FFFF Start v0.4.0 development by DaoyuanLi2816 · Pull Request #29 · DaoyuanLi2816/mini-verl · GitHub
[go: up one dir, main page]

Skip to content
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
11 changes: 9 additions & 2 deletions PROJECT_STATE.md
Original file line number Diff line number Diff line change
Expand Up @@ -6,15 +6,15 @@ and what it printed.

Last updated: 2026-08-01.

## v0.3.0 RecoveryBench development
## v0.3.0 RecoveryBench release

| item | current state |
| --- | --- |
| audited baseline | fetched all remotes and started `agent/v0.3-recoverybench` from clean public `main` at `6c79c1fcc5a6e55dd9f4af843d446ece1c454431`; no PR was open, main CI/build were green, and development advanced directly from `0.2.7.dev0` to `0.3.0.dev0` without a `0.2.7` release |
| available execution environment | Windows checkout with Python 3.12, Torch 2.13.0+cu130 and an NVIDIA GeForce RTX 4080 (16376 MiB); GitHub and Hugging Face access are available and publication remains restricted to exact validated tags |
| immutable baseline | calculator benchmark SHA-256 is `53fc1d4d5b7adee09618d77ad62d4086ba56b78569832d6fc7c3bcd5c2695bbc`; calculator protocol-teacher HEAD is `23323751318135484c06c043b1f9b9e7016dd89f`; existing tags remain untouched |
| phase boundary | RecoveryBench must be preregistered, fully measured with the sequential engine, frozen, merged, released as `v0.3.0`, publicly verified and state-synced to `0.4.0.dev0` before padded batching begins |
| current highest-risk work | final experiment, artifact audit and data-bound publication are complete; remaining work is release validation, integration, Trusted Publishing verification and the required `0.4.0.dev0` state sync |
| current highest-risk work | RecoveryBench, release validation, integration and Trusted Publishing verification are complete; the next phase is the separately gated v0.4 consumer batched runtime |
| RecoveryBench environment | `sqlite_recovery` has 12 structurally disjoint versioned templates, deterministic controlled/natural/no-intervention subsets, structured tool-error provenance, executable recovery oracles, and exact recovery metrics; focused environment and backward-read tests pass |
| preregistration | public commit `7087b3a333463b88a62ffed73daee2c85d039145` and revision-1.3 digest `9c4c2ec19a56cebb2b2c1c0f3c7e504a9285467c99ae1590488251fbf2ff3934` bind the final procedure; the later public wall-budget amendment was reverted, its partial replacement stopped, and the already-complete frozen experiment was retained without rerun |
| teacher selection so far | the historical calculator teacher completed 96 eval tasks and failed the gate at 25.0% strict, 10.7% recovery, 95.8% parse validity and 14.4% tool execution; candidate A trained for 64 QLoRA SFT updates and its in-process NF4 eval measured 86.5%, 78.1%, 100% and 87.5%; a separately loaded full-precision reapplication failed at 65.6%, 65.6%, 100% and 62.7% and is retained as a noncanonical diagnostic |
Expand All @@ -32,6 +32,13 @@ Last updated: 2026-08-01.
| secondary budgets | all equal-selected-position arms crossed the 6,224 target after eight steps with recorded overshoot; the nominal 50-second artifact is explicitly a cycle-capped wall diagnostic because SFT and frozen KD completed eight cycles while fresh OPD crossed the target in one indivisible step |
| immutable RecoveryBench results | equal-updates `6ce2e6837e12b99ebc4fad6d27ce3e69c92e295ff3b9b60e0f68c2d308022384`; equal-selected-positions `fe4c9afc799724dfe7a32e631676a1e5177c44559a7374d2ea31da135354f137`; wall diagnostic `425b0fa568f37b09e61af731d3da5009bd3833bddde6efaf2c66e9dba8355cbe`; task JSONL `aff96bffc6da27240a852410ac041bd4d95badf34cad030e6f437be1491a55ad`; paired analysis `8a6891f74aed80f07ec00d5ea1909895c579346e1abbb1d5d95a354bb46c6b81` |
| technical publication | the generated Markdown analysis, three SVGs and deterministic six-page PDF report are data-bound to the frozen JSON; native/README-width SVG inspection and every-page PDF inspection found no clipping or overlap |
| integration source | RecoveryBench PR [#27](https://github.com/DaoyuanLi2816/mini-verl/pull/27) was squash-merged as `bee82d3a6e2c8a5c3f4c62c0d3827a07483a1977`; release-metadata PR [#28](https://github.com/DaoyuanLi2816/mini-verl/pull/28) was squash-merged as `624c6a352db53e0bd2038c4aeb0669a16a402239`; annotated tag `v0.3.0` resolves to the latter exact commit |
| version transition | `v0.3.0` is immutable and public; this state-sync change identifies subsequent development as `0.4.0.dev0` |
| release execution | tag run [`30772772078`](https://github.com/DaoyuanLi2816/mini-verl/actions/runs/30772772078) passed metadata, full tests, one-time build, OIDC publication with attestations, public verification, exact clean install and GitHub Release creation |
| published wheel | [`miniverl-0.3.0-py3-none-any.whl`](https://pypi.org/project/miniverl/0.3.0/), SHA-256 `e42404ced88b75ba4ff31541cb3df697da0eee09a68f603e4eadbf69a11d2032` |
| published sdist | [`miniverl-0.3.0.tar.gz`](https://pypi.org/project/miniverl/0.3.0/), SHA-256 `006ce418243286dd27e0731090bf9fa1711a1abc962ecfcc4d39807257539cb2` |
| independent public verification | GitHub Release downloads reproduce both public hashes; a no-cache Windows Python 3.10 install from `https://pypi.org/simple` reported `miniverl 0.3.0`, passed JSON `doctor`, and kept torch, Transformers, PEFT and bitsandbytes absent |
| release state | complete; PyPI and [`miniVERL v0.3.0`](https://github.com/DaoyuanLi2816/mini-verl/releases/tag/v0.3.0) expose the same verified wheel and sdist |

## v0.2.6 concurrency, lifecycle and privacy correctness release

Expand Down
62 changes: 31 additions & 31 deletions PYPI.md
F438
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
<p align="center">
<img src="https://raw.githubusercontent.com/DaoyuanLi2816/mini-verl/v0.3.0/docs/banner.svg" alt="miniVERL — auditable online post-training on one GPU" width="880">
<img src="https://raw.githubusercontent.com/DaoyuanLi2816/mini-verl/main/docs/banner.svg" alt="miniVERL — auditable online post-training on one GPU" width="880">
</p>

<div align="center">
Expand All @@ -8,14 +8,14 @@
[![Build](https://github.com/DaoyuanLi2816/mini-verl/actions/workflows/build.yml/badge.svg)](https://github.com/DaoyuanLi2816/mini-verl/actions/workflows/build.yml)
[![PyPI](https://img.shields.io/pypi/v/miniverl.svg)](https://pypi.org/project/miniverl/)
[![Python](https://img.shields.io/badge/python-3.10%20%7C%203.11%20%7C%203.12%20%7C%203.13-blue)](https://www.python.org)
[![License](https://img.shields.io/badge/license-Apache--2.0-blue)](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/LICENSE)
[![License](https://img.shields.io/badge/license-Apache--2.0-blue)](https://github.com/DaoyuanLi2816/mini-verl/blob/main/LICENSE)

</div>

<p align="center">
<a href="https://pypi.org/project/miniverl/"><strong>PyPI package</strong></a> ·
<a href="#single-gpu-quickstart">Install &amp; train</a> ·
<a href="https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/single-gpu-guide.md">Bring your own GPU</a> ·
<a href="https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/single-gpu-guide.md">Bring your own GPU</a> ·
<a href="#recoverybench-do-fresh-on-policy-states-justify-their-cost">Measured result</a>
</p>

Expand Down Expand Up @@ -59,7 +59,7 @@ validate artifacts without downloading a multi-gigabyte ML stack; use
[Run the local demo](#local-toy-demo) ·
[Train on your GPU](#single-gpu-quickstart) ·
[Inspect the measured result](#recoverybench-do-fresh-on-policy-states-justify-their-cost) ·
[Read the math](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/math.md)
[Read the math](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/math.md)

## Why miniVERL exists

Expand Down Expand Up @@ -98,7 +98,7 @@ keeps the whole lifecycle in one readable single-GPU process.
| Calculator, JSON-navigation and SQLite environments | yes, deterministic with exact verifiers |
| Exact checkpoint/resume | yes, asserted parameter-for-parameter |
| Self-contained offline HTML report with token-level divergence | yes |
| Ray, FSDP, DeepSpeed, vLLM, VLMs, cross-tokenizer, PPO/GRPO | **no** — see [limitations](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/limitations.md) |
| Ray, FSDP, DeepSpeed, vLLM, VLMs, cross-tokenizer, PPO/GRPO | **no** — see [limitations](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/limitations.md) |

## RecoveryBench: do fresh on-policy states justify their cost?

Expand All @@ -122,7 +122,7 @@ All three seeds and all completed negative results are retained.
| strict fresh-state OPD | 10.9% | 9.1% | 686.8 s |
| budget-50 fresh-state OPD | 27.3% | 20.7% | 720.8 s |

![RecoveryBench three-seed result](https://raw.githubusercontent.com/DaoyuanLi2816/mini-verl/v0.3.0/docs/recoverybench/recovery-success.svg)
![RecoveryBench three-seed result](https://raw.githubusercontent.com/DaoyuanLi2816/mini-verl/main/docs/recoverybench/recovery-success.svg)

The equal-selected-position view reached the 6,224-position boundary after
eight updates for every core method, so its quality result matches the primary
Expand All @@ -132,9 +132,9 @@ did not reduce wall time because teacher backbone forwards were unchanged. The
evidence**: SFT and frozen KD completed their eight-cycle ceiling, while fresh
OPD crossed the target in one indivisible 88-121 second update.

Read the [full analysis](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/recoverybench/recoverybench-v1.md), the
[data-bound technical report](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/paper/recoverybench-v1/recoverybench-v1.pdf), or
the [immutable schema-v3 artifacts](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/benchmarks/README.md#recoverybench-v1).
Read the [full analysis](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/recoverybench/recoverybench-v1.md), the
[data-bound technical report](https://github.com/DaoyuanLi2816/mini-verl/blob/main/paper/recoverybench-v1/recoverybench-v1.pdf), or
the [immutable schema-v3 artifacts](https://github.com/DaoyuanLi2816/mini-verl/blob/main/benchmarks/README.md#recoverybench-v1).
The result is scoped to one Qwen3 pair, one task family, three seeds and one RTX
4080. It does not show that OPD is universally ineffective or that offline KD
always wins.
Expand All @@ -148,18 +148,18 @@ time. Two protocol-naive controls completed normally at 0%; they were not
configuration failures. Both used the ambiguous historical protocol-v1 prompt,
so the failure cannot be attributed solely to intrinsic teacher behavior.

![Two-seed protocol-teacher benchmark](https://raw.githubusercontent.com/DaoyuanLi2816/mini-verl/v0.3.0/docs/gpu-calc-hard-equal-update-v2.svg)
![Two-seed protocol-teacher benchmark](https://raw.githubusercontent.com/DaoyuanLi2816/mini-verl/main/docs/gpu-calc-hard-equal-update-v2.svg)

| Artifact | Role |
| --- | --- |
| [Default recipe](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/recipes/qwen_consumer_gpu_calc.yaml) | protocol-qualified default |
| [Schema-v2 benchmark](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/benchmarks/results/gpu-calc-hard-equal-update-v2.json) | frozen five-arm result |
| [Raw-teacher recipe](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/recipes/qwen_consumer_gpu_calc_raw_teacher.yaml) | historical control; not default |
| [Default recipe](https://github.com/DaoyuanLi2816/mini-verl/blob/main/recipes/qwen_consumer_gpu_calc.yaml) | protocol-qualified default |
| [Schema-v2 benchmark](https://github.com/DaoyuanLi2816/mini-verl/blob/main/benchmarks/results/gpu-calc-hard-equal-update-v2.json) | frozen five-arm result |
| [Raw-teacher recipe](https://github.com/DaoyuanLi2816/mini-verl/blob/main/recipes/qwen_consumer_gpu_calc_raw_teacher.yaml) | historical control; not default |

The teacher gate and downstream comparison reused the same 24-task v0.2 test
set, so this is evidence for qualification in that setup, not a general OPD
advantage. The separate schema-v1 481-second smoke proves the pipeline, not OPD
over SFT. [Full diagnosis and caveats](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/rtx4080-baselines.md).
over SFT. [Full diagnosis and caveats](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/rtx4080-baselines.md).

</details>

Expand Down Expand Up @@ -236,7 +236,7 @@ bf16, while older CUDA cards such as Titan V use fp16. RTX 3070, Titan V,
RTX 4080 and RTX 5090-class cards all enter the same code path; only the
RTX 4080 result is measured here. Exact fit is governed by VRAM, model sizes,
drivers and token budgets, not the card's marketing name. See the
[`single-GPU guide`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/single-gpu-guide.md) before changing the recipe.
[`single-GPU guide`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/single-gpu-guide.md) before changing the recipe.

```bash
git clone https://github.com/DaoyuanLi2816/mini-verl.git
Expand Down Expand Up @@ -293,7 +293,7 @@ Layer boundaries are strict, and the first layer never imports torch:
6. `evaluation/`, `reporting/` — measurement.
7. `cli.py` — a thin shell that calls one library function per command.

See [`docs/design.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/design.md).
See [`docs/design.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/design.md).

## Exact versus top-k + tail

Expand Down Expand Up @@ -323,7 +323,7 @@ the teacher still runs a full forward pass to produce the hidden states. Reports
therefore say `teacher_queried_position_ratio`, never "teacher compute saved".

Top-k + tail targets are not a new idea — TRL's `ServerDistillationTrainer` has
`loss_top_k` with an optional tail bucket. See [`docs/math.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/math.md).
`loss_top_k` with an optional tail bucket. See [`docs/math.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/math.md).

## Tool-token masking

Expand All @@ -348,18 +348,18 @@ documented — see `tests/unit/test_token_provenance.py`.
## Benchmark results

Every number below was produced by the commands in
[`docs/benchmarking.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/benchmarking.md) on the hardware recorded in each
[`docs/benchmarking.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/benchmarking.md) on the hardware recorded in each
result file. Nothing is estimated or extrapolated.

* **RTX 4080, real models** — [`docs/rtx4080-baselines.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/rtx4080-baselines.md)
* **RTX 4080, real models** — [`docs/rtx4080-baselines.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/rtx4080-baselines.md)
has measured peak VRAM, decode throughput, the full-recipe run, the two-seed
schema-v2 protocol-teacher comparison, and the preserved legacy comparison.
* **CPU, toy models** — `recipes/toy_cpu.yaml` moves task success from 0.0% to
91.7% in 192 s, and `benchmarks/results/` holds the legacy equal-update parity
run.
The parity run's accuracy differences are **within noise**; it exists to show
that all seven arms run to completion under identical budgets, not to rank
them. See [`benchmarks/README.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/benchmarks/README.md) for why the toy
them. See [`benchmarks/README.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/benchmarks/README.md) for why the toy
backend cannot rank methods.

## Installation
Expand Down Expand Up @@ -464,11 +464,11 @@ distribution before handing it over, and asserts that the result still trains.

For a standard frozen PEFT teacher adapter, including the Qwen3 protocol-SFT
recipe, export command, compatibility checks and policy-competence gate, see
[`docs/teacher-adapters.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/teacher-adapters.md).
[`docs/teacher-adapters.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/teacher-adapters.md).

## Limitations

The short version; the full list is in [`docs/limitations.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/limitations.md).
The short version; the full list is in [`docs/limitations.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/limitations.md).

* Same tokenizer only. Cross-tokenizer distillation is rejected with an error.
* One trajectory per forward pass — `gradient_accumulation_steps` *is* the batch
Expand Down Expand Up @@ -512,8 +512,8 @@ still retain the local state required for exact resume. Redaction is a
best-effort sharing defense, not permission to place real credentials in any
config, run artifact or report.

See [`docs/reproducibility.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/reproducibility.md) and the concise
[`compatibility policy`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/compatibility.md).
See [`docs/reproducibility.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/reproducibility.md) and the concise
[`compatibility policy`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/compatibility.md).

## Roadmap

Expand All @@ -534,7 +534,7 @@ multi-turn tool use — at cluster scale, with Ray. If you have a cluster, use i
miniVERL exists for the case where you have one personal GPU and want to read
every line of what is happening. That can be an older 12 GiB card or a current
high-end card; the repository claims measured performance only for hardware it
actually ran. See [`docs/comparisons.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/docs/comparisons.md).
actually ran. See [`docs/comparisons.md`](https://github.com/DaoyuanLi2816/mini-verl/blob/main/docs/comparisons.md).

## Citation

Expand All @@ -548,13 +548,13 @@ actually ran. See [`docs/comparisons.md`](https://github.com/DaoyuanLi2816/mini-
}
```

See [CITATION.cff](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/CITATION.cff) and [CHANGELOG.md](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/CHANGELOG.md).
Contributions: [CONTRIBUTING.md](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/CONTRIBUTING.md). Security:
[SECURITY.md](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/SECURITY.md).
See [CITATION.cff](https://github.com/DaoyuanLi2816/mini-verl/blob/main/CITATION.cff) and [CHANGELOG.md](https://github.com/DaoyuanLi2816/mini-verl/blob/main/CHANGELOG.md).
Contributions: [CONTRIBUTING.md](https://github.com/DaoyuanLi2816/mini-verl/blob/main/CONTRIBUTING.md). Security:
[SECURITY.md](https://github.com/DaoyuanLi2816/mini-verl/blob/main/SECURITY.md).

## License

Apache-2.0. See [LICENSE](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/LICENSE) and
[THIRD_PARTY_NOTICES.md](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/THIRD_PARTY_NOTICES.md).
Apache-2.0. See [LICENSE](https://github.com/DaoyuanLi2816/mini-verl/blob/main/LICENSE) and
[THIRD_PARTY_NOTICES.md](https://github.com/DaoyuanLi2816/mini-verl/blob/main/THIRD_PARTY_NOTICES.md).

Chinese translation: [README.zh-CN.md](https://github.com/DaoyuanLi2816/mini-verl/blob/v0.3.0/README.zh-CN.md).
Chinese translation: [README.zh-CN.md](https://github.com/DaoyuanLi2816/mini-verl/blob/main/README.zh-CN.md).
2 changes: 1 addition & 1 deletion docs/generated/quality.json
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
{
"schema_version": 1,
"release": "0.3.0",
"status": "validated",
"status": "released",
"measured_commit": "9175bbe88e19e96834766a34dc7c9513c46513c7",
"measured_at": "2026-08-02T15:55:58-07:00",
"cpu_non_gpu_non_network": {
Expand Down
15 changes: 12 additions & 3 deletions docs/release-checklist.md
Original file line number Diff line number Diff line change
Expand Up @@ -48,9 +48,18 @@ after the exact release commit and its remote checks are green.

## After the tag

- [ ] OIDC publication, public hashes/attestations, clean install and one
GitHub Release are verified from the same distributions.
- [ ] Post-release state-sync PR advances clean `main` to `0.4.0.dev0`.
- [x] OIDC publication, public hashes/attestations, clean install and one
GitHub Release were verified from the same distributions by tag run
[`30772772078`](https://github.com/DaoyuanLi2816/mini-verl/actions/runs/30772772078).
- [x] This post-release state-sync change advances development to
`0.4.0.dev0`; it merges only after remote checks are green.

Published artifact identity:

- `miniverl-0.3.0-py3-none-any.whl`: SHA-256
`e42404ced88b75ba4ff31541cb3df697da0eee09a68f603e4eadbf69a11d2032`
- `miniverl-0.3.0.tar.gz`: SHA-256
`006ce418243286dd27e0731090bf9fa1711a1abc962ecfcc4d39807257539cb2`

## Historical v0.2.6 release record

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__ = "0.3.0"
__version__ = "0.4.0.dev0"

__all__ = ["__version__"]
0