Model card: quantized export of fastino/GLiNER2.5-multi-Decide (Apache-2.0)
Browse files
README.md
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: fastino/GLiNER2.5-multi-Decide
|
| 4 |
+
tags:
|
| 5 |
+
- onnx
|
| 6 |
+
- gliner2
|
| 7 |
+
- decision-model
|
| 8 |
+
- on-device
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# river-decider
|
| 12 |
+
|
| 13 |
+
Mobile ONNX encoder for the on-device agent of River (MonkeyMesh). It is a quantized export of
|
| 14 |
+
[fastino/GLiNER2.5-multi-Decide](https://huggingface.co/fastino/GLiNER2.5-multi-Decide) (Apache-2.0),
|
| 15 |
+
derived from the community ONNX conversion
|
| 16 |
+
[DanKau/gliner2.5-multi-decide-onnx](https://huggingface.co/DanKau/gliner2.5-multi-decide-onnx).
|
| 17 |
+
|
| 18 |
+
- `encoder.onnx`: int8 embedding table, weights stored fp16 and computed in fp32. 364 MB.
|
| 19 |
+
sha256 `3078c2351e140f0c4f6f7a6d0349c167dd74f289c425e98490e9e807a0b5ec2a`.
|
| 20 |
+
- Max probability difference to the fp32 reference on 119 test posts: 0.054; labels 474/476 identical.
|
| 21 |
+
- Use with `onnx/classifier.onnx` and `tokenizer.json` from the DanKau repo.
|
| 22 |
+
|
| 23 |
+
License: Apache-2.0, same as the original model. All credit for the model goes to Fastino.
|