8000 Release miniVERL v0.5.0 by DaoyuanLi2816 · Pull Request #34 · DaoyuanLi2816/mini-verl · GitHub
[go: up one dir, main page]

Skip to content
8000
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 4 additions & 1 deletion CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,8 @@ All notable changes to miniVERL are recorded here. The format follows

## [Unreleased]

## [0.5.0] - 2026-08-02

### Added

- `miniverl align` for explicit base → SFT checkpoint → teacher/reference →
Expand Down Expand Up @@ -526,7 +528,8 @@ Same-tokenizer only; one trajectory per forward pass; `swap` unavailable for
quantized models; only Qwen3 and Qwen2 architectures tested; single-seed GPU
results. The full list is in `docs/limitations.md`.

[Unreleased]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.4.0...HEAD
[Unreleased]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.5.0...HEAD
[0.5.0]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.4.0...v0.5.0
[0.4.0]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.3.0...v0.4.0
[0.3.0]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.2.6...v0.3.0
[0.2.6]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.2.5...v0.2.6
Expand Down
4 changes: 2 additions & 2 deletions CITATION.cff
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@ cff-version: 1.2.0
title: "miniVERL: On-policy distillation for tool-using agents on one GPU"
message: "If you use miniVERL in your work, please cite it as below."
type: software
version: 0.4.0
version: 0.5.0
date-released: 2026-08-02
license: Apache-2.0
repository-code: "https://github.com/DaoyuanLi2816/mini-verl"
Expand Down Expand Up @@ -98,4 +98,4 @@ references:
url: "https://arxiv.org/abs/2603.07079"
notes: >-
Motivates recording per-token teacher entropy. Entropy-aware divergence
mixing is a roadmap item and is not implemented in miniVERL v0.4.0.
mixing is a roadmap item and is not implemented in miniVERL v0.5.0.
3 changes: 2 additions & 1 deletion PROJECT_STATE.md
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,8 @@ Last updated: 2026-08-02.
| presentation | four data-bound dark SVGs passed native and 820-pixel inspection; banner and social preview passed raster inspection; the report PDF is deterministic, six pages, metadata-valid and visually inspected page by page |
| local release gates | ruff, format, mypy, actionlint, generated-artifact checks, Markdown links and package/twine checks pass; the full non-GPU/non-network suite passes 1508 tests with 6 deselected at 86.13% branch coverage; the RTX 4080 gate passes 5 and the network gate passes 3; fresh core and CPU-training wheel installs, a real no-network toy demo, and all 1508 tests from an isolated extracted sdist pass |
| immutable baseline | calculator benchmark remains byte-identical at `53fc1d4d5b7adee09618d77ad62d4086ba56b78569832d6fc7c3bcd5c2695bbc`; v0.3/v0.4 tags, RecoveryBench artifacts and public adapter revisions remain unchanged |
| integration state | development branch `v0.5-alignment-lab` contains the frozen execution commit `8ffb2d9c3094fb6580e5afd89c861c8e0511ef78`; implementation and local release gates are complete, while PR review, merge, release and state sync remain pending |
| integration source | Alignment Lab PR [#33](https://github.com/DaoyuanLi2816/mini-verl/pull/33) was squash-merged as `f9dae54a203d303f6562101500343ead310d99e8`; synchronized main CI and build runs `30788116397` and `30788116373` are green |
| release state | development integration and post-merge validation are complete; this focused release-metadata change binds exact `0.5.0`, after which annotated-tag publication and `0.6.0.dev0` state sync remain |

## v0.4.0 Consumer Runtime release

Expand Down
86 changes: 43 additions & 43 deletions PYPI.md

Large diffs are not rendered by default.

2 changes: 1 addition & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@ safety-aligned teacher—without Ray or a cluster.
**Measure alignment, over-refusal, retained utility and cost before choosing
SFT, DPO or OPD.**

PyPI `v0.4.0` is the stable release; `main` is development and may be ahead.
PyPI `v0.5.0` is the stable release; `main` is development and may be ahead.

miniVERL is an independent one-GPU companion for prototyping, diagnosing and
validating online post-training workflows before scale-out. SFT establishes
Expand Down
2 changes: 1 addition & 1 deletion README.zh-CN.md
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,7 @@

**先同时测量对齐、过度拒绝、保留效用与成本,再选择 SFT、DPO 或 OPD。**

PyPI `v0.4.0` 是稳定发布版;`main` 是开发分支,可能领先于稳定版。
PyPI `v0.5.0` 是稳定发布版;`main` 是开发分支,可能领先于稳定版。

miniVERL 是一个独立的单卡配套工具,用于在扩展前原型化、诊断和验证在线
后训练流程。SFT 负责建立任务与协议能力;OPD 在学生实际访问的状态上转移
Expand Down
18 changes: 9 additions & 9 deletions docs/generated/quality.json
Original file line number Diff line number Diff line change
@@ -1,22 +1,22 @@
{
"schema_version": 1,
"release": "0.4.0",
"status": "released",
"measured_commit": "09df1ac1f8a8c33fa07bffdf3f24dad8000cc500",
"measured_at": "2026-08-02T18:19:00-07:00",
"release": "0.5.0",
"status": "validated",
"measured_commit": "f9dae54a203d303f6562101500343ead310d99e8",
"measured_at": "2026-08-02T22:50:12-07:00",
"cpu_non_gpu_non_network": {
"passed": 1448,
"passed": 1508,
"deselected": 6,
"branch_coverage_percent": 86.26
"branch_coverage_percent": 86.13
},
"gpu": {
"passed": 5,
"deselected": 1449,
"deselected": 1509,
"hardware": "NVIDIA GeForce RTX 4080"
},
"network": {
"passed": 3,
"deselected": 1451
"deselected": 1511
},
"quality_floor": "1,440+ tests and 86%+ branch coverage at v0.4.0"
"quality_floor": "1,500+ tests and 86%+ branch coverage at v0.5.0"
}
36 changes: 36 additions & 0 deletions docs/release-checklist.md
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,42 @@ This is the release gate and publication record for miniVERL. A checked item
names an invariant exercised on the stated source. Publication begins only
after the exact release commit and its remote checks are green.

## v0.5.0 One-GPU Alignment Lab release

- [x] Public preregistration revision 1.4 freezes three seeds, the common SFT
checkpoint, four continuation updates, 48 test tasks and the disjoint
seed-1234 recovery rule at SHA-256
`71307dbfe9a5bb20c686307cafce8bd254c07af8b69c1bf1c6ec0dbf53a8cde0`.
- [x] All 18 method-by-seed arms completed with 48 ordered paired tasks each;
864 task rows, strict freshness, verifier decisions and DPO provenance
pass exact publication checks. No completed final arm was rerun.
- [x] The saturated 100% SFT checkpoint, ties from DPO and offline soft
distillation, and regressions from continued SFT, standard OPD and
verifier-gated OPD are all reported without suppressing negative results.
- [x] The frozen result, task rows, state diagnostic and six-page PDF have
SHA-256 values `584752dccb91654109c357b8ebb12681a12a9c1476a9ba539dd35e4d860a22ef`,
`8d7fc723436d7377d196fc44046d960e3cb7f0aa81e03d49ef05b627eb84630f`,
`9e08129ba4cd9e460c189b94b4e421d881ba69e3938f02eac95d251f50c88788`
and `adbffa967f6b9a25d2cdb0cc4464a93c13db4615a1e91499585fb199285d980b`.
- [x] The legacy calculator JSON remains byte-identical at SHA-256
`53fc1d4d5b7adee09618d77ad62d4086ba56b78569832d6fc7c3bcd5c2695bbc`;
v0.3/v0.4 tags, artifacts and public adapter revisions are unchanged.
- [x] Ruff check/format, mypy, actionlint, Markdown links, generated-artifact
byte comparison and `git diff --check` pass; four SVGs, banner, social
preview and every PDF page were visually inspected.
- [x] Full non-GPU/non-network suite: **1508 passed**, 6 deselected, **86.13%**
branch coverage.
- [x] Available RTX 4080 GPU suite: **5 passed**, 1509 deselected.
- [x] Network suite: **3 passed**, 1511 deselected, including immutable public
adapters used by the release.
- [x] Wheel and sdist build and pass Twine; clean core and CPU-training wheel
installs, a real no-network toy demo, and the isolated extracted-sdist
suite pass at `miniverl 0.5.0.dev0`.
- [x] Focused PR #33 is green and merged as `f9dae54`; synchronized main CI
and build runs `30788116397` and `30788116373` are green.
- [x] Release metadata declares exact `0.5.0`; the intended annotated tag is
exactly `v0.5.0` and will use the existing OIDC-only release workflow.

## v0.4.0 Consumer Runtime release

- [x] Preregistration revisions 1.1 and 1.2 were public before the sole
Expand Down
2 changes: 1 addition & 1 deletion src/miniverl/__init__.py
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,6 @@

from __future__ import annotations

__version__ = "0.5.0.dev0"
__version__ = "0.5.0"

__all__ = ["__version__"]
4 changes: 2 additions & 2 deletions tests/unit/test_packaging.py
Original file line number Diff line number Diff line change
Expand Up @@ -386,7 +386,7 @@ def test_generated_pypi_readme_is_byte_bound_and_has_only_navigable_project_link

@pytest.mark.parametrize(
("version", "ref"),
[("0.4.0", "v0.4.0"), ("0.5.0.dev0", "main")],
[("0.5.0", "v0.5.0"), ("0.6.0.dev0", "main")],
)
def test_pypi_generator_rewrites_nested_images_and_every_project_target(
tmp_path: Path,
Expand Down Expand Up @@ -446,7 +446,7 @@ def test_release_quality_has_one_version_bound_machine_readable_record() -> None
(REPO_ROOT / "docs" / "generated" / "quality.json").read_text(encoding="utf-8")
)
assert record["schema_version"] == 1
assert record["quality_floor"] == "1,440+ tests and 86%+ branch coverage at v0.4.0"
assert record["quality_floor"] == "1,500+ tests and 86%+ branch coverage at v0.5.0"
if ".dev" in miniverl.__version__:
assert record["status"] in {"candidate", "released"}
if record["status"] == "candidate":
Expand Down
0