contrastive_training_objective
Compute and test CLIP-style symmetric contrastive loss with normalized image-text similarities.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- contrastive_training_objective
- description
- Compute and test CLIP-style symmetric contrastive loss with normalized image-text similarities.
02 资源文件 2
03 SKILL.md 正文
Contrastive Training Objective
When To Use
Use this skill when reproducing or testing the CLIP paper mechanism in a bounded setting. It is designed for generated recovery workflows and does not require the original CLIP repository.
Inputs
- Small image-text feature records or class prompt features.
- A module plan target describing whether the run is full or reduced/proxy.
- Standard Python 3; no required external packages.
Outputs
- Deterministic validation results or helper values suitable for recovery artifacts.
- Clear errors when the input violates the CLIP contract.
Workflow
- Keep image and text rows aligned so the diagonal represents positive matches.
- Normalize embeddings before similarity scoring.
- Use symmetric image-to-text and text-to-image evidence for training objectives when applicable.
- For zero-shot evaluation, synthesize classifier weights from prompt text rather than supervised downstream labels.
- Record whether the run is a reduced proxy and never claim full ImageNet-scale recovery without real model and data evidence.
Validation
Run python /share/project/yuyang/workspace/Paper2Skills/Paper2Skills-Agent/src/packages/paper2skills-agent/src/paper2skills/skills/module-to-skill/scripts/validate_skill_tree.py <skill_dir> --run-tests.
Limitations
This skill preserves the paper mechanism but does not include OpenAI CLIP weights, ImageNet data, or the original repository. Full reproduction requires those external resources and a much larger runtime.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).