headroom-winnow 0.2.0__tar.gz

headroom_winnow-0.2.0/.gitignore

@@ -0,0 +1,11 @@
+__pycache__/
+*.py[cod]
+*.egg-info/
+build/
+dist/
+.venv/
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+benchmarks/results/
+site/

headroom_winnow-0.2.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

headroom_winnow-0.2.0/NOTICE

@@ -0,0 +1,17 @@
+headroom-winnow
+Copyright 2026 headroom-winnow contributors
+
+This product is a plugin for Headroom (https://github.com/headroomlabs-ai/headroom),
+licensed under the Apache License, Version 2.0. It implements Headroom's
+``headroom.compressor`` contract and follows its coding conventions and tool
+configuration.
+
+This product uses the Squeez approach and models by KRLabs
+(https://github.com/KRLabsOrg/squeez), licensed under the Apache License,
+Version 2.0, including the model KRLabsOrg/verbatim-rag-modern-bert-v2 and the
+dataset KRLabsOrg/tool-output-extraction-swebench.
+
+The pooled model distributed in this project's releases is fine-tuned from
+jhu-clsp/ettin-encoder-32m (https://huggingface.co/jhu-clsp/ettin-encoder-32m),
+licensed under the MIT License, on the dataset above using Squeez's training
+code.

headroom_winnow-0.2.0/PKG-INFO

@@ -0,0 +1,279 @@
+Metadata-Version: 2.5
+Name: headroom-winnow
+Version: 0.2.0
+Summary: Winnow: task-aware pruning of coding-agent tool output for Headroom, powered by Squeez models
+Project-URL: Homepage, https://github.com/RecepKurtulus/headroom-winnow
+Project-URL: Issues, https://github.com/RecepKurtulus/headroom-winnow/issues
+Project-URL: Benchmarks, https://github.com/RecepKurtulus/headroom-winnow/blob/main/benchmarks/RESULTS.md
+Author: Recep Kurtulus
+License-Expression: Apache-2.0
+License-File: LICENSE
+License-File: NOTICE
+Keywords: agents,compression,context,headroom,llm,squeez,winnow
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Requires-Dist: headroom-ai<0.40,>=0.39
+Provides-Extra: bench
+Requires-Dist: datasets>=2.18; extra == 'bench'
+Provides-Extra: dev
+Requires-Dist: mypy>=1.10; extra == 'dev'
+Requires-Dist: numpy<2.3; extra == 'dev'
+Requires-Dist: pytest>=8.0; extra == 'dev'
+Requires-Dist: ruff>=0.6; extra == 'dev'
+Provides-Extra: model
+Requires-Dist: torch>=2.1; extra == 'model'
+Requires-Dist: transformers>=4.48; extra == 'model'
+Description-Content-Type: text/markdown
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/RecepKurtulus/headroom-winnow/main/docs/assets/logo.png" width="112" alt="Winnow logo">
+</p>
+
+<h1 align="center">Winnow</h1>
+
+<p align="center">
+  <b>Separate the grain from the chaff in your coding agent's tool output.</b><br>
+  A task-aware pruner for <a href="https://github.com/headroomlabs-ai/headroom">Headroom</a>, powered by <a href="https://github.com/KRLabsOrg/squeez">Squeez</a> models.
+</p>
+
+<p align="center">
+  <a href="https://github.com/RecepKurtulus/headroom-winnow/actions/workflows/ci.yml"><img src="https://github.com/RecepKurtulus/headroom-winnow/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
+  <img src="https://img.shields.io/badge/python-3.10%2B-blue" alt="Python 3.10+">
+  <img src="https://img.shields.io/badge/license-Apache--2.0-green" alt="Apache-2.0">
+  <a href="https://recepkurtulus.github.io/headroom-winnow/"><img src="https://img.shields.io/badge/docs-site-5c6bc0" alt="Docs"></a>
+</p>
+
+Coding agents spend most of their context on tool output: test runs, builds,
+grep hits, logs. Usually only a handful of those lines matter for the agent's
+next step. Winnow plugs a small model trained on real agent traces
+into Headroom's compression pipeline. It keeps the lines the task needs, never
+drops an error or a traceback, and turns everything else into retrievable
+markers.
+
+<p align="center"><img src="https://raw.githubusercontent.com/RecepKurtulus/headroom-winnow/main/docs/assets/demo.gif" width="760" alt="Winnow pruning a go build output from 1,838 to 479 tokens"></p>
+
+> **41% fewer tokens with 0.85 gold-line recall, zero lost error lines, and
+> 0.29 s median latency on a consumer GPU.** At equal compression it keeps
+> 15-27 points more of the relevant lines than Headroom's built-in
+> `relevance_split`. [Full results →](https://github.com/RecepKurtulus/headroom-winnow/blob/main/benchmarks/RESULTS.md)
+
+---
+
+## Example
+
+The agent asks *"Find the build output block that reports the missing
+`GetObject` method in the `UserHandler` implementation of
+`storage.StorageService`"* and runs `go build ./...`:
+
+<table>
+<tr><th>Before: 119 lines, 1,838 tokens</th><th>After: 32 lines, 479 tokens (−74%)</th></tr>
+<tr><td>
+
+```text
+$ go build ./...
+# .../handlers/user_handler.go:23:12: cannot use h
+  (type *UserHandler) as type storage.StorageService:
+  *UserHandler does not implement storage.StorageService
+  (missing GetObject method)
+# .../handlers/order_handler.go:45:9: cannot use orderSvc ...
+# .../auth/auth.go:12:5: import cycle not allowed
+# .../permissions/perm.go:8:2: import cycle not allowed
+# .../api/v2/client.go:67:15: undefined: storage.ObjectMetadata
+# .../api/v2/client.go:88:20: cannot assign string literal ...
+# .../database/migration.go:33:10: cannot find module ...
+# .../cache/lru_test.go:78:5: race detector: data race
+Read at 0x00c0000a1230 by goroutine 9:
+  github.com/example/project/internal/cache.(*LRUCache).Get()
+      .../internal/cache/lru.go:44 +0x7c
+... 100 more lines of goroutine dumps,
+    unrelated packages and test output
+```
+
+</td><td>
+
+```text
+$ go build ./...
+# .../handlers/user_handler.go:23:12: cannot use h
+  (type *UserHandler) as type storage.StorageService:
+  *UserHandler does not implement storage.StorageService
+  (missing GetObject method)
+# .../handlers/order_handler.go:45:9: cannot use orderSvc ...
+# .../auth/auth.go:12:5: import cycle not allowed
+	imports github.com/example/project/internal/permissions
+<<ccr:e2e73bc3f2f290a77efc7426 52_lines_offloaded>>
+--- FAIL: TestUserHandler_Get (0.00s)
+    user_handler_test.go:45:
+        Error:        Not equal:
+                        expected: &storage.Object{...}
+                        actual  : <nil>
+FAIL
+exit status 1
+<<ccr:0bac82ae58f9e3ba08a67445 15_lines_offloaded>>
+--- FAIL: TestConcurrentAccess (0.01s)
+        panic: runtime error: invalid memory address ...
+FAIL
+exit status 1
+<<ccr:407242e9f3f948833f58dd56 23_lines_offloaded>>
+...
+```
+
+</td></tr>
+</table>
+
+The block the agent asked for is kept verbatim, every failure line survives,
+and each `<<ccr:…>>` marker can be expanded through Headroom's retrieval tool if
+the agent needs the dropped lines after all. (Real model output on an example
+from the Squeez test set; long paths and lines shortened to fit.)
+
+## How it works
+
+```mermaid
+flowchart LR
+    A[Agent tool call] --> B[Headroom proxy]
+    B --> C{Content router}
+    C -- "log / search / diff / text" --> D[WinnowCompressor]
+    C -- "JSON / code / HTML" --> E[Headroom built-ins]
+    D --> F[Span model<br/>which lines matter?]
+    F --> G[Safety rules<br/>errors, tracebacks,<br/>±2 context, edges]
+    G --> H[Render<br/>kept lines verbatim +<br/>ccr markers]
+    H --> I[(CCR store<br/>hash → original)]
+    H --> J[Pruned context to LLM]
+    D -. "no query / too short /<br/>too large / low savings" .-> E
+```
+
+1. **Gate.** Blocks without a task query, under 40 lines, over the device's token
+   budget, or where pruning would save less than 20% pass through untouched, so
+   Headroom's own compressors handle them.
+2. **Score.** A span model reads the task query and the whole output and marks
+   the lines that matter for the next step.
+3. **Protect.** Lines with errors, failures, exit codes, panics or pytest `E`
+   details are always kept, as are whole Python tracebacks, two neighbours of
+   every kept line, and the first and last two lines.
+4. **Render.** Kept lines stay byte-identical. Each dropped run becomes one
+   `<<ccr:HASH N_lines_offloaded>>` marker whose original text goes into the
+   CCR store. Hashes are deterministic, so identical input yields identical
+   output and prompt caches stay warm.
+
+The plugin never raises. Any failure (missing torch, model download error,
+inference error) falls back to Headroom's own path.
+
+## Installation
+
+```bash
+pip install "headroom-winnow[model]"
+```
+
+Installing changes nothing until you opt in:
+
+```bash
+headroom proxy --compressor winnow
+```
+
+or in Python:
+
+```python
+from headroom.transforms.content_router import ContentRouter, ContentRouterConfig
+
+router = ContentRouter(ContentRouterConfig(active_external_compressors=["winnow"]))
+```
+
+On first use Winnow downloads its model,
+[`rbk4209/winnow-pooled-32m`](https://huggingface.co/rbk4209/winnow-pooled-32m)
+(128 MB), from the Hugging Face Hub at a pinned commit. A GPU is strongly
+recommended; on CPU Winnow only scores shorter outputs and leaves the rest to
+Headroom.
+
+### Models
+
+| Backend | Model | When to use |
+|---|---|---|
+| `pooled` (default) | [`rbk4209/winnow-pooled-32m`](https://huggingface.co/rbk4209/winnow-pooled-32m), 32M line classifier trained for Winnow | Fast enough to score any tool output on a GPU |
+| `highlighter` | [`KRLabsOrg/verbatim-rag-modern-bert-v2`](https://huggingface.co/KRLabsOrg/verbatim-rag-modern-bert-v2), 150M span model | Squeez's original extractive model; slower, so limited to short outputs |
+
+### Configuration
+
+| Variable | Default | Meaning |
+|---|---|---|
+| `HEADROOM_WINNOW_BACKEND` | `pooled` | `pooled` or `highlighter` |
+| `HEADROOM_WINNOW_MODEL` | the backend's published model | Hub id or local path |
+| `HEADROOM_WINNOW_REVISION` | pinned commit of the default model | Model revision |
+| `HEADROOM_WINNOW_DEVICE` | `auto` | `auto` (CUDA if available), `cuda` or `cpu` |
+| `HEADROOM_WINNOW_DTYPE` | `float32` | `float16` is faster only on GPUs with tensor cores |
+| `HEADROOM_WINNOW_MAX_TOKENS` | per backend and device | Largest output the model scores; larger ones go to Headroom |
+
+## Results
+
+On the 618-example test split of the Squeez dataset:
+
+| Method | Recall | Token reduction | Error lines lost | p50 latency |
+|---|---|---|---|---|
+| Headroom `relevance_split` (BM25) | 0.725 | 58.6%¹ | 7,953 | 1 ms |
+| Headroom `relevance_split` (hybrid) | 0.753 | 57.6%¹ | 7,771 | 2.2 s |
+| **Winnow (32M pooled)** | **0.847** | **41.4%** | **0** | **0.29 s** |
+
+¹ Upper bound: the dropped tail is Kompressed inside Headroom, not removed.
+
+At equal compression (50 / 70 / 90% of lines dropped) the pooled model keeps
+0.89 / 0.84 / 0.68 of the gold lines, against 0.73 / 0.63 / 0.46 for
+`relevance_split`. Details, the 150M highlighter numbers and reproduction
+commands are in [`benchmarks/RESULTS.md`](https://github.com/RecepKurtulus/headroom-winnow/blob/main/benchmarks/RESULTS.md).
+
+## Training your own model
+
+[`training/kaggle_train_pooled.ipynb`](https://github.com/RecepKurtulus/headroom-winnow/blob/main/training/kaggle_train_pooled.ipynb)
+trains the 32M pooled classifier on a free Kaggle T4 in about six hours. It
+uses [Squeez](https://github.com/KRLabsOrg/squeez)'s own training code at a
+pinned commit and evaluates on the same test split as the benchmark. Swap
+`--base-model` to try other encoders.
+
+## Project layout
+
+```
+headroom_winnow/
+  compressor.py    WinnowCompressor: the headroom.compressor contract, gates, fail-open
+  selection.py     spans → lines, safety rules, rendering with markers
+  markers.py       CCR marker format and deterministic hashes
+  backends.py      highlighter and pooled backends: lazy, pinned, device-aware
+benchmarks/        compare.py and RESULTS.md
+training/          Kaggle notebook for the pooled model
+upstream/          patch proposed to Headroom (see below)
+tests/             unit tests with a fake model, router end-to-end, real-model tests
+```
+
+## Limitations
+
+- **Headroom's lossless fold runs first.** Headroom applies its byte-exact fold
+  (and, for logs and search results, `relevance_split`) before the external
+  compressor hook and returns early when either succeeds, so such blocks never
+  reach this plugin. See `tests/test_router.py::test_known_gap_lossless_fold_preempts_external`.
+- **Passthroughs skip Headroom's compressors in 0.39.1.** When the plugin
+  declines a block, the router still adopts the unchanged block. The one-line
+  fix is in [`upstream/router-respect-passthrough.patch`](https://github.com/RecepKurtulus/headroom-winnow/blob/main/upstream/router-respect-passthrough.patch);
+  the matching test is a strict `xfail` until it lands.
+
+## Development
+
+```bash
+pip install -e ".[dev,model]"
+ruff check . && ruff format --check . && mypy headroom_winnow tests benchmarks
+pytest               # fast tests, no downloads
+pytest -m slow       # downloads and runs the real highlighter
+```
+
+## Acknowledgements
+
+- [Headroom](https://github.com/headroomlabs-ai/headroom) for the proxy, CCR
+  store and the external compressor contract this plugin implements.
+- [Squeez](https://github.com/KRLabsOrg/squeez) by KRLabs for the task-conditioned
+  pruning approach, the training code, the highlighter model and the dataset.
+- [Ettin](https://huggingface.co/jhu-clsp/ettin-encoder-32m) for the 32M encoder
+  the pooled model is fine-tuned from.
+
+## License
+
+Apache-2.0. See [`LICENSE`](https://github.com/RecepKurtulus/headroom-winnow/blob/main/LICENSE) and [`NOTICE`](https://github.com/RecepKurtulus/headroom-winnow/blob/main/NOTICE).