laya-mnn 0.1.0__tar.gz

laya_mnn-0.1.0/LICENSE

@@ -0,0 +1,176 @@
+                                 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

laya_mnn-0.1.0/NOTICE

@@ -0,0 +1,14 @@
+laya-mnn
+Copyright 2026 yunfengwang
+
+This product includes software and model weights derived from Laya
+(https://github.com/NandhaKishorM/laya, https://huggingface.co/convaiinnovations/laya),
+Copyright Convai Innovations, licensed under the Apache License, Version 2.0.
+
+The prompt construction (build_sequence / render_options), confidence and
+temperature handling in src/laya_mnn/runtime.py are ports of the reference
+implementation in that project, kept numerically identical so that converted
+models reproduce the original outputs. The converted MNN weights are a
+quantized derivative of the published Laya checkpoint.
+
+See LICENSE for the full Apache License, Version 2.0 text.

laya_mnn-0.1.0/PKG-INFO

@@ -0,0 +1,108 @@
+Metadata-Version: 2.4
+Name: laya-mnn
+Version: 0.1.0
+Summary: Run the Laya System 1 decision model locally on MNN (CPU, no PyTorch, no API)
+Author: yunfengwang
+License: Apache-2.0
+Project-URL: Homepage, https://huggingface.co/yunfengwang/laya-MNN-fp16
+Project-URL: Source, https://huggingface.co/yunfengwang/laya-MNN-fp16
+Keywords: mnn,onnx,decision-model,system-one,laya,calibration,edge-inference
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Operating System :: MacOS
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Operating System :: Microsoft :: Windows
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: NOTICE
+Requires-Dist: mnn>=3.6.1
+Requires-Dist: tokenizers>=0.15
+Requires-Dist: numpy>=1.21
+Dynamic: license-file
+
+# laya-mnn
+
+Run the [Laya](https://huggingface.co/convaiinnovations/laya) System 1 decision model **locally on MNN** — no PyTorch, no GPU, no API key.
+
+Laya takes a **state** (email, ticket, JSON) plus **typed questions** and returns typed answers with calibrated probabilities in one forward pass. It never generates text, so there is nothing to parse and nothing to hallucinate. This package runs a converted Laya checkpoint on Alibaba's [MNN](https://github.com/alibaba/MNN) runtime.
+
+```bash
+uvx laya-mnn decide \
+  --state "Hi, we were billed twice for March. Please refund the duplicate or we will cancel our plan." \
+  --questions '{"department":{"type":"choice","instructions":"Which department should handle this?","criteria":{"billing":"invoices, payments, refunds","technical":"bugs, outages","other":"everything else"}},
+                "churn_risk":{"type":"noul","instructions":"Does the user threaten to cancel?"}}'
+```
+
+```json
+{
+  "model": "laya-mnn",
+  "answers": {
+    "department": {
+      "type": "choice", "choice": "billing",
+      "probabilities": { "billing": 0.9631, "technical": 0.0165, "other": 0.0204 },
+      "confidence": 0.8067, "action": { "act_probability": 0.9977 }
+    },
+    "churn_risk": { "type": "noul", "noul": 0.9476, "confidence": 0.9476, "action": { "act_probability": 0.9977 } }
+  }
+}
+```
+
+Weights download automatically on first use (about 850 MB for `fp16`).
+
+## Question types
+
+| type | answer | use for |
+|---|---|---|
+| `choice` | one of the `criteria` keys, plus a probability per option | routing, intent, classification |
+| `score` | expected value over ordered `criteria` levels (0-based) | severity, urgency, sentiment strength |
+| `noul` | `P(true)` | yes/no judgements |
+
+`criteria` is defined **at request time** — new schemas do not need retraining. Up to 64 options per question with the shipped checkpoints.
+
+## CLI
+
+```
+laya-mnn decide --state S --questions Q [--precision fp16|int8] [--threads N] [--quiet]
+laya-mnn download [--precision ...]      # pre-fetch weights
+laya-mnn info     [--precision ...]      # cache dir + manifest
+```
+
+`--state` accepts literal text, inline JSON, or `@path`. `--questions` accepts inline JSON or `@path`.
+
+## Python
+
+```python
+from laya_mnn import LayaMNN, ensure_model
+
+model = LayaMNN(ensure_model("fp16"), threads=4)
+result = model.system_one(
+    {"subject": "Duplicate charge", "body": "We were billed twice..."},
+    {"department": {"type": "choice", "instructions": "Which department?", "criteria": ["billing", "technical", "other"]}},
+)
+print(result["answers"]["department"]["choice"], result["answers"]["department"]["probabilities"])
+```
+
+## Checkpoints
+
+| precision | size | notes |
+|---|---|---|
+| `fp16` | 846 MB | weights stored in half precision; probability deltas vs. the original PyTorch model are below 1e-3 |
+| `int8` | 581 MB | weight-only 8-bit quantization; larger deltas, check it against your own eval |
+
+Both are conversions of `convaiinnovations/laya` (English, ModernBERT-large backbone, 421M parameters). The model was exported to ONNX (`opset 17`, eager attention, static 512-token sequence) and converted with the MNN converter 3.6.1.
+
+Hosted on [Hugging Face](https://huggingface.co/yunfengwang/laya-MNN-fp16) and [ModelScope](https://modelscope.cn/models/yunfeng/laya-MNN-fp16); `modelscope` is tried first, then Hugging Face (override with `HF_ENDPOINT`).
+Set `LAYA_MNN_CACHE` to change the cache directory (default `~/.cache/laya-mnn`).
+
+## Notes and limits
+
+- **CPU is the supported backend.** The MNN Metal backend was measured to return all-zero logits for this graph in some builds of the 3.6.1 wheel; `--backend metal` is exposed but not recommended.
+- A 512-token forward pass costs roughly 1.3 s on an Apple Silicon CPU. Shorter states are cheaper — the graph is exported at a fixed 512 tokens, so the CLI pads and masks.
+- Max 64 options per question, 512 tokens of context, English only. The multilingual checkpoint (`mmBERT-base`) is not converted here.
+- Percentile-latency and calibration in the upstream model card are for the original PyTorch model; quantized results will differ slightly.
+
+## License and attribution
+
+Apache-2.0. The model, its prompt format and the calibration scheme come from [NandhaKishorM/laya](https://github.com/NandhaKishorM/laya) by Convai Innovations (Apache-2.0); this package is an independent MNN port and is not affiliated with or endorsed by them. See `NOTICE`.

laya_mnn-0.1.0/README.md

@@ -0,0 +1,84 @@
+# laya-mnn
+
+Run the [Laya](https://huggingface.co/convaiinnovations/laya) System 1 decision model **locally on MNN** — no PyTorch, no GPU, no API key.
+
+Laya takes a **state** (email, ticket, JSON) plus **typed questions** and returns typed answers with calibrated probabilities in one forward pass. It never generates text, so there is nothing to parse and nothing to hallucinate. This package runs a converted Laya checkpoint on Alibaba's [MNN](https://github.com/alibaba/MNN) runtime.
+
+```bash
+uvx laya-mnn decide \
+  --state "Hi, we were billed twice for March. Please refund the duplicate or we will cancel our plan." \
+  --questions '{"department":{"type":"choice","instructions":"Which department should handle this?","criteria":{"billing":"invoices, payments, refunds","technical":"bugs, outages","other":"everything else"}},
+                "churn_risk":{"type":"noul","instructions":"Does the user threaten to cancel?"}}'
+```
+
+```json
+{
+  "model": "laya-mnn",
+  "answers": {
+    "department": {
+      "type": "choice", "choice": "billing",
+      "probabilities": { "billing": 0.9631, "technical": 0.0165, "other": 0.0204 },
+      "confidence": 0.8067, "action": { "act_probability": 0.9977 }
+    },
+    "churn_risk": { "type": "noul", "noul": 0.9476, "confidence": 0.9476, "action": { "act_probability": 0.9977 } }
+  }
+}
+```
+
+Weights download automatically on first use (about 850 MB for `fp16`).
+
+## Question types
+
+| type | answer | use for |
+|---|---|---|
+| `choice` | one of the `criteria` keys, plus a probability per option | routing, intent, classification |
+| `score` | expected value over ordered `criteria` levels (0-based) | severity, urgency, sentiment strength |
+| `noul` | `P(true)` | yes/no judgements |
+
+`criteria` is defined **at request time** — new schemas do not need retraining. Up to 64 options per question with the shipped checkpoints.
+
+## CLI
+
+```
+laya-mnn decide --state S --questions Q [--precision fp16|int8] [--threads N] [--quiet]
+laya-mnn download [--precision ...]      # pre-fetch weights
+laya-mnn info     [--precision ...]      # cache dir + manifest
+```
+
+`--state` accepts literal text, inline JSON, or `@path`. `--questions` accepts inline JSON or `@path`.
+
+## Python
+
+```python
+from laya_mnn import LayaMNN, ensure_model
+
+model = LayaMNN(ensure_model("fp16"), threads=4)
+result = model.system_one(
+    {"subject": "Duplicate charge", "body": "We were billed twice..."},
+    {"department": {"type": "choice", "instructions": "Which department?", "criteria": ["billing", "technical", "other"]}},
+)
+print(result["answers"]["department"]["choice"], result["answers"]["department"]["probabilities"])
+```
+
+## Checkpoints
+
+| precision | size | notes |
+|---|---|---|
+| `fp16` | 846 MB | weights stored in half precision; probability deltas vs. the original PyTorch model are below 1e-3 |
+| `int8` | 581 MB | weight-only 8-bit quantization; larger deltas, check it against your own eval |
+
+Both are conversions of `convaiinnovations/laya` (English, ModernBERT-large backbone, 421M parameters). The model was exported to ONNX (`opset 17`, eager attention, static 512-token sequence) and converted with the MNN converter 3.6.1.
+
+Hosted on [Hugging Face](https://huggingface.co/yunfengwang/laya-MNN-fp16) and [ModelScope](https://modelscope.cn/models/yunfeng/laya-MNN-fp16); `modelscope` is tried first, then Hugging Face (override with `HF_ENDPOINT`).
+Set `LAYA_MNN_CACHE` to change the cache directory (default `~/.cache/laya-mnn`).
+
+## Notes and limits
+
+- **CPU is the supported backend.** The MNN Metal backend was measured to return all-zero logits for this graph in some builds of the 3.6.1 wheel; `--backend metal` is exposed but not recommended.
+- A 512-token forward pass costs roughly 1.3 s on an Apple Silicon CPU. Shorter states are cheaper — the graph is exported at a fixed 512 tokens, so the CLI pads and masks.
+- Max 64 options per question, 512 tokens of context, English only. The multilingual checkpoint (`mmBERT-base`) is not converted here.
+- Percentile-latency and calibration in the upstream model card are for the original PyTorch model; quantized results will differ slightly.
+
+## License and attribution
+
+Apache-2.0. The model, its prompt format and the calibration scheme come from [NandhaKishorM/laya](https://github.com/NandhaKishorM/laya) by Convai Innovations (Apache-2.0); this package is an independent MNN port and is not affiliated with or endorsed by them. See `NOTICE`.

laya_mnn-0.1.0/pyproject.toml

@@ -0,0 +1,36 @@
+[build-system]
+requires = ["setuptools>=68", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "laya-mnn"
+version = "0.1.0"
+description = "Run the Laya System 1 decision model locally on MNN (CPU, no PyTorch, no API)"
+readme = "README.md"
+requires-python = ">=3.9"
+license = { text = "Apache-2.0" }
+authors = [{ name = "yunfengwang" }]
+keywords = ["mnn", "onnx", "decision-model", "system-one", "laya", "calibration", "edge-inference"]
+classifiers = [
+    "Programming Language :: Python :: 3",
+    "License :: OSI Approved :: Apache Software License",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Operating System :: MacOS",
+    "Operating System :: POSIX :: Linux",
+    "Operating System :: Microsoft :: Windows",
+]
+dependencies = [
+    "mnn>=3.6.1",
+    "tokenizers>=0.15",
+    "numpy>=1.21",
+]
+
+[project.urls]
+Homepage = "https://huggingface.co/yunfengwang/laya-MNN-fp16"
+Source = "https://huggingface.co/yunfengwang/laya-MNN-fp16"
+
+[project.scripts]
+laya-mnn = "laya_mnn.cli:main"
+
+[tool.setuptools.packages.find]
+where = ["src"]

laya_mnn-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

laya_mnn-0.1.0/src/laya_mnn/__init__.py

@@ -0,0 +1,6 @@
+"""Laya decision model on MNN: fast, non-autoregressive, calibrated decisions."""
+from .download import DEFAULT_PRECISION, REPOS, cache_dir, ensure_model
+from .runtime import LayaMNN
+
+__version__ = "0.1.0"
+__all__ = ["LayaMNN", "ensure_model", "cache_dir", "DEFAULT_PRECISION", "REPOS", "__version__"]

laya_mnn-0.1.0/src/laya_mnn/cli.py

@@ -0,0 +1,78 @@
+"""uvx-friendly CLI: download the converted weights, then answer typed questions."""
+import argparse
+import json
+import os
+import sys
+
+from .download import DEFAULT_PRECISION, REPOS, ensure_model
+
+
+def _load_arg(value: str, what: str):
+    """Accept inline JSON, a bare path, or @path."""
+    if value.startswith("@"):
+        value = value[1:]
+    if os.path.exists(value):
+        with open(value, encoding="utf-8") as f:
+            text = f.read().strip()
+        if what == "state":
+            try:
+                return json.loads(text)
+            except json.JSONDecodeError:
+                return text
+        return json.loads(text)
+    if what == "state":
+        try:
+            return json.loads(value)
+        except json.JSONDecodeError:
+            return value
+    return json.loads(value)
+
+
+def main(argv=None):
+    p = argparse.ArgumentParser(prog="laya-mnn",
+                                description="Run the Laya System 1 decision model locally on MNN.")
+    sub = p.add_subparsers(dest="cmd")
+
+    d = sub.add_parser("decide", help="answer typed questions about a state")
+    d.add_argument("--state", required=True, help="email/ticket/JSON text, or @file, or inline JSON")
+    d.add_argument("--questions", required=True, help="questions JSON ({id: {type,instructions,criteria}}), or @file")
+    sub.add_parser("download", help="pre-fetch model weights").add_argument(
+        "--precision", default=DEFAULT_PRECISION, choices=sorted(REPOS))
+    sub.add_parser("info", help="show cache location and model manifest").add_argument(
+        "--precision", default=DEFAULT_PRECISION, choices=sorted(REPOS))
+
+    for sp in (d,):
+        sp.add_argument("--precision", default=DEFAULT_PRECISION, choices=sorted(REPOS))
+        sp.add_argument("--threads", type=int, default=0, help="CPU threads (default 4)")
+        sp.add_argument("--backend", default="cpu", choices=["cpu", "metal"],
+                        help="metal is experimental and may return all-zero logits in some builds")
+        sp.add_argument("--quiet", action="store_true", help="suppress download progress")
+
+    args = p.parse_args(argv)
+    if not args.cmd:
+        p.print_help()
+        return 2
+
+    if args.cmd == "info":
+        d_ = ensure_model(args.precision, quiet=True)
+        print(json.dumps({"cache": d_, "manifest": json.load(open(os.path.join(d_, "manifest.json")))},
+                         indent=2, ensure_ascii=False))
+        return 0
+    if args.cmd == "download":
+        print(ensure_model(args.precision))
+        return 0
+
+    state = _load_arg(args.state, "state")
+    questions = _load_arg(args.questions, "questions")
+
+    from .runtime import LayaMNN
+    model_dir = ensure_model(args.precision, quiet=args.quiet)
+    runtime = LayaMNN(model_dir, threads=args.threads, backend=args.backend)
+    result = runtime.system_one(state, questions)
+    json.dump(result, sys.stdout, indent=2, ensure_ascii=False)
+    sys.stdout.write("\n")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

laya_mnn-0.1.0/src/laya_mnn/download.py

@@ -0,0 +1,118 @@
+"""Fetch converted Laya weights from ModelScope (fast in CN) or Hugging Face."""
+import hashlib
+import json
+import os
+import sys
+import urllib.error
+import urllib.request
+
+DEFAULT_PRECISION = "fp16"
+
+# (registry, repo template) tried in order; ModelScope first because it is
+# reliably fast from mainland China while huggingface.co is often throttled.
+SOURCES = [
+    ("modelscope", "https://modelscope.cn/models/{repo}/resolve/master/{file}"),
+    ("hf", "{endpoint}/{repo}/resolve/main/{file}"),
+]
+
+REPOS = {
+    "fp16": {"hf": "yunfengwang/laya-MNN-fp16", "modelscope": "yunfeng/laya-MNN-fp16"},
+    "int8": {"hf": "yunfengwang/laya-MNN-int8", "modelscope": "yunfeng/laya-MNN-int8"},
+}
+
+
+def cache_dir(precision: str = DEFAULT_PRECISION) -> str:
+    root = os.environ.get("LAYA_MNN_CACHE") or os.path.join(
+        os.environ.get("XDG_CACHE_HOME", os.path.expanduser("~/.cache")), "laya-mnn")
+    return os.path.join(root, precision)
+
+
+def _url(source: str, repo: str, file: str) -> str:
+    endpoint = os.environ.get("HF_ENDPOINT", "https://huggingface.co").rstrip("/")
+    return SOURCES[[s[0] for s in SOURCES].index(source)][1].format(repo=repo, file=file, endpoint=endpoint)
+
+
+def _stream(url: str, dest: str):
+    """Download with resume; returns without error if the server ignores Range."""
+    pos = os.path.getsize(dest) if os.path.exists(dest) else 0
+    req = urllib.request.Request(url, headers={"User-Agent": "laya-mnn"})
+    if pos:
+        req.add_header("Range", "bytes=%d-" % pos)
+    try:
+        with urllib.request.urlopen(req, timeout=60) as r:
+            total = int(r.headers.get("Content-Length", 0)) + (pos if r.status == 206 else 0)
+            if r.status != 206:
+                pos = 0
+            mode = "ab" if pos else "wb"
+            done = pos
+            with open(dest, mode) as f:
+                last = 0.0
+                while True:
+                    chunk = r.read(1 << 20)
+                    if not chunk:
+                        break
+                    f.write(chunk)
+                    done += len(chunk)
+                    if total and done - last > 32 << 20:
+                        last = done
+                        print("  %.0f/%.0f MB (%.0f%%)" % (done / 1e6, total / 1e6, 100.0 * done / total),
+                              file=sys.stderr, flush=True)
+    except urllib.error.HTTPError as e:
+        if e.code == 416 and pos:
+            return
+        raise
+
+
+def fetch_file(precision: str, file: str, dest_dir: str, quiet: bool = False) -> str:
+    os.makedirs(dest_dir, exist_ok=True)
+    dest = os.path.join(dest_dir, file)
+    errors = []
+    for source, _ in SOURCES:
+        repo = REPOS[precision][source]
+        url = _url(source, repo, file)
+        try:
+            if not quiet:
+                print("[laya-mnn] %s <- %s" % (file, url), file=sys.stderr, flush=True)
+            _stream(url, dest)
+            return dest
+        except Exception as e:
+            errors.append("%s: %s" % (source, e))
+    raise RuntimeError("could not download %s\n  %s" % (file, "\n  ".join(errors)))
+
+
+def _sha256(path: str) -> str:
+    h = hashlib.sha256()
+    with open(path, "rb") as f:
+        for chunk in iter(lambda: f.read(1 << 22), b""):
+            h.update(chunk)
+    return h.hexdigest()
+
+
+def ensure_model(precision: str = DEFAULT_PRECISION, quiet: bool = False) -> str:
+    """Return a local dir containing manifest.json, tokenizer.json and the .mnn weights."""
+    if precision not in REPOS:
+        raise ValueError("unknown precision %r (expected one of %s)" % (precision, ", ".join(REPOS)))
+    d = cache_dir(precision)
+    manifest_path = os.path.join(d, "manifest.json")
+    if not os.path.exists(manifest_path):
+        fetch_file(precision, "manifest.json", d, quiet)
+    with open(manifest_path) as f:
+        manifest = json.load(f)
+
+    tok = os.path.join(d, "tokenizer.json")
+    if not os.path.exists(tok):
+        fetch_file(precision, "tokenizer.json", d, quiet)
+
+    weights = os.path.join(d, manifest["model_file"])
+    want = manifest.get("sha256")
+    if os.path.exists(weights) and want:
+        got = _sha256(weights)
+        if got != want:
+            if not quiet:
+                print("[laya-mnn] checksum mismatch, re-downloading", file=sys.stderr)
+            os.remove(weights)
+    if not os.path.exists(weights):
+        fetch_file(precision, manifest["model_file"], d, quiet)
+        if want and _sha256(weights) != want:
+            raise RuntimeError("checksum mismatch after download for %s" % weights)
+    return d

laya_mnn-0.1.0/src/laya_mnn/runtime.py

@@ -0,0 +1,204 @@
+"""Tokenizer, prompt construction and MNN inference for the Laya decision model.
+
+The prompt format and confidence/temperature handling are ported from
+NandhaKishorM/laya (Apache-2.0) so that this runtime matches the reference
+implementation bit for bit; see NOTICE.
+"""
+import json
+import math
+from typing import Any, Dict, List, Optional, Union
+
+import numpy as np
+
+QTYPES = {"choice": 0, "score": 1, "noul": 2}
+QTYPE_NAMES = {v: k for k, v in QTYPES.items()}
+
+CLS_ID, SEP_ID, PAD_ID, MASK_ID = 50281, 50282, 50283, 50284
+
+TEMP_MIN, TEMP_MAX = 0.5, 5.0
+
+
+def render_criterion(value: Any) -> str:
+    if isinstance(value, str):
+        return value
+    return json.dumps(value, ensure_ascii=False, separators=(", ", ": "), default=str)
+
+
+def render_options(q: Dict) -> List[str]:
+    t, crit = q["t"], q.get("crit")
+    if t == "choice":
+        return [k if v is None or v == "" else "%s: %s" % (k, render_criterion(v)) for k, v in crit.items()]
+    if t == "score":
+        return ["level %d: %s" % (i, render_criterion(c)) for i, c in enumerate(crit)]
+    crit = crit or {}
+    false_crit, true_crit = crit.get("false"), crit.get("true")
+    return [
+        "false: " + (render_criterion(false_crit) if false_crit not in (None, "") else "no, the statement does not hold"),
+        "true: " + (render_criterion(true_crit) if true_crit not in (None, "") else "yes, the statement holds"),
+    ]
+
+
+def _encode(tok, text: str) -> List[int]:
+    return tok.encode(text, add_special_tokens=False).ids
+
+
+def build_sequence(tok, state, q, max_len=512, head_max_len=192):
+    """[CLS] <type> question: <instructions> [SEP] [MASK] opt0 [MASK] opt1 ... [SEP] state [SEP]"""
+    opts = render_options(q)
+    order = list(range(len(opts)))
+    ins = str(q["ins"]).replace("[MASK]", " ")
+    head_ids = _encode(tok, "%s question: %s" % (q["t"], ins))
+    opt_ids = []
+    for i in order:
+        opt_ids.append([MASK_ID] + _encode(tok, " " + opts[i].replace("[MASK]", " "))[:48])
+    opt_budget = head_max_len - sum(len(o) for o in opt_ids)
+    if opt_budget < 16:
+        per = max(4, (head_max_len - 16) // max(1, len(opt_ids)))
+        opt_ids = [o[:per] for o in opt_ids]
+        opt_budget = head_max_len - sum(len(o) for o in opt_ids)
+    head_ids = head_ids[: max(8, opt_budget)]
+    ids = [CLS_ID] + head_ids + [SEP_ID]
+    markers = []
+    for o in opt_ids:
+        markers.append(len(ids))
+        ids.extend(o)
+    ids.append(SEP_ID)
+    room = max(0, max_len - len(ids) - 1)
+    st = _encode(tok, state.replace("[MASK]", " "))
+    st = st[:room]
+    ids = ids + st + [SEP_ID]
+    return ids[:max_len], [m for m in markers if m < max_len]
+
+
+def confidence_from_probs(p: np.ndarray, k: int) -> float:
+    if k < 2:
+        return 1.0
+    p = p[:k]
+    ent = -(p * np.log(np.clip(p, 1e-12, 1.0))).sum()
+    return float(np.clip(1.0 - ent / math.log(k), 0.0, 1.0))
+
+
+def temp_bucket(qtype: int, k: int) -> str:
+    size = "2" if k <= 2 else "3-5" if k <= 5 else "6-10" if k <= 10 else "11+"
+    return "%s:%s" % (QTYPE_NAMES[int(qtype)], size)
+
+
+def clamp_temperature(t, lo: float = TEMP_MIN, hi: float = TEMP_MAX) -> float:
+    try:
+        t = float(t)
+    except (TypeError, ValueError):
+        return 1.0
+    if t != t or t in (float("inf"), float("-inf")):
+        return 1.0
+    return min(hi, max(lo, t))
+
+
+def _to_internal(qdef: Dict) -> Dict:
+    t = qdef["type"]
+    crit = qdef.get("criteria")
+    if t == "choice" and isinstance(crit, list):
+        crit = {c: None for c in crit}
+    ins = qdef["instructions"]
+    if not isinstance(ins, str):
+        ins = json.dumps(ins)
+    return {"t": t, "ins": ins, "crit": crit}
+
+
+class LayaMNN:
+    """Runs a converted Laya checkpoint on MNN."""
+
+    def __init__(self, model_dir: str, threads: int = 0, backend: str = "cpu"):
+        import MNN
+        import MNN.expr as F
+        from tokenizers import Tokenizer
+
+        self._F = F
+        with open("%s/manifest.json" % model_dir) as f:
+            self.manifest = json.load(f)
+        self.cfg = self.manifest["config"]
+        self.tok = Tokenizer.from_file("%s/tokenizer.json" % model_dir)
+        self.max_len = self.cfg.get("max_len", 512)
+        self.head_max_len = self.cfg.get("head_max_len", 192)
+        self.max_options = self.manifest.get("max_options", 16)
+        self.temperature = [clamp_temperature(t) for t in self.cfg.get("temperature", [1.0, 1.0, 1.0])]
+        self.temperature_by_options = {k: clamp_temperature(v) for k, v in self.cfg.get("temperature_by_options", {}).items()}
+
+        kw = {"backend": F.Backend.METAL if backend == "metal" else F.Backend.CPU,
+              "thread_num": threads or 4}
+        if backend == "metal":
+            kw["precision_mode"] = F.PrecisionMode.Low
+        names = ["input_ids", "attention_mask", "marker_pos", "marker_mask", "qtype"]
+        self.model = MNN.nn.load_module_from_file(
+            "%s/%s" % (model_dir, self.manifest["model_file"]), names, ["logits", "act_logits"], **kw)
+        self._MNN = MNN
+
+    def _infer(self, items):
+        """items: list of {ids, markers, qtype} -> list of (logits [k], act [2])
+
+        Questions run one at a time: the graph is exported for batch 1, and feeding
+        MNN a wider batch makes its shape inference mishandle the padding mask.
+        Marker tensors are padded to the graph's fixed width; the extra entries stay
+        masked and are dropped here.
+        """
+        F = self._F
+        width = self.max_options
+        out_list = []
+        for it in items:
+            k = len(it["markers"])
+            if k > width:
+                raise ValueError("question has %d options but this model was converted for at most %d; "
+                                 "re-convert with a larger --max-options or split the option set"
+                                 % (k, width))
+            seq = it["ids"][: self.max_len]
+            pad = self.max_len - len(seq)
+            ids = np.array([list(seq) + [PAD_ID] * pad], dtype=np.int32)
+            att = np.array([[1] * len(seq) + [0] * pad], dtype=np.int32)
+            mpos = np.array([it["markers"][:k] + [0] * (width - k)], dtype=np.int32)
+            mmask = np.array([[1] * k + [0] * (width - k)], dtype=np.int32)
+            qtype = np.array([it["qtype"]], dtype=np.int32)
+            inps = [F.const(x, list(x.shape), F.NCHW, F.int) for x in (ids, att, mpos, mmask, qtype)]
+            out = self.model.forward(inps)
+            act = np.array(out[1].read(), dtype=np.float32).reshape(2)
+            act = np.exp(act - act.max())
+            out_list.append((np.array(out[0].read(), dtype=np.float32).reshape(width)[:k],
+                             act / act.sum()))
+        return out_list
+
+    def system_one(self, state, questions: Dict[str, Dict[str, Any]]) -> Dict[str, Any]:
+        state = state if isinstance(state, str) else json.dumps(state, ensure_ascii=False)
+        ids = list(questions.keys())
+        items = []
+        for qid in ids:
+            q = _to_internal(questions[qid])
+            seq, markers = build_sequence(self.tok, state, q, self.max_len, self.head_max_len)
+            if len(markers) != len(render_options(q)):
+                raise ValueError("question %r options exceed head_max_len=%d" % (qid, self.head_max_len))
+            items.append({"ids": seq, "markers": markers, "qtype": QTYPES[q["t"]]})
+
+        inferred = self._infer(items)
+        answers = {}
+        for r, qid in enumerate(ids):
+            q = _to_internal(questions[qid])
+            logit_row, act_row = inferred[r]
+            k = len(items[r]["markers"])
+            qt = QTYPES[q["t"]]
+            t_scale = self.temperature_by_options.get(temp_bucket(qt, k), self.temperature[qt])
+            z = logit_row / t_scale
+            p = np.exp(z - z.max())
+            p = p / p.sum()
+            conf = round(confidence_from_probs(p, k), 4)
+            ext = {"act_probability": round(float(act_row[0]), 4)}
+            if q["t"] == "choice":
+                keys = list(q["crit"].keys())
+                answers[qid] = {"type": "choice", "choice": keys[int(p.argmax())],
+                                "probabilities": {kk: round(float(v), 4) for kk, v in zip(keys, p)},
+                                "confidence": conf, "action": ext}
+            elif q["t"] == "score":
+                answers[qid] = {"type": "score", "score": round(float((np.arange(k) * p).sum()), 4),
+                                "legend": {str(i): c for i, c in enumerate(q["crit"])},
+                                "probabilities": {str(i): round(float(v), 4) for i, v in enumerate(p)},
+                                "confidence": conf, "action": ext}
+            else:
+                answers[qid] = {"type": "noul", "noul": round(float(p[1]), 4),
+                                "confidence": round(max(float(p[1]), 1.0 - float(p[1])), 4), "action": ext}
+        return {"model": "laya-mnn", "answers": answers}

laya_mnn-0.1.0/src/laya_mnn.egg-info/PKG-INFO

@@ -0,0 +1,108 @@
+Metadata-Version: 2.4
+Name: laya-mnn
+Version: 0.1.0
+Summary: Run the Laya System 1 decision model locally on MNN (CPU, no PyTorch, no API)
+Author: yunfengwang
+License: Apache-2.0
+Project-URL: Homepage, https://huggingface.co/yunfengwang/laya-MNN-fp16
+Project-URL: Source, https://huggingface.co/yunfengwang/laya-MNN-fp16
+Keywords: mnn,onnx,decision-model,system-one,laya,calibration,edge-inference
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Operating System :: MacOS
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Operating System :: Microsoft :: Windows
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: NOTICE
+Requires-Dist: mnn>=3.6.1
+Requires-Dist: tokenizers>=0.15
+Requires-Dist: numpy>=1.21
+Dynamic: license-file
+
+# laya-mnn
+
+Run the [Laya](https://huggingface.co/convaiinnovations/laya) System 1 decision model **locally on MNN** — no PyTorch, no GPU, no API key.
+
+Laya takes a **state** (email, ticket, JSON) plus **typed questions** and returns typed answers with calibrated probabilities in one forward pass. It never generates text, so there is nothing to parse and nothing to hallucinate. This package runs a converted Laya checkpoint on Alibaba's [MNN](https://github.com/alibaba/MNN) runtime.
+
+```bash
+uvx laya-mnn decide \
+  --state "Hi, we were billed twice for March. Please refund the duplicate or we will cancel our plan." \
+  --questions '{"department":{"type":"choice","instructions":"Which department should handle this?","criteria":{"billing":"invoices, payments, refunds","technical":"bugs, outages","other":"everything else"}},
+                "churn_risk":{"type":"noul","instructions":"Does the user threaten to cancel?"}}'
+```
+
+```json
+{
+  "model": "laya-mnn",
+  "answers": {
+    "department": {
+      "type": "choice", "choice": "billing",
+      "probabilities": { "billing": 0.9631, "technical": 0.0165, "other": 0.0204 },
+      "confidence": 0.8067, "action": { "act_probability": 0.9977 }
+    },
+    "churn_risk": { "type": "noul", "noul": 0.9476, "confidence": 0.9476, "action": { "act_probability": 0.9977 } }
+  }
+}
+```
+
+Weights download automatically on first use (about 850 MB for `fp16`).
+
+## Question types
+
+| type | answer | use for |
+|---|---|---|
+| `choice` | one of the `criteria` keys, plus a probability per option | routing, intent, classification |
+| `score` | expected value over ordered `criteria` levels (0-based) | severity, urgency, sentiment strength |
+| `noul` | `P(true)` | yes/no judgements |
+
+`criteria` is defined **at request time** — new schemas do not need retraining. Up to 64 options per question with the shipped checkpoints.
+
+## CLI
+
+```
+laya-mnn decide --state S --questions Q [--precision fp16|int8] [--threads N] [--quiet]
+laya-mnn download [--precision ...]      # pre-fetch weights
+laya-mnn info     [--precision ...]      # cache dir + manifest
+```
+
+`--state` accepts literal text, inline JSON, or `@path`. `--questions` accepts inline JSON or `@path`.
+
+## Python
+
+```python
+from laya_mnn import LayaMNN, ensure_model
+
+model = LayaMNN(ensure_model("fp16"), threads=4)
+result = model.system_one(
+    {"subject": "Duplicate charge", "body": "We were billed twice..."},
+    {"department": {"type": "choice", "instructions": "Which department?", "criteria": ["billing", "technical", "other"]}},
+)
+print(result["answers"]["department"]["choice"], result["answers"]["department"]["probabilities"])
+```
+
+## Checkpoints
+
+| precision | size | notes |
+|---|---|---|
+| `fp16` | 846 MB | weights stored in half precision; probability deltas vs. the original PyTorch model are below 1e-3 |
+| `int8` | 581 MB | weight-only 8-bit quantization; larger deltas, check it against your own eval |
+
+Both are conversions of `convaiinnovations/laya` (English, ModernBERT-large backbone, 421M parameters). The model was exported to ONNX (`opset 17`, eager attention, static 512-token sequence) and converted with the MNN converter 3.6.1.
+
+Hosted on [Hugging Face](https://huggingface.co/yunfengwang/laya-MNN-fp16) and [ModelScope](https://modelscope.cn/models/yunfeng/laya-MNN-fp16); `modelscope` is tried first, then Hugging Face (override with `HF_ENDPOINT`).
+Set `LAYA_MNN_CACHE` to change the cache directory (default `~/.cache/laya-mnn`).
+
+## Notes and limits
+
+- **CPU is the supported backend.** The MNN Metal backend was measured to return all-zero logits for this graph in some builds of the 3.6.1 wheel; `--backend metal` is exposed but not recommended.
+- A 512-token forward pass costs roughly 1.3 s on an Apple Silicon CPU. Shorter states are cheaper — the graph is exported at a fixed 512 tokens, so the CLI pads and masks.
+- Max 64 options per question, 512 tokens of context, English only. The multilingual checkpoint (`mmBERT-base`) is not converted here.
+- Percentile-latency and calibration in the upstream model card are for the original PyTorch model; quantized results will differ slightly.
+
+## License and attribution
+
+Apache-2.0. The model, its prompt format and the calibration scheme come from [NandhaKishorM/laya](https://github.com/NandhaKishorM/laya) by Convai Innovations (Apache-2.0); this package is an independent MNN port and is not affiliated with or endorsed by them. See `NOTICE`.

laya_mnn-0.1.0/src/laya_mnn.egg-info/SOURCES.txt

@@ -0,0 +1,14 @@
+LICENSE
+NOTICE
+README.md
+pyproject.toml
+src/laya_mnn/__init__.py
+src/laya_mnn/cli.py
+src/laya_mnn/download.py
+src/laya_mnn/runtime.py
+src/laya_mnn.egg-info/PKG-INFO
+src/laya_mnn.egg-info/SOURCES.txt
+src/laya_mnn.egg-info/dependency_links.txt
+src/laya_mnn.egg-info/entry_points.txt
+src/laya_mnn.egg-info/requires.txt
+src/laya_mnn.egg-info/top_level.txt

laya_mnn-0.1.0/src/laya_mnn.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

laya_mnn-0.1.0/src/laya_mnn.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+laya-mnn = laya_mnn.cli:main

laya_mnn-0.1.0/src/laya_mnn.egg-info/requires.txt

@@ -0,0 +1,3 @@
+mnn>=3.6.1
+tokenizers>=0.15
+numpy>=1.21

laya_mnn-0.1.0/src/laya_mnn.egg-info/top_level.txt

@@ -0,0 +1 @@
+laya_mnn