decideit 7.0.1__py3-none-any.whl

decideit-7.0.1.dist-info/METADATA

@@ -0,0 +1,275 @@
+Metadata-Version: 2.4
+Name: decideit
+Version: 7.0.1
+Summary: RSI-Jev typed-decision models: a Jev-compatible server, a Python API and the training code
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/Shanghua-Gao/RSI-Jev
+Project-URL: Weights, https://huggingface.co/shgao
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: torch>=2.13; sys_platform != "darwin" or platform_machine != "arm64"
+Requires-Dist: transformers<5.18,>=5.17; sys_platform != "darwin" or platform_machine != "arm64"
+Requires-Dist: mlx>=0.32; sys_platform == "darwin" and platform_machine == "arm64"
+Requires-Dist: tokenizers>=0.22; sys_platform == "darwin" and platform_machine == "arm64"
+Requires-Dist: numpy; sys_platform == "darwin" and platform_machine == "arm64"
+Requires-Dist: pillow; sys_platform == "darwin" and platform_machine == "arm64"
+Requires-Dist: safetensors
+Requires-Dist: huggingface_hub
+Requires-Dist: fastapi
+Requires-Dist: uvicorn
+Provides-Extra: fast
+Requires-Dist: flash-linear-attention<0.6,>=0.5.2; extra == "fast"
+Requires-Dist: fla-core<0.6,>=0.5.2; extra == "fast"
+Provides-Extra: vision
+Requires-Dist: pillow; extra == "vision"
+Requires-Dist: torchvision; extra == "vision"
+Provides-Extra: http
+Requires-Dist: orjson>=3.10; extra == "http"
+Requires-Dist: uvloop>=0.21; sys_platform != "win32" and extra == "http"
+Requires-Dist: httptools>=0.6; extra == "http"
+Provides-Extra: test
+Requires-Dist: pytest; extra == "test"
+Requires-Dist: httpx; extra == "test"
+Provides-Extra: torch
+Requires-Dist: torch>=2.13; extra == "torch"
+Requires-Dist: transformers<5.18,>=5.17; extra == "torch"
+Provides-Extra: mlx
+Requires-Dist: mlx>=0.32; extra == "mlx"
+Requires-Dist: numpy; extra == "mlx"
+Requires-Dist: pillow; extra == "mlx"
+Dynamic: license-file
+
+<h1 align="center">RSI-Jev</h1>
+
+<p align="center">
+<b>A recursively self-improving research loop that builds
+<a href="https://docs.typesafe.ai/api">Jev</a>-style <i>System One</i> decision models.</b>
+</p>
+
+<p align="center">
+  <b><a href="https://shanghua-gao.github.io/RSI-Jev/">🌐 shanghua-gao.github.io/RSI-Jev</a></b>
+</p>
+
+<p align="center">
+  <a href="https://huggingface.co/shgao">🤗 Models</a> ·
+  <a href="docs/inference.md">Docs</a> ·
+  <a href="EXPLORE.md">Every experiment</a> ·
+  <a href="versions/v7.0-vl-3b.md">Release notes</a>
+</p>
+
+<p align="center">
+  <a href="LICENSE"><img src="https://img.shields.io/badge/code-MIT-black" alt="MIT"></a>
+  <a href="https://huggingface.co/shgao/rsi-jev-v7.0-vl-4b"><img src="https://img.shields.io/badge/weights-Apache--2.0-black" alt="Apache-2.0 weights"></a>
+  <a href="serve/README.md"><img src="https://img.shields.io/badge/API-Jev%20compatible-black" alt="Jev-compatible API"></a>
+</p>
+
+AI agents propose the hypotheses, register their predictions before spending GPU time, run the
+experiments, and retire their own champions when the evidence says to. Every release, and every
+experiment that failed on the way, is published with its numbers. The loop is the next version of
+[AutoScientists](https://github.com/mims-harvard/AutoScientists).
+
+- **9 releases in 41 days**, from v1.0 to v7.0-VL, each trained, evaluated and documented by the loop.
+- **801 experiments**, every one written up, failures included: [EXPLORE.md](EXPLORE.md).
+- **v6.1-VL 27B has the highest score on the Decision Index 0.3 public set: 65.64**, with `--effort auto`. The official score is pending.
+- **v7.0-VL 4B scores 60.63 on the public Decision Index 0.3**, up from 50.98 for v6.1-VL 4B.
+- **v7.0-VL 3B**, the 4B cut at layer 20, scores 60.01 at two thirds of its latency.
+
+<p align="center">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" srcset="assets/board-v6.1-27b-dark.svg">
+    <img src="assets/board-v6.1-27b-light.svg" width="760" alt="Decision Index 0.3 public set: RSI-Jev v6.1-VL 27B 65.64, Torchcast Decision 27B 65.10, Perplexity Decider v1.1 (27B) 62.25, Cloudflare clef 61.71, Jev 57.96, RSI-Jev v6.1-VL 4B 50.98">
+  </picture>
+</p>
+
+## What it builds
+
+Models that answer a yes/no, pick-one-of-*k* or rate-on-a-rubric question about a document, a chat
+or an image. One forward pass returns a calibrated probability for every option. Nothing is
+generated, so there are no reasoning tokens to spend.
+
+What v7.0-VL can spend is depth. It has answer heads at layers 16, 20 and 32, and `effort` picks
+how many layers a request uses:
+
+| effort | layers | median / mean latency | use it for |
+|---|---|---|---|
+| `low` | 16 | 24 / 65 ms | intent, routing, retrieval |
+| `medium` | 20 | 28 / 78 ms | classification and judgement |
+| `high` | 32 | 42 / 117 ms | knowledge, multi-step reasoning, code |
+| `auto` | 16, 20 or 32 | 41 / 85 ms | mixed traffic: stops at the first layer that is confident enough |
+
+v7.0-VL 3B is the same model cut at layer 20: `low` stops at 16, `medium` and `high` at 20.
+v6.1-VL 27B has the same levels at layers 48, 56 and 64.
+
+Fast when it's easy, deep when it's hard. Which tasks need which depth:
+[docs/effort-depth.md](docs/effort-depth.md).
+
+<p align="center">
+  <picture><source media="(prefers-color-scheme: dark)" srcset="site/assets/v4/capsules-dark.webp"><img src="site/assets/v4/capsules-light.webp" width="49%" alt="A tray of gel capsules, one leaking: defective, 68% sure"></picture>
+  <picture><source media="(prefers-color-scheme: dark)" srcset="site/assets/v4/keep-file-dark.webp"><img src="site/assets/v4/keep-file-light.webp" width="49%" alt="A delete-file dialog; the user said keep the file: Cancel, 99% sure"></picture>
+  <br><sub>Capsules photo from VisA (Zou et al. 2022, CC BY 4.0, resized).</sub>
+</p>
+
+## Quickstart
+
+```bash
+pip install "rsi-jev[fast,vision] @ git+https://github.com/Shanghua-Gao/RSI-Jev"
+rsi-jev serve v7.0-vl-4b --effort auto --port 8000
+```
+
+`rsi-jev serve v7.0-vl-3b` serves the 3B (6.8 GB). On a 96 GB GPU, `rsi-jev serve v6.1-vl-27b --effort auto` serves the 27B.
+
+On an Apple Silicon Mac (the 3B runs in about 2 GB) or on Linux, nothing else needs to be installed first:
+`curl -fsSL https://github.com/Shanghua-Gao/RSI-Jev/releases/latest/download/install.sh | sh`, then
+`git diff | decide "Does this change touch authentication?"`. Building on it: [decideit](docs/decideit.md);
+on a Mac, see also [the Mac guide](docs/mac.md).
+
+```bash
+curl localhost:8000/v1/systemone -H 'Content-Type: application/json' -d '{
+  "model": "jev-latest",
+  "state": "I was charged twice. Please refund.",
+  "questions": {"refund": {"type": "noul", "instructions": "Does the user request a refund?"}}
+}'
+```
+
+Or in Python, with no server:
+
+```python
+from rsijev import Decider
+
+d = Decider("v7.0-vl-4b")
+d.decide("Customer photo: <image>\nThe customer says it arrived damaged.",
+         {"damaged": {"type": "noul", "instructions": "Does the photo show visible damage?"}},
+         images=["photo.jpg"])
+```
+
+The server speaks Jev's API, so an existing Jev client works once its base URL points here. Every
+response reports which layer answered and how confident it was. Options and limits:
+[`serve/README.md`](serve/README.md); setups and what each costs: [`docs/inference.md`](docs/inference.md).
+
+## Releases
+
+| Model | Input | Decision Index 0.3 | Decision Index 0.2.1 | Held-out set | |
+|---|---|---|---|---|---|
+| **v7.0-VL-4B** | text, images | 60.63 | 59.63 | **0.811** | [🤗](https://huggingface.co/shgao/rsi-jev-v7.0-vl-4b) |
+| v7.0-VL-3B | text, images | 60.01 | 58.84 | 0.797 | [🤗](https://huggingface.co/shgao/rsi-jev-v7.0-vl-3b) |
+| v6.1-VL-27B, `auto` | text, images | **65.64** | **64.92** | 0.808 | [🤗](https://huggingface.co/shgao/rsi-jev-v6.1-vl-27b) |
+| v6.1-VL-4B | text, images | 50.98 | 50.74 | 0.729 | [🤗](https://huggingface.co/shgao/rsi-jev-v6.1-vl-4b) |
+| v6.0-VL-4B | text, images | 46.23 | 46.24 | 0.698 | [🤗](https://huggingface.co/shgao/rsi-jev-v6.0-vl-4b) |
+| v5.0-VL-3B | text, images | – | 38.38 | 0.689 | [🤗](https://huggingface.co/shgao/rsi-jev-v5.0-vl-3b) |
+| v4.0-VL-2B | text, images | – | – | 0.653 | [🤗](https://huggingface.co/shgao/rsi-jev-v4.0-vl-qwen3.5-2b) |
+| v3.0-2B | text | – | – | 0.649 | [🤗](https://huggingface.co/shgao/rsi-jev-v3.0-qwen3.5-2b) |
+| v2.1-2B | text | – | – | 0.633 | [🤗](https://huggingface.co/shgao/rsi-jev-v2.1-qwen3.5-2b) |
+| v2.0-2B | text | – | – | – | [🤗](https://huggingface.co/shgao/rsi-jev-v2.0-qwen3.5-2b) |
+| v1.0-2B | text | – | – | 0.604 | [🤗](https://huggingface.co/shgao/rsi-jev-v1.0-qwen3.5-2b) |
+| v1.0-0.8B | text | – | – | – | [🤗](https://huggingface.co/shgao/rsi-jev-v1.0-qwen3.5-0.8b) |
+
+Decision Index is the public benchmark, full runs of the default configuration (the 27B with
+`--effort auto`, which we recommend): the 0.3 public
+index (public set; v6.1-VL 4B scored with the pre-release 0.3 kit) and 0.2.1. The held-out set
+is our zero-shot check: no release trained on it. The 4B and smaller are fine-tuned from Qwen3.5
+Base; v6.0-VL and v6.1-VL 4B run the whole Qwen3.5-4B-Base and can answer at layer 16, 20 or 32
+of it. v7.0-VL 4B starts from the instruction-tuned Qwen3.5-4B, repeats the options and reads
+them with attention that runs both ways over the prompt; it answers at layer 16, 20 or 32 the same
+way. v7.0-VL 3B is v7.0-VL 4B cut at layer 20, with no retraining. v6.1-VL 27B is fine-tuned from Qwen3.8-27B and answers at layer 48, 56 or 64. Every
+number, caveat and per-benchmark score is in the [release records](versions/).
+
+- **2026-11-06 · v7.0-VL 3B** is v7.0-VL 4B cut at layer 20, no retraining: the 4B's answers at `medium`, 3.30B parameters. Decision Index 0.3 60.01.
+- **2026-11-05 · v7.0-VL 4B** reads the options in a new way (option echo, prefix-bidirectional attention) and is trained once on 2.76M examples. Decision Index 0.3 50.98 → 60.63.
+- **2026-10-08 · v6.1-VL 27B** brings the same method to Qwen3.8-27B. Decision Index 0.3 public set 65.64 with `auto`.
+- **2026-10-07 · v6.1-VL 4B** averages v6.0-VL with a second fine-tune of the same base. Decision Index 0.3 46.23 → 50.98.
+- **2026-10-06 · v6.0-VL 4B** chooses its depth per question. Decision Index 38.38 → 46.24.
+- **2026-10-02 · v5.0-VL 3B** cuts the model to the first 20 of 32 layers and says "unknown" when a question has no answer.
+- **2026-10-01 · v4.0-VL** reads images, with text held at v3.0's level.
+- **2026-09-28 · v3.0** is the first release where reinforcement learning helps.
+
+## How it got here
+
+<p align="center">
+  <img src="assets/loop-social.gif" width="680" alt="The champion line climbs from v1.0 (0.622) to v2.0 (0.709), v2.1 (0.736) and v3.0 (0.756) on the 15-benchmark suite; grey dots are the experiments that did not clear it">
+  <br><sub>The loop's first three releases: each grey dot is an experiment that did not beat the champion.</sub>
+</p>
+
+The experiments that made a release are in [docs/history.md](docs/history.md); all the others,
+with why each failed, are in [EXPLORE.md](EXPLORE.md); the reinforcement-learning arms are in
+[docs/rl.md](docs/rl.md).
+
+<details>
+<summary><b>Each experiment that moved a release, from v1.0 to v7.0-VL</b></summary>
+
+| where it went | result | what it established |
+|---|---|---|
+| **v1.0** | typed-decisions **0.662** | the bar to clear, and it had never seen a benchmark's train split |
+| train on the benchmark's own train split | typed-decisions 0.7794 | worth +0.12 — and it costs general knowledge, so it was **rejected** |
+| the same + 15% general-knowledge replay | typed-decisions 0.796 | the replay pays that cost back. The first keeper |
+| the other benchmarks' train splits too, 3k per source | suite 0.7049 | the lever is not specific to one benchmark |
+| a per-question confidence head, forbidden to sharpen score questions | suite ECE 0.0736 | confidence that means something, and it never changes which option wins. Shipped |
+| **v2.0** | suite **0.7056**, ECE **0.0546** | capability was v1.0's lever; **what it trains on, and calibrating it afterwards, is a second one** — and free at inference. But the gain was on the benchmarks it had trained for |
+| the replay weighted toward science, half of it recast to ten options | MMLU-Pro 0.331 | more and better-shaped data barely moves general knowledge: +0.008 |
+| **freeze** the tower's lower third instead | MMLU-Pro 0.339 | holding those layers still does recover some of it — and breaks three of the decision benchmarks, because they do need to adapt |
+| the bottom 8 layers at **one tenth** the learning rate | MMLU-Pro 0.359 | it was never a data problem. Fitting decisions through the whole tower at one rate was overwriting what the knowledge sat in — frozen layers cannot adapt, slow ones can |
+| **v2.1** | suite **0.7291**, MMLU-Pro **0.383** | and it travels: ahead of v1.0 on all three benchmarks it never trained on, where v2.0 was not |
+| wider coverage (data-scale-2): +84k questions, emotion/SST-5/ANLI and more tasksource and Open-Jev train items | suite +0.011 / +0.013 on two seeds | coverage is still the lever that moves the suite |
+| a two-layer MLP combining the option and cross-attention features | suite 0.7589 on the new 15-benchmark suite | the readout had a little left in it: +0.002 |
+| a listwise reranking reward (NDCG@5 over 16 candidates) on top, against a KL anchor | hippo R@1 0.192 → 0.308, suite −0.003 | **the first RL stage kept**: a reward over the *order* of candidates carries what per-item labels cannot |
+| **v3.0** | suite **0.756**, ECE **0.066** | **the first release where RL works**: +60% reranking R@1 over its supervised parent, the rest held level |
+| image training on the train splits of twelve public datasets, with text replay | held-out image top-1 0.809 against 0.796 for a text-only control | the model reads images through the base model's frozen vision tower |
+| generated reasoning questions with soft targets where the model is at chance | held-out ECE after calibration, BBH excluded, 0.096 → 0.075 | the image rounds had made it overconfident; this repairs most of it |
+| an RL stage charging a confident mistake 4x a timid correct answer | held-out ECE before calibration 0.200 → 0.082 | against a matched supervised stage: better accuracy and calibration, no detectable ranking difference (one seed) |
+| **v4.0-VL** | image top-1 **0.803** held out, suite **0.756**, ECE **0.043** | **reads images**, with text held at v3.0's level |
+| the same recipe on Qwen3.5-4B, read at layer 16, 20 or 24 of its 32 | suite 0.744 · 0.760 · 0.760 | a System One model doesn't need the deep layers: quality flattens by layer 20 |
+| images on the layer-20 model, plus questions whose right answer is "unknown" | KoBBQ unknown-when-ambiguous 0.679 → 0.891 | images cost the model its "unknown" answer until the data asked for it |
+| pool each option over its own tokens, not the separator that follows the previous one | CLINC150 0.383 → 0.753 | a readout bug found by an outside benchmark; the same weights read correctly once each option is pooled over its own tokens |
+| retrain only the decision head for that readout, tower frozen, 600 steps | final ECE after calibration 0.080 → 0.042, Decision Index 38.38 | the head had learned the old pooling; a short head-only stage recovers the calibration |
+| **v5.0-VL** | MMLU-Pro **0.429**, Decision Index **38.38**, KoBBQ unknown **0.932** | **3B: the first 20 of 32 layers** |
+| early-exit heads that read a detached copy of their layer, so no gradient from them reaches the trunk | MMLU-Pro 0.443 at layer 32 | retuning heads on a trunk trained with attached exits had not recovered depth: the trunk had lost it |
+| 10,000 steps on a cleaned corpus instead of 40,000 on the old one | held-out +0.026, MMLU-Pro +0.043 | longer training overfits the suite; shorter and cleaner wins held out |
+| one temperature per exit and an exit policy fitted on held-out proxies of the test mix | every release bar passes at 20.9 layers on average | policies tuned on in-distribution rows stop too early on hard questions |
+| **v6.0-VL** | Decision Index **46.24**, MMLU-Pro **0.440** | **4B: answers at layer 16, 20 or 32, whichever is confident first** |
+| v6.0-VL averaged, weight 0.5 each, with a second fine-tune of the same base trained on other data | suite 0.793 and held-out 0.728 at layer 32, against 0.770 and 0.695 | the average beats both members; the gain is in the tower, the heads add nothing |
+| one temperature per exit and both thresholds chosen again for the average | `auto` at 0.85 / 0.50 matches layer 32 at 19.5 layers | averaging shrinks the logits: at v6.0-VL's temperatures it was underconfident; calibration ends worse than v6.0-VL's (final ECE 0.048 vs 0.036) |
+| **v6.1-VL** | Decision Index 0.3 **50.98**, held-out **0.729** | **4B: v6.0-VL averaged with a second fine-tune** |
+| two LoRA fine-tunes of Qwen3.8-27B, averaged weight 0.5 each, chosen from 9 full Decision Index 0.3 reads | Decision Index 0.3 65.59, against 64.83 and 65.30 for the two members | averaging carries over to 27B; uneven weights and a third ingredient read 65.37 to 65.58 |
+| short inputs right-padded to a multiple of 64 tokens and replayed as CUDA graphs | short one-question requests in about half the time | at one request a time the 27B is bound by kernel launches, not compute |
+| **v6.1-VL 27B** | Decision Index 0.3 public set **65.64** with `auto`, held-out **0.808** | **27B: the same method on Qwen3.8-27B, answers at layer 48, 56 or 64** |
+| repeat the option list once and read the second copy; let the prompt and the options attend both ways | our 4B pick set +3.5 and +2.6, +3.8 over v6.1-VL together with an option-order consistency loss | what the readout sees mattered more than ~150 later loss and schedule variants |
+| start from the instruction-tuned model, keep the base model's reasoning traces as a training signal | held level at 4B, +1.78 at 2B | reasoning stays a training-time teacher; nothing is generated at inference |
+| data rounds 7 to 12: each round's weakest families get training data built from public train splits | Decision Index 0.3 53.06 → 60.62 | targeted data moves its targets and dilutes the rest; rounds are judged per benchmark |
+| **v7.0-VL** | Decision Index 0.3 **60.63**, held-out **0.811** | **4B: option echo and prefix-bidirectional attention, instruction-tuned start, one end-to-end run** |
+| **v7.0-VL 3B** | Decision Index 0.3 **60.01**, held-out **0.797**, median latency 19.6 ms against 29.4 ms | **3B: v7.0-VL 4B cut at layer 20, no retraining** |
+
+</details>
+
+## Documentation
+
+| | |
+|---|---|
+| [Inference guide](docs/inference.md) | setups, measured latency, speed-ups |
+| [Mac (MLX)](docs/mac.md) | the 8-bit and 4-bit Apple-silicon builds and what each costs |
+| [Effort and depth](docs/effort-depth.md) | which tasks and questions need the deep layers |
+| [HTTP API](serve/README.md) | the Jev-compatible server |
+| [Benchmarks](BENCHMARKS.md) | what each number means and how to reproduce it |
+| [Release records](versions/) | how each release was built, what it scores, its limitations |
+| [Reinforcement learning](docs/rl.md) | where RL beat supervised training, and where it didn't |
+| [Speed](docs/speed.md) | how inference was made faster |
+| [Code](rsijev/README.md) | training and the files the loop may rewrite |
+
+## Contributing
+
+A case where the model is confidently wrong, or a variant you tried that failed, is the most useful
+thing you can send: [report a wrong answer](https://github.com/Shanghua-Gao/RSI-Jev/issues/new?template=wrong-answer.yml)
+· [report an experiment](https://github.com/Shanghua-Gao/RSI-Jev/issues/new?template=experiment.yml).
+Each becomes a registered prediction tested in the next version. See
+[`CONTRIBUTING.md`](CONTRIBUTING.md). To collaborate or support the work with compute,
+contact [Shanghua Gao](https://shgao.site).
+
+## Acknowledgements
+
+Thanks to **HP** and **NVIDIA** for providing the **HP ZGX Nano AI Station**, powered by the
+**NVIDIA GB10 Grace Blackwell Superchip**.
+
+## License
+
+Code: MIT. Weights: Apache-2.0, following the base model; some image training sources are
+non-commercial, listed on each model card. Not affiliated with TypeSafe AI. To cite, use
+[`CITATION.cff`](CITATION.cff).

decideit-7.0.1.dist-info/RECORD

@@ -0,0 +1,60 @@
+agentguard/agentguard.py,sha256=kokw2sZq24G2_UypVDHklRRUCH5Wbm8z9fMX-7KIOhI,21697
+decideit-7.0.1.dist-info/licenses/LICENSE,sha256=EQOzpvYbfhf58xZP9kwabyQI6_9Z-kPIKyZ9PKdFrn4,1069
+rsijev/__init__.py,sha256=C4s6KTw_WJzY_1FnTSLEy9lwjWoB5vKAYHvHLlGBYrU,1003
+rsijev/adaptive.py,sha256=eYS3u6xuJHTS5ZTwr2tksDVdCKDpgHVUoUDlrVe52BI,33710
+rsijev/arch.py,sha256=nOgJSpcGk8TR9vEJQCw2oIHz2EXPB3JZOnLYwzJfZM8,61917
+rsijev/calibrate.py,sha256=4iqkDMNGmTNxUY79dcsvkWxl3f6N6y7mm4IB4YYRLcE,18069
+rsijev/contract.py,sha256=1BL53ZjyT9w_Cqv6O8aySBzojJn36EmBQKTB0jUuodc,5162
+rsijev/data.py,sha256=vp6Hx2uPHB4-jxuR8uY3az7tE0l8IqNFfX06hj8hQqE,11797
+rsijev/encode.py,sha256=sT1PIuveLZs-gsGMxFNtQyHez7FTQvub9QjGOD7hUtM,24281
+rsijev/evaluate.py,sha256=Tw3JgsmPxqNSVUc5tIvfjr5ACJzr9PGRYzPU1t0dwMs,27810
+rsijev/exit_policy.py,sha256=YqA-oRipUdrvpdgOkk4byUwArh7q7d5qnROSrZ-pdTg,38934
+rsijev/fit.py,sha256=MjS72ssUvmTcnk17AMuRXYKZBQEyWkGZBeby-WH71QE,57398
+rsijev/frozen_tok.py,sha256=R2m696zSvAmwn27HjJuQ8fDViYBLpIWmxVuIt5_kZjs,2736
+rsijev/image_text.py,sha256=JB7t6WsNg7i7jVtbWEPMw2zu_1I81mBf2SzBcZapmVE,1355
+rsijev/metrics.py,sha256=56C5cnoryWg8Ys0N4ozyMvcJhLCoqQLUumVioT7Z9Sk,6579
+rsijev/rl2.py,sha256=WWbjCQ2xfy77pbxgpvy4qsq9DhRY4_2PynK973jKS9U,52253
+rsijev/targets.py,sha256=ZkA1zfZvC-o21Q_oAqr0Af7oFGvT80JRzayr0ISjNi0,5115
+rsijev/targets_suite.py,sha256=MGFym3l-Fszh7S_CbxM9Vx9YOcd-08M2SrJozwN34A4,42289
+rsijev/train.py,sha256=A_a6BMDv2BYkFc1ip6tPEOpSXKMSUHtxhJJ42FEd22c,9975
+rsijev/vision.py,sha256=x_vj6U51GnNGFcQPXQzaIoMykTr6WFiOtH4M6bKNtiA,15333
+rsijev/vision_fit.py,sha256=V7-lcmHp1XiqMCw7qOF-TBNq9hbG11qH9OiV5iGSrVQ,6609
+rsijev/mlx/__init__.py,sha256=Z-5EvZ3Qsm6O8jHAALYFG8PYSo2Ii5kXoyMq4x-wgOU,1160
+rsijev/mlx/collate.py,sha256=mgxsKQoano34qB7I_wL66fm5aY7tSeU2VGJAWtWnHNE,3183
+rsijev/mlx/convert.py,sha256=pu8L84a3jEp97ZsMRpJBQkHcIpM0OP_yUWZs6kMMimc,17445
+rsijev/mlx/heads.py,sha256=j0kUnaCda2anmvkqOdGfbb7BjeuMmx2FYmyu3WMvDcI,10209
+rsijev/mlx/model.py,sha256=IWrgDctxTQNK058wiyKqPMPcuiwoWaxRc744E2fDN5c,16320
+rsijev/mlx/quant.py,sha256=IQCURFUWD-DV6jV7qTIiLTQU_Cb0wtfeEErI2FHhz7M,7297
+rsijev/mlx/serving.py,sha256=koNwg--rmpQS6-leoTOm7JGiX6KNcSGpz6QakoMfCKI,23987
+rsijev/mlx/text.py,sha256=AhyeWxZshvdaubAVXl4nTXAMre7wC8VYItVQWJIpajw,30396
+rsijev/mlx/vision.py,sha256=FyoFNq57wDZDzvF9hmgOf5jYBQ-ZhH_jXVCdd0vz7SA,16057
+serve/__init__.py,sha256=MkGgPISXMwTV_k4Ffv3nIpA6mFuUvN4f54ioIHUpkCo,67
+serve/accel.py,sha256=dbGn4I8gtAEvwWj3VfWF9P0P6l9LxiSZW7G3stiMbBs,2993
+serve/app.py,sha256=xpkR1503dr-UljKlgaRJ_LI8pP8s4kATBRl3_SYNRT8,15965
+serve/autostart.py,sha256=H8E-iu92o0haArczmo1puJFvz5OKYz35bOIF639LLqs,7855
+serve/batcher.py,sha256=iRLa-Oiw7chq6-Ra6ESojZ9E_tk-bQETv-wFoQmWRwI,14751
+serve/bench.py,sha256=xEahsqczy2dg78530MyOw1Ktrlv6d4KP9N4bRUDkebw,8283
+serve/cli.py,sha256=PCvTAqCOeZ9K16Mv8hbKwub3HPRxFO1LkSc8WAhm_p0,2621
+serve/daemon.py,sha256=tUxXi_v5PNwUUQe-NOMvqafxcoLZM_kliV025ebX9V8,4111
+serve/decide_cli.py,sha256=RPyApGd3XpkSZWCzswWNQQRNqPPYk95ofZQU-9ZD6XM,8867
+serve/decider.py,sha256=yQA3JHEcdXMoMt0jrKFTRsxkhvnu0uCBaxSo0amRIfw,5441
+serve/demo.py,sha256=Q6C8ksLlG_VheBnrWAbqnK8rVN6MropFJzdaV2GQEc8,7207
+serve/demo_images.py,sha256=ELMcUI0aNN4OiXWY19lKuZUYHxKzbSdUSfpNqIk8zZw,12209
+serve/effort.py,sha256=fqKucC2-Q4VL3eD0HLFk9uXaKkZEOjemDZgdO0mqzvw,7574
+serve/examples.json,sha256=4bMLn1qhYIh5EMv3_RrwoOjDOqaIgqOZSGxEWnlQ3Nk,113904
+serve/images.py,sha256=BUREyY8oV16WFAxS7egHqwruvOKgPuKOMJ8jQ4IgGWk,9458
+serve/infer.py,sha256=ZwGSiiqTPx3V5FM2yAFPUV_Bt9_IJ8VrnsC3BxlDYjY,50740
+serve/kernels.py,sha256=ghBaH4Nc6ltMIaIdiHTvwYDWF6ZAJj9KYh6Zjvn1yEc,4017
+serve/padded.py,sha256=wcvq8J2UW6E4N0xnd_THAo777-1R4xwDAqcR8lt2fB0,21588
+serve/plan.py,sha256=q7kPdGz6FtKR7a7yo7-RxNOqQ43ySuaJDfWYeuzV2JU,12332
+serve/release.py,sha256=l8-NlIx98wCjvxFoISWotSCXcb2-3Xf_Xrq96_DQ_4w,29648
+serve/runtime.py,sha256=ycUDpeEZI4Yq7chEK1jYTXJgFdBbhri-EwzmKlbDOkE,11028
+serve/serve_tune.py,sha256=59aPXfAfYg1EVOhzCK1HUaiUeAnZWLV-GviryQoyaQc,5592
+serve/server.py,sha256=7P3vdY9C6HOaRYQAFIEvt9eJXc-Yr8WLKlr31u5wtiI,25058
+serve/ui.html,sha256=CEe8fsqE2cn8VE10kAyhrKizBO4AoQZKGmcG8u14hcI,60973
+serve/wire.py,sha256=GLCO-DX9TZTirXxVd9NZbd_QSEtVw_kQyStBkf-_jXc,10424
+decideit-7.0.1.dist-info/METADATA,sha256=KeW3a-Hzssep5nVqVO3w8nWwMLkaNi6Ny055biAVxN4,20180
+decideit-7.0.1.dist-info/WHEEL,sha256=mOdXR9C3wMVov7cSY8PY-XCzo3SmmWCHvGQr5csJoao,91
+decideit-7.0.1.dist-info/entry_points.txt,sha256=DOPDWjgoZB-lReKw4hHGQUo4oGvCls_ORR_W0YTz7k0,147
+decideit-7.0.1.dist-info/top_level.txt,sha256=aJzPLsHgp6OzliBarNts_hDSjnt7e46niuMOvrn75_E,24
+decideit-7.0.1.dist-info/RECORD,,

decideit-7.0.1.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (84.2.1)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

decideit-7.0.1.dist-info/entry_points.txt

@@ -0,0 +1,5 @@
+[console_scripts]
+agentguard = agentguard.agentguard:main
+decide = serve.decide_cli:main
+decideit = serve.decide_cli:main
+rsi-jev = serve.cli:main

decideit-7.0.1.dist-info/licenses/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Shanghua Gao
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

decideit-7.0.1.dist-info/top_level.txt

@@ -0,0 +1,3 @@
+agentguard
+rsijev
+serve

rsijev/__init__.py

@@ -0,0 +1,21 @@
+"""Typed-decision models: a fine-tuned tower with a trained option scorer.
+
+The file split is a guard rail, not a style. An automated experiment may rewrite
+the EDITABLE modules and may not touch the PROTECTED ones, so a change to the
+model, the data or the training recipe can never quietly become a change to how
+it is scored. `rsijev/README.md` has the table.
+
+    PROTECTED   contract.py  encode.py  evaluate.py  metrics.py  targets.py
+    EDITABLE    arch.py      data.py    train.py     (fit.py: bookkeeping is not)
+"""
+__all__ = ["contract", "metrics", "arch", "data", "train"]
+
+
+def __getattr__(name):
+    # `from rsijev import Decider` is the one-line way in. It lives with the server
+    # (serve/decider.py) because it IS the server's code path; imported lazily so
+    # that training and evaluation never pull in the HTTP stack.
+    if name == "Decider":
+        from serve.decider import Decider
+        return Decider
+    raise AttributeError(f"module 'rsijev' has no attribute {name!r}")