Text Classification
Transformers
ONNX
Safetensors
English
modernbert
rag
governance
hallucination-detection
evidence-verification
pyrrho
fitz-gov-v2
multi-label-classification
text-embeddings-inference
Instructions to use yafitzdev/pyrrho-v2-nano-g1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use yafitzdev/pyrrho-v2-nano-g1 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="yafitzdev/pyrrho-v2-nano-g1")# Load model directly from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("yafitzdev/pyrrho-v2-nano-g1") model = AutoModelForSequenceClassification.from_pretrained("yafitzdev/pyrrho-v2-nano-g1") - Notebooks
- Google Colab
- Kaggle
Update Pyrrho v2 model card terminology
Browse files
README.md
CHANGED
|
@@ -14,8 +14,8 @@ tags:
|
|
| 14 |
|
| 15 |
# pyrrho-v2-nano-g1
|
| 16 |
|
| 17 |
-
`pyrrho-v2-nano-g1` is
|
| 18 |
-
It
|
| 19 |
|
| 20 |
- `evidence_verdict`: `INSUFFICIENT`, `DISPUTED`, `SUFFICIENT`
|
| 21 |
- `failure_mode`: `none`, `unresolved_conflict`, `missing_or_incomplete_evidence`, `wrong_scope_or_version`, `ambiguous_request`
|
|
@@ -24,13 +24,11 @@ It replaces the older Pyrrho g5 multitask shape with four v2 heads:
|
|
| 24 |
|
| 25 |
The model is intended for local governance in `fitz-sage`: decide whether
|
| 26 |
retrieved evidence is sufficient, insufficient, or disputed, and expose
|
| 27 |
-
actionable retrieval/failure metadata.
|
| 28 |
-
post-retrieval evidence governance. Query-only pre-retrieval planning remains
|
| 29 |
-
disabled until a query-trained v2 head is available.
|
| 30 |
|
| 31 |
## Inputs
|
| 32 |
|
| 33 |
-
|
| 34 |
|
| 35 |
```text
|
| 36 |
Question: <user query>
|
|
@@ -40,9 +38,6 @@ Sources:
|
|
| 40 |
[2] <retrieved source text>
|
| 41 |
```
|
| 42 |
|
| 43 |
-
The raw model can be run on query-only text, but `fitz-sage` does not use those
|
| 44 |
-
logits for first-pass retrieval planning in this release.
|
| 45 |
-
|
| 46 |
## Output Decoding
|
| 47 |
|
| 48 |
The 18 logits are not one flat softmax. Decode them by group:
|
|
@@ -100,7 +95,6 @@ kinds = [
|
|
| 100 |
- Dataset: `fitz-gov-v2`
|
| 101 |
- Clean active training rows: 41,358
|
| 102 |
- Training source pointer: `fitz_gov_v2_41358_20260703`
|
| 103 |
-
- Poisoned/quarantined later data is excluded.
|
| 104 |
- Base model: `answerdotai/ModernBERT-base`
|
| 105 |
- Seed: 42
|
| 106 |
|
|
@@ -123,7 +117,7 @@ Fitz-sage benchmark check for this release candidate:
|
|
| 123 |
|
| 124 |
| Benchmark | Result |
|
| 125 |
| --- | ---: |
|
| 126 |
-
| balanced fixed-evidence
|
| 127 |
| live fitz-sage benchmark | 86/120 |
|
| 128 |
|
| 129 |
The live benchmark result is the practical integration target; the fixed-evidence
|
|
|
|
| 14 |
|
| 15 |
# pyrrho-v2-nano-g1
|
| 16 |
|
| 17 |
+
`pyrrho-v2-nano-g1` is the Pyrrho v2 ModernBERT-base classifier for Fitz RAG
|
| 18 |
+
governance. It emits four native v2 heads:
|
| 19 |
|
| 20 |
- `evidence_verdict`: `INSUFFICIENT`, `DISPUTED`, `SUFFICIENT`
|
| 21 |
- `failure_mode`: `none`, `unresolved_conflict`, `missing_or_incomplete_evidence`, `wrong_scope_or_version`, `ambiguous_request`
|
|
|
|
| 24 |
|
| 25 |
The model is intended for local governance in `fitz-sage`: decide whether
|
| 26 |
retrieved evidence is sufficient, insufficient, or disputed, and expose
|
| 27 |
+
actionable retrieval/failure metadata.
|
|
|
|
|
|
|
| 28 |
|
| 29 |
## Inputs
|
| 30 |
|
| 31 |
+
Governance input:
|
| 32 |
|
| 33 |
```text
|
| 34 |
Question: <user query>
|
|
|
|
| 38 |
[2] <retrieved source text>
|
| 39 |
```
|
| 40 |
|
|
|
|
|
|
|
|
|
|
| 41 |
## Output Decoding
|
| 42 |
|
| 43 |
The 18 logits are not one flat softmax. Decode them by group:
|
|
|
|
| 95 |
- Dataset: `fitz-gov-v2`
|
| 96 |
- Clean active training rows: 41,358
|
| 97 |
- Training source pointer: `fitz_gov_v2_41358_20260703`
|
|
|
|
| 98 |
- Base model: `answerdotai/ModernBERT-base`
|
| 99 |
- Seed: 42
|
| 100 |
|
|
|
|
| 117 |
|
| 118 |
| Benchmark | Result |
|
| 119 |
| --- | ---: |
|
| 120 |
+
| balanced fixed-evidence governance sanity suite | 120/120 |
|
| 121 |
| live fitz-sage benchmark | 86/120 |
|
| 122 |
|
| 123 |
The live benchmark result is the practical integration target; the fixed-evidence
|