Ollaya package for Contrastive-LM/CLM-v0.1-8B

8b/calibration.json

@@ -0,0 +1,9 @@
+{
+  "temperature": [
+    1.0,
+    1.0,
+    1.0
+  ],
+  "temperature_by_options": {},
+  "source": "none upstream: the probabilities are softmax(scale * cosine), scale = exp(logit_scale) capped at 100"
+}

8b/decision.json

@@ -0,0 +1,71 @@
+{
+  "engine": "onnx",
+  "family": "clm",
+  "layout": "clm-v1",
+  "upstream": {
+    "repo": "Contrastive-LM/CLM-v0.1-8B",
+    "revision": "e939398d4556fcd9400c76fa8c5a513202f42b0a",
+    "head": "CLM_v0.1-8B.pt",
+    "base": "Qwen/Qwen3-8B",
+    "base_revision": "b968826d9c46dd6066d109eabc6255188de91218",
+    "code": "https://github.com/Contrastive-LM/CLM"
+  },
+  "contract": {
+    "inputs": {
+      "input_ids": {
+        "dtype": "int64",
+        "shape": [
+          "rows",
+          "seq"
+        ],
+        "note": "one text per row; right-pad with any id"
+      },
+      "last_pos": {
+        "dtype": "int64",
+        "shape": [
+          "rows"
+        ],
+        "note": "position of the row's last token"
+      },
+      "action": {
+        "dtype": "int64",
+        "shape": [
+          "rows"
+        ],
+        "note": "1 = option text (action head), 0 = state text"
+      }
+    },
+    "outputs": {
+      "z": {
+        "dtype": "float32",
+        "shape": [
+          "rows",
+          512
+        ],
+        "note": "L2-normalised projection"
+      }
+    },
+    "positions": "0..seq-1, implicit",
+    "attention": "causal; no mask input (right padding cannot reach earlier positions)"
+  },
+  "scale": 100.0,
+  "max_text_tokens": 2048,
+  "special_tokens": {
+    "pad": 151643,
+    "add_special_tokens": false
+  },
+  "templates": {
+    "state": "clm.schema.state_text: to_text(state).strip() + '\\n\\n' + to_text(instructions).strip()",
+    "choice_option": "to_text(description), or the key when the description is null or ''",
+    "score_option": "to_text(level)",
+    "noul_options": [
+      "false: {description or 'No. This is false: ' + instructions}",
+      "true: {description or 'Yes. This is true: ' + instructions}"
+    ],
+    "to_text": "clm.schema.to_text: str verbatim; bool true/false; numbers Python str(); dict 'k: v' (top level joined by a blank line, nested indented 2); list '- item' lines"
+  },
+  "option_logits": "scale * z[option row] . z[state row], in option order",
+  "opset": 20,
+  "precision": "fp32 compute; Qwen3-8B weights BF16, widened by Cast; heads F32",
+  "weights_in_memory": "bf16"
+}

8b/model-fp32.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f7b6781f50025560de5a4b2aa60fa4f1c4752559ebeccc5daa00b66a91e684e
+size 7000661

README.md

@@ -0,0 +1,45 @@
+---
+license: apache-2.0
+base_model:
+- Contrastive-LM/CLM-v0.1-8B
+- Qwen/Qwen3-8B
+library_name: onnx
+tags:
+- ollaya
+- onnx
+- decision-model
+- system-one
+pipeline_tag: text-classification
+---
+
+# clm for Ollaya
+
+[Ollaya](https://github.com/ollaya-dev/ollaya) package of **[Contrastive-LM/CLM-v0.1-8B](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B)** and **[Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B)** by Contrastive-LM (projection heads) and the Qwen team (Qwen3-8B encoder).
+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 clm
+```
+
+## 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 |
+|---|---|---|
+| `clm:8b` | [Contrastive-LM/CLM-v0.1-8B@e939398](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B/tree/e939398d4556fcd9400c76fa8c5a513202f42b0a), [Qwen/Qwen3-8B@b968826](https://huggingface.co/Qwen/Qwen3-8B/tree/b968826d9c46dd6066d109eabc6255188de91218) | `8b/model-fp32.onnx`, `8b/decision.json`, `8b/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
+
+Ollaya's Rust runtime matches the reference (upstream clm.schema and heads, the Qwen3-8B encoder in fp32 on its BF16 weights) on 480 questions from 117 requests, on CPU and CUDA: the same state and option texts and token ids, the same rejections, the same decision on every question, option logits within 1.3e-4 and probabilities within 2.9e-5.
+
+## License
+
+Same as the upstream model (Apache-2.0). Ollaya itself is Apache-2.0.