RLVR Training And Evaluation¶
Trace separates task generation from model training and external evaluation.
The main branch provides deterministic task generation, metadata-backed
verification, packaged resources, dataset validation, and export. The
rlvr branch provides the
Qwen2.5-VL 3B/7B training workflow, TRACE validation, and the 24-benchmark
trace_eval_v1 external evaluation used in the TRACE paper.
Training Contract¶
The two reproducible training profiles are:
trace-qwen2.5-vl-3b, based onQwen/Qwen2.5-VL-3B-Instruct@66285546d2b821cf421d4f5eb2576359d3770cd3;trace-qwen2.5-vl-7b, based onQwen/Qwen2.5-VL-7B-Instruct@cc594898137f460bfe9f0759e9844b3ce807cfb5.
Both profiles use
maveryn/trace@dataset-v1,
select prompt_answer, score answer_gt, run GRPO for 500 steps, and verify
the training prompt with SHA-256
f394927d9abcfb7a1e43ef48a30c29b8c70e6facdbda314d7b27c59d8c3ae900.
The machine-facing configurations pin the tag's immutable commit. A checked-in
equivalence receipt verifies that every model-consumed value, image byte, and
row position matches the historical training input. The optional
trace_supervision_mode column is advisory and is ignored by these profiles.
Environment setup, input preparation, smoke runs, full runs, checkpoint merge,
and receipt validation are documented in the
rlvr guide.
Verification Boundaries¶
Native Trace rewards score typed answers and optional annotations against the metadata contract emitted by the task generator. Those APIs are documented in Verifier contracts and covered by the package tests.
External benchmarks retain their official prompt, parsing, and scoring
contracts from pinned evaluator revisions. The trace_eval_v1 workflow records
every narrowly scoped adapter or post-run reproducibility patch. It does not
substitute a native Trace verifier for an official benchmark scorer.
Canonical External Evaluation¶
The evaluation covers 24 benchmarks for:
- Qwen2.5-VL-3B base and TRACE;
- Qwen2.5-VL-7B base, TRACE, VERO, Game-RL, Sphinx, and PCGRPO;
- decoding seeds
42,43, and44.
The generated
results.json
is the machine-readable result authority. It retains all 576
model/seed/benchmark scores and binds each aggregate to immutable run artifacts.
The adjacent
suite manifest,
benchmark provenance,
and
evaluation guide
define benchmark sources, revisions, splits, licenses, generation settings,
answer extraction, scoring, retry behavior, aggregation, and artifact checks.
Validate the frozen training inputs and evaluation metadata with:
python rlvr/train.py check --config trace-qwen2.5-vl-3b
python rlvr/train.py check --config trace-qwen2.5-vl-7b
python rlvr/evaluation/scripts/validate_release_inputs.py
Use immutable dataset, model, evaluator, judge, prompt, and result revisions in every run receipt. Do not replace them with mutable branch references or machine-specific paths.
TRACE Validation¶
The paper also evaluates the same eight checkpoints on two unseen instances
from each of TRACE's 1,000 task programs. This 2,000-instance, seed-42 campaign
uses its own frozen suite, preparation, generation, extraction, scoring, and
verification interfaces under
rlvr/evaluation/trace_validation.
It measures new realizations within the TRACE task distributions and is not a
held-out-task or external-transfer benchmark.
The checked-in result source contains the eight aggregate scores. The
corresponding response and score artifacts are pinned at
maveryn/trace-eval-runs@cf0d14aed86db2661d397ce8b68b36171873478d.
Test Boundary¶
The main branch tests task generation, schemas, verifiers, installation, CLI
behavior, manifests, and repository hygiene. The rlvr branch adds focused
training-profile and evaluator validation. Model-specific compatibility checks
belong beside the affected launcher when a reported result depends on them.