skills.rest
ProgMastermind
ProgMastermindCOMMUNITY
Explore all skills by this author →
Website𝕏@PrakashCollymo1GitHub

dump-bisect-debug

Compare intermediate tensor outputs between target and reference neural network implementations.

Updated Jul 15, 2026
Search Tags:#neural network#bug debugging#tensor comparison#bisecting methodology#batch invariance
One-click install
npx skills add https://github.com/ProgMastermind/ATOM --skill dump-bisect-debug
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: dump-bisect-debug
Source: https://github.com/ProgMastermind/ATOM/tree/main/.claude/skills/dump-bisect-debug
Command: npx skills add https://github.com/ProgMastermind/ATOM --skill dump-bisect-debug

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

The dump-bisect-debug Skill identifies and resolves forward numerical bugs in neural network models by comparing the outputs of the target implementation against a known-good reference implementation.

Core Features & Use Cases

  • Identify Bugs: Locate forward numerical bugs by comparing intermediate tensors from the target implementation and a known-good reference.
  • Bisecting Methodology: Utilize a methodology that reduces the time to find bugs from hours to minutes by systematically bisecting layers and sub-stages.
  • Batch Invariance: Handle batch-invariance bisect for models expecting identical outputs across different batch slots.
  • Use Case: When dealing with a model producing incorrect outputs but unable to pinpoint the issue through code review, the Skill can trace the problem back to its root.

Quick Start

Use the dump-bisect-debug Skill to bisect and debug a neural network model by following the outlined methodology and using provided tools.

Frequently Asked Questions about dump-bisect-debug

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I debug forward numerical bugs in a neural network model?▼

Bisecting methodology reduces debugging time by systematically narrowing down layers and sub-stages, comparing intermediate tensors between target and reference implementations to isolate the exact location of forward numerical bugs.

Can I trace batch invariance issues when comparing neural network tensors?▼

Yes, you can handle batch-invariance bisecting for models that expect identical outputs across different batch slots, ensuring tensor comparisons remain accurate when validating your target implementation against the reference.

What do I need to debug forward numerical bugs using tensor comparison?▼

You need knowledge of the model architecture and access to a known-good reference implementation to compare intermediate tensor outputs and successfully bisect layers to locate forward numerical bugs.

Why does my neural network produce incorrect outputs when the code looks correct?▼

Forward numerical bugs often hide in intermediate tensor computations rather than visible code logic; bisecting layer outputs against a reference implementation traces the problem back to its exact root cause.

What is the best way to locate a numerical bug across multiple neural network layers?▼

The best way is a bisecting methodology that systematically splits layers and sub-stages, comparing intermediate tensors from target and reference implementations to reduce bug isolation time from hours to minutes.

Related Skills

Explore More Software Engineering Skills→
DeepLink-orgDeepLink-org
74

precision-align

Diagnose precision mismatches between lmdeploy+dlinfer and reference implementations on Ascend, CAMB, and MACA backends.

Official
Advanced
Devsoul2026Devsoul2026
150

kanban-codex-lane

Spawn Codex in an isolated worktree for Hermes Kanban tasks.

Community
Advanced
dyyounggdyyoungg

debug-cuda-crash

Capture CUDA crash boundaries and tensor metadata via kernel API logging in SGLang.

Community
Advanced
nanorepublicananorepublica
61

Testing Unit Tests

Standardize unit-test coding practices across Claude Code projects.

Community
Basic