fudge_logit_rescoring
Apply the FUDGE decoding rule by combining base candidate logits with future-discriminator probabilities and renormalizing.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- fudge_logit_rescoring
- description
- Apply the FUDGE decoding rule by combining base candidate logits with future-discriminator probabilities and renormalizing.
02 资源文件 2
- 脚本scripts/rescore.py1.0 KB
- 脚本tests/test_rescoring.py712 B
03 SKILL.md 正文
FUDGE Logit Rescoring
Use this skill when a base autoregressive generator supplies candidate next-token logits and a future discriminator can score each candidate-extended prefix. The skill implements logit' = base_logit + strength * log P(attribute | prefix+candidate).
Inputs
prefix_tokens.- Mapping of candidate token to base logit.
- Mapping/callable of candidate token to future probability.
- Optional
strength,top_k, and clipping epsilon.
Outputs
- Normalized probabilities over retained candidates.
- Audit trace with base logit, future probability, future log score, adjusted logit, and final probability.
Workflow
- Restrict to top-k base candidates if requested.
- Score each extended prefix with the future discriminator.
- Clip probabilities for numerical safety.
- Add future log scores to base logits.
- Softmax over retained candidates.
- Inspect the audit trace before claiming FUDGE was used.
Validation
Run python tests/test_rescoring.py.
Limitations
This helper does not load a language model. It is a deterministic FUDGE rescoring kernel.
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).