Text Classification
Transformers
ONNX
Safetensors
GLiClass
rlcd
typesafe-ai
jev
decision-engine
system-1
modernbert
gliclass
non-autoregressive
zero-token-generation
structured-outputs
calibration
expected-calibration-error
ece
brier-score
proper-scoring-rules
webgpu
edge-ai
fast-inference
banking77
Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: knowledgator/gliclass-modern-base-v2.0
|
| 4 |
+
tags:
|
| 5 |
+
- decision-engine
|
| 6 |
+
- rlcd
|
| 7 |
+
- modernbert
|
| 8 |
+
- calibration
|
| 9 |
+
- onnx
|
| 10 |
+
- webgpu
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# Verdict-Open-Jev / RLCD ModernBERT (151M)
|
| 14 |
+
|
| 15 |
+
Lightweight, non-autoregressive decision model running client-side in browsers via WebGPU/WASM and locally on PyTorch/ONNX.
|
| 16 |
+
|
| 17 |
+
## Architecture
|
| 18 |
+
- Backbone: knowledgator/gliclass-modern-base-v2.0 (ModernBERT-base, 151M parameters)
|
| 19 |
+
- Loss: Cross-Entropy + Brier Score
|
| 20 |
+
- Post-hoc Calibration: Temperature scaling via L-BFGS
|
| 21 |
+
|
| 22 |
+
## Limitations and Failure Cases
|
| 23 |
+
- Bounded candidate set: Supports up to 24 substantive choices plus 1 explicit abstention option.
|
| 24 |
+
- See failure gallery and full audit reports in the repository.
|