future_discriminator_data
Build prefix-level future-discriminator examples for FUDGE-style controlled generation from completed token sequences and sequence/target-word labels.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- future_discriminator_data
- description
- Build prefix-level future-discriminator examples for FUDGE-style controlled generation from completed token sequences and sequence/target-word labels.
02 资源文件 2
- 脚本scripts/prefix_data.py1.8 KB
- 脚本tests/test_prefix_data.py651 B
03 SKILL.md 正文
Future Discriminator Prefix Data
Use this skill when implementing or testing FUDGE-like future discriminators. It converts completed token sequences into prefix examples whose labels indicate whether an attribute or target word will hold in the completed future.
Do not use this skill to perform decoding or choose output tokens; its output is only training/evaluation data for a discriminator.
Inputs
- Completed token sequences as lists of tokens.
mode:whole_sequenceorfuture_suffix.- For
whole_sequence: one binary label per sequence. - For
future_suffix: a target token/word.
Outputs
A JSON-compatible object with examples and metadata. Each example contains prefix_tokens, label, position, target, and source_sequence_id.
Workflow
- Normalize tokens consistently before calling the script if case-insensitive labels are wanted.
- Expand every non-empty prefix of every completed sequence.
- In whole-sequence mode, copy the sequence attribute to every prefix.
- In future-suffix mode, label a prefix positive if the target appears in the remaining suffix including the current prefix-final token.
- Use the examples to train or audit a lightweight future discriminator.
Validation
Run python tests/test_prefix_data.py from this skill directory.
Limitations
The helper operates on already-tokenized sequences. It does not train a neural model and does not read the original paper repository.
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).