Ollaya package for EldanRing/Winnow-12B

12b/calibration.json

@@ -0,0 +1,9 @@
+{
+  "temperature": [
+    1.0,
+    1.0,
+    1.0
+  ],
+  "temperature_by_options": {},
+  "source": "Winnow's default decision temperature 1.0; no fitted post-hoc map (release-manifest.json)"
+}

12b/decision.json

@@ -0,0 +1,162 @@
+{
+ "engine": "llama",
+ "family": "winnow",
+ "layout": "winnow-v1",
+ "labels": {
+  "strings": [
+   "A",
+   "B",
+   "C",
+   "D",
+   "E",
+   "F",
+   "G",
+   "H",
+   "I",
+   "J",
+   "K",
+   "L",
+   "M",
+   "N",
+   "O",
+   "P",
+   "Q",
+   "R",
+   "S",
+   "T",
+   "U",
+   "V",
+   "W",
+   "X",
+   "Y",
+   "Z",
+   "AA",
+   "AB",
+   "AC",
+   "AD",
+   "AE",
+   "AF",
+   "AG",
+   "AH",
+   "AI",
+   "AJ",
+   "AK",
+   "AL",
+   "AM",
+   "AN",
+   "AO",
+   "AP",
+   "AQ",
+   "AR",
+   "AS",
+   "AT",
+   "AU",
+   "AV",
+   "AW",
+   "AX",
+   "AY",
+   "AZ",
+   "BA",
+   "BB",
+   "BC",
+   "BD",
+   "BE",
+   "BF",
+   "BG",
+   "BH",
+   "BI",
+   "BJ",
+   "BK",
+   "BL"
+  ],
+  "ids": [
+   236776,
+   236799,
+   236780,
+   236796,
+   236788,
+   236811,
+   236823,
+   236814,
+   236777,
+   236863,
+   236855,
+   236798,
+   236792,
+   236797,
+   236806,
+   236791,
+   236935,
+   236794,
+   236773,
+   236774,
+   236836,
+   236847,
+   236824,
+   236917,
+   236874,
+   236953,
+   8686,
+   3066,
+   2714,
+   2466,
+   14196,
+   11558,
+   2807,
+   20211,
+   12553,
+   57902,
+   9978,
+   1565,
+   1936,
+   1536,
+   70713,
+   2663,
+   80055,
+   1425,
+   2371,
+   1375,
+   22258,
+   7008,
+   35066,
+   26428,
+   5790,
+   25862,
+   6552,
+   20804,
+   5901,
+   11217,
+   12051,
+   28671,
+   54684,
+   48367,
+   11526,
+   127017,
+   74610,
+   8017
+  ]
+ },
+ "thought": true,
+ "max_state_tokens": 6144,
+ "upstream": {
+  "inference": "https://github.com/EldanRing/winnow-inference",
+  "commit": "6c2b3c04e248a319f2cb43832628eba03e55fe38",
+  "source": "native/protocol.h"
+ },
+ "gguf": {
+  "repo": "EldanRing/Winnow-12B",
+  "revision": "b6ac22b0d51b69b18200acacb3fbdd98073fffe8",
+  "path": "gguf/Winnow-12B-Q8_0.gguf",
+  "sha256": "b710efc4c0d048ee61eed92c5fef5ce323a4d17e7c51f9f0533cc72ae50818ea",
+  "size": 12669646592,
+  "quantization": "Q8_0",
+  "architecture": "gemma4",
+  "url": "https://huggingface.co/EldanRing/Winnow-12B/resolve/b6ac22b0d51b69b18200acacb3fbdd98073fffe8/gguf/Winnow-12B-Q8_0.gguf"
+ },
+ "llama": {
+  "n_ctx": 8192,
+  "swa_full": true,
+  "plan": "prefix",
+  "build": "b11146"
+ }
+}

README.md

@@ -0,0 +1,46 @@
+---
+license: apache-2.0
+base_model:
+- EldanRing/Winnow-12B
+- EldanRing/Winnow-E4B
+library_name: onnx
+tags:
+- ollaya
+- onnx
+- decision-model
+- system-one
+pipeline_tag: text-classification
+---
+
+# winnow for Ollaya
+
+[Ollaya](https://github.com/ollaya-dev/ollaya) package of **[EldanRing/Winnow-12B](https://huggingface.co/EldanRing/Winnow-12B)** and **[EldanRing/Winnow-E4B](https://huggingface.co/EldanRing/Winnow-E4B)** by EldanRing.
+Ollaya runs open decision models locally, the way Ollama runs LLMs: typed questions in,
+calibrated answers out, behind a TypeSafe-compatible API.
+
+```sh
+ollaya run winnow
+```
+
+## What is in this repository
+
+This repository holds only the files Ollaya derives, with no weights. Each graph is an ONNX export of the
+original model whose weights **reference the authors' own weight files by byte offset**,
+so `ollaya pull` downloads the weights from the upstream repositories, unmodified and pinned to a
+commit, and verifies their sha256.
+
+| Tag | Upstream | Files |
+|---|---|---|
+| `winnow:12b` | [EldanRing/Winnow-12B@b6ac22b](https://huggingface.co/EldanRing/Winnow-12B/tree/b6ac22b0d51b69b18200acacb3fbdd98073fffe8) | `12b/decision.json`, `12b/calibration.json` |
+| `winnow:e4b` | [EldanRing/Winnow-E4B@734302f](https://huggingface.co/EldanRing/Winnow-E4B/tree/734302fe5fbfeb3f21a7ece62653c9539be4aaf3) | `e4b/decision.json`, `e4b/calibration.json` |
+
+Each tag has an fp32 graph, used on CPU and GPU. Each tag also has `decision.json` (sequence layout, special tokens) and
+`calibration.json` (temperatures).
+
+## Parity
+
+PARITY-PENDING
+
+## License
+
+Same as the upstream model (Apache-2.0). Ollaya itself is Apache-2.0.

e4b/calibration.json

@@ -0,0 +1,9 @@
+{
+  "temperature": [
+    1.2574172017327816,
+    1.2574172017327816,
+    1.2574172017327816
+  ],
+  "temperature_by_options": {},
+  "source": "Winnow-E4B's fitted decision temperature for the Q8_0 GGUF (release-manifest.json decision_calibration, 778 rows)"
+}

e4b/decision.json

@@ -0,0 +1,162 @@
+{
+ "engine": "llama",
+ "family": "winnow",
+ "layout": "winnow-v1",
+ "labels": {
+  "strings": [
+   "A",
+   "B",
+   "C",
+   "D",
+   "E",
+   "F",
+   "G",
+   "H",
+   "I",
+   "J",
+   "K",
+   "L",
+   "M",
+   "N",
+   "O",
+   "P",
+   "Q",
+   "R",
+   "S",
+   "T",
+   "U",
+   "V",
+   "W",
+   "X",
+   "Y",
+   "Z",
+   "AA",
+   "AB",
+   "AC",
+   "AD",
+   "AE",
+   "AF",
+   "AG",
+   "AH",
+   "AI",
+   "AJ",
+   "AK",
+   "AL",
+   "AM",
+   "AN",
+   "AO",
+   "AP",
+   "AQ",
+   "AR",
+   "AS",
+   "AT",
+   "AU",
+   "AV",
+   "AW",
+   "AX",
+   "AY",
+   "AZ",
+   "BA",
+   "BB",
+   "BC",
+   "BD",
+   "BE",
+   "BF",
+   "BG",
+   "BH",
+   "BI",
+   "BJ",
+   "BK",
+   "BL"
+  ],
+  "ids": [
+   236776,
+   236799,
+   236780,
+   236796,
+   236788,
+   236811,
+   236823,
+   236814,
+   236777,
+   236863,
+   236855,
+   236798,
+   236792,
+   236797,
+   236806,
+   236791,
+   236935,
+   236794,
+   236773,
+   236774,
+   236836,
+   236847,
+   236824,
+   236917,
+   236874,
+   236953,
+   8686,
+   3066,
+   2714,
+   2466,
+   14196,
+   11558,
+   2807,
+   20211,
+   12553,
+   57902,
+   9978,
+   1565,
+   1936,
+   1536,
+   70713,
+   2663,
+   80055,
+   1425,
+   2371,
+   1375,
+   22258,
+   7008,
+   35066,
+   26428,
+   5790,
+   25862,
+   6552,
+   20804,
+   5901,
+   11217,
+   12051,
+   28671,
+   54684,
+   48367,
+   11526,
+   127017,
+   74610,
+   8017
+  ]
+ },
+ "thought": false,
+ "max_state_tokens": 6144,
+ "upstream": {
+  "inference": "https://github.com/EldanRing/winnow-inference",
+  "commit": "77d14580c6732ca2f3745750c1dc1fd446d8bcee",
+  "source": "native/protocol.h"
+ },
+ "gguf": {
+  "repo": "EldanRing/Winnow-E4B",
+  "revision": "734302fe5fbfeb3f21a7ece62653c9539be4aaf3",
+  "path": "gguf/Winnow-E4B-Q8_0.gguf",
+  "sha256": "840e3f50e5a9c218727f44e121d1b37cc9e2c3b318c8eb422ba6ef2e27b618a2",
+  "size": 8005437472,
+  "quantization": "Q8_0",
+  "architecture": "gemma4",
+  "url": "https://huggingface.co/EldanRing/Winnow-E4B/resolve/734302fe5fbfeb3f21a7ece62653c9539be4aaf3/gguf/Winnow-E4B-Q8_0.gguf"
+ },
+ "llama": {
+  "n_ctx": 8192,
+  "swa_full": true,
+  "plan": "prefix",
+  "build": "b11146"
+ }
+}