u002FSkillLearnBench","https:\u002F\u002Fgithub.com\u002Fcxcscmu","2026-10-02 22:55:36",{"source":8,"skillName":5,"commands":40,"prompt":50,"zipAvailable":51},[41,44,47],{"tool":42,"command":43},"npx","npx skills add cxcscmu\u002FSkillLearnBench --skill pytorch-loss-implementation",{"tool":45,"command":46},"bunx","bunx skills add cxcscmu\u002FSkillLearnBench --skill pytorch-loss-implementation",{"tool":48,"command":49},"pnpm","pnpm dlx skills add cxcscmu\u002FSkillLearnBench --skill pytorch-loss-implementation","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fcxcscmu\u002FSkillLearnBench\u002Fblob\u002Fmain\u002Fskills\u002Fb1-one-shot-claude-haiku-4-5\u002Fnlp-paper-reproduction\u002Fpytorch-loss-implementation\u002FSKILL.md",true,[],[54,72,97,115,135,147],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":65,"publishedAt":26},"01M6KM5KH175P5ABP5JMX2SBDQ","object_counter","Count occurrences of an object in the image using computer vision algorithm.","https:\u002F\u002Fgithub.com\u002Fcxcscmu\u002FSkillLearnBench\u002Fblob\u002Fmain\u002Fskills\u002Fhuman_authored\u002Fvideo-object-counting\u002Fobject_counter\u002FSKILL.md",[60],{"categoryNo":61,"categoryName":62,"taxonomyType":63,"taxonomyTypeDesc":64},"01JBSEEDCAT000000000000016","图像与多媒体","SCENARIO","使用场景",[66,67,68,69,70,71],"object counter","object","vision","computer","Computer Vision","image",{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":81,"publishedAt":26},"01M6KM51TMVXG79MMPCS6Y8G86","travel-planning-workflow","Invoke this skill when performing travel planning. It defines a structured workflow for generating itineraries. Following this workflow is essential to ensure the final plan satisfies user requirements.","https:\u002F\u002Fgithub.com\u002Fcxcscmu\u002FSkillLearnBench\u002Fblob\u002Fmain\u002Fskills\u002Fhuman_authored\u002Ftravel-planning\u002Ftravel-planning-workflow\u002FSKILL.md",[78],{"categoryNo":79,"categoryName":80,"taxonomyType":63,"taxonomyTypeDesc":64},"01JBSEEDCAT000000000000023","Skill与智能体开发",[82,83,84,85,86,87,88,89,90,91,92,93,94,95,96],"workflow","planning","planning workflow","travel","travel planning","travel planning workflow","structured","performing","plan","for","user","it","requirements","essential","itineraries",{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":26},"01M6KM4DY820XHS9YTSK5QP5HD","bundled-dataset-search-tool","invoke this skill when you need to perform database search for travel planning. This skill provides some useful pre-packaged tools to look up accommodations, attractions, cities, driving distance, flights, and restaurants from the bundled dataset.","https:\u002F\u002Fgithub.com\u002Fcxcscmu\u002FSkillLearnBench\u002Fblob\u002Fmain\u002Fskills\u002Fhuman_authored\u002Ftravel-planning\u002Fbundled-dataset-search-tool\u002FSKILL.md",[],[104,105,106,107,108,109,110,85,111,112,86,113,91,114,83],"search","tool","dataset","bundled dataset search tool","restaurants","Tools","up","look","flights","you","database",{"skillNo":116,"name":117,"nameZh":6,"description":118,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":119,"categories":120,"keywords":121,"publishedAt":26},"01M6KM465BB9PQ0B548HQSVAWZ","glm-output","Output-processing guidance for GLM tasks. Especially useful after glm-basics and glm-calibration when you need verifier-matching metrics from output.nc and a final \u002Froot\u002Fmetrics.json.","https:\u002F\u002Fgithub.com\u002Fcxcscmu\u002FSkillLearnBench\u002Fblob\u002Fmain\u002Fskills\u002Fhuman_authored\u002Ftemperature-simulation\u002Fglm-output\u002FSKILL.md",[],[122,123,124,91,125,126,127,128,129,130,131,132,113,133,134],"glm output","output","glm","tasks","metrics","json","basics","guidance","matching","verifier","processing","root","calibration",{"skillNo":136,"name":137,"nameZh":6,"description":138,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":139,"categories":140,"keywords":142,"publishedAt":26},"01M6KM40KY3QWDW5W97S39R2BC","glm-calibration","Calibration guidance for GLM tasks. Often most effective after glm-basics has clarified the setup; glm-output is the companion skill for exact final metric computation.","https:\u002F\u002Fgithub.com\u002Fcxcscmu\u002FSkillLearnBench\u002Fblob\u002Fmain\u002Fskills\u002Fhuman_authored\u002Ftemperature-simulation\u002Fglm-calibration\u002FSKILL.md",[141],{"categoryNo":79,"categoryName":80,"taxonomyType":63,"taxonomyTypeDesc":64},[143,134,124,123,129,128,125,144,145,91,146],"glm calibration","setup","companion","metric",{"skillNo":148,"name":149,"nameZh":6,"description":150,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":151,"categories":152,"keywords":153,"publishedAt":26},"01M6KM3VHHVSGMGHJQB3PPT6YG","glm-basics","A good starting point for GLM calibration tasks. Use to inspect glm3.nml, confirm how GLM runs, and identify the relevant files before moving on to calibration and output evaluation.","https:\u002F\u002Fgithub.com\u002Fcxcscmu\u002FSkillLearnBench\u002Fblob\u002Fmain\u002Fskills\u002Fhuman_authored\u002Ftemperature-simulation\u002Fglm-basics\u002FSKILL.md",[],[154,124,128,155,143,134,91,123,156,157,158,159,160,161,125],"glm basics","inspect","relevant","files","good","use","how","Evaluation"]