AGENT SKILL

qdagger_loss

Compute the QDagger objective for policy-to-value reincarnating RL using n-step TD loss plus teacher-policy distillation.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
qdagger_loss
description
Compute the QDagger objective for policy-to-value reincarnating RL using n-step TD loss plus teacher-policy distillation.

02 资源文件 2

03 SKILL.md 正文

QDagger Loss

When To Use

Use this skill when implementing or validating a reduced or full QDagger-style policy-to-value reincarnating RL experiment. It is appropriate when you have transition records, student Q-values, and teacher action probabilities. Do not use it as a standalone imitation-only objective; the TD and distillation terms must remain separately visible.

Inputs

  • q_values: mapping from state ids to action-value lists.
  • transitions: records with state, action, n_step_return, discount, next_max_q, and teacher_policy.
  • temperature: positive softmax temperature for deriving the student policy from Q-values.
  • lambda_t: current distillation coefficient.

Outputs

  • td_loss: mean squared Bellman error.
  • distillation_loss: mean teacher cross-entropy against the student softmax policy.
  • combined_loss: td_loss + lambda_t * distillation_loss.
  • examples: per-transition targets, predictions, and loss terms.

Workflow

  1. Validate that every transition has a known state, valid action index, and normalized teacher policy.
  2. Compute each n-step TD target as n_step_return + discount * next_max_q.
  3. Compute mean squared TD loss over selected actions.
  4. Convert each student Q-vector into softmax(q / temperature).
  5. Compute teacher-policy cross-entropy.
  6. Return both component losses and the combined QDagger loss.

Validation

Run:

python scripts/qdagger_loss.py --self-test
python tests/test_qdagger_loss.py

Limitations

This implementation is deterministic and dependency-free. It is intended for recovery harnesses, tests, and small tabular/scalar checks, not high-throughput deep RL training.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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).

★ 351