Text Classification
Transformers
Safetensors
gemma4
gevva
cross-encoder
nli
gemma-4
system1
decision-engine
fast-inference
multimodal
vision
long-context
128k
zero-shot
tool-routing
reranking
hallucination-detection
Eval Results (legacy)
Instructions to use davidburhans/gevva-e2b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use davidburhans/gevva-e2b with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="davidburhans/gevva-e2b")# Load model directly from transformers import AutoProcessor, AutoModelForSequenceClassification processor = AutoProcessor.from_pretrained("davidburhans/gevva-e2b") model = AutoModelForSequenceClassification.from_pretrained("davidburhans/gevva-e2b", device_map="auto") - Notebooks
- Google Colab
- Kaggle
fix: set default optimal_temperature to 1.0 to preserve calibration on general NLI data
Browse files- calibration.json +6 -8
calibration.json
CHANGED
|
@@ -1,16 +1,14 @@
|
|
| 1 |
{
|
| 2 |
"model_name": "Gevva e2b",
|
| 3 |
"project": "Gevva",
|
| 4 |
-
"optimal_temperature": 1.
|
| 5 |
-
"
|
| 6 |
-
"
|
| 7 |
-
"
|
| 8 |
-
"
|
| 9 |
-
"val_brier_before": 0.2067,
|
| 10 |
-
"val_brier_after": 0.2063,
|
| 11 |
"jevbench_hard_ece_at_t10": 0.1604,
|
| 12 |
"jevbench_hard_ece_at_t16": 0.0655,
|
| 13 |
"composite_jevbench_score_t10": 72.94,
|
| 14 |
"composite_jevbench_score_t16": 77.54,
|
| 15 |
-
"
|
| 16 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"model_name": "Gevva e2b",
|
| 3 |
"project": "Gevva",
|
| 4 |
+
"optimal_temperature": 1.0,
|
| 5 |
+
"fitted_validation_temperature_nll": 0.8496,
|
| 6 |
+
"jevbench_public_temperature": 1.6,
|
| 7 |
+
"val_ece_at_t10": 0.0237,
|
| 8 |
+
"val_brier_at_t10": 0.2067,
|
|
|
|
|
|
|
| 9 |
"jevbench_hard_ece_at_t10": 0.1604,
|
| 10 |
"jevbench_hard_ece_at_t16": 0.0655,
|
| 11 |
"composite_jevbench_score_t10": 72.94,
|
| 12 |
"composite_jevbench_score_t16": 77.54,
|
| 13 |
+
"notes": "Default inference runs at standard T=1.0 to preserve calibration on general NLI data. T=1.6 was fitted specifically on JevBench Public hard items."
|
| 14 |
}
|