Update README.md
Browse files
README.md
CHANGED
|
@@ -26,8 +26,9 @@ The model is trained jointly on **English, German, Italian, French, and Spanish*
|
|
| 26 |
|
| 27 |
As a baseline, we compare against **`urchade/gliner_multi-v2.1`**, which is built on **`microsoft/mdeberta-v3-base`**, a multilingual DeBERTa v3 model that extends BERT/RoBERTa with **disentangled attention** and an **enhanced mask decoder**, architectural features that improve token-level representations and benefit sequence-labeling tasks such as NER.
|
| 28 |
|
|
|
|
| 29 |
> 🔍 **Demo:** You can try the model directly in the browser via our Hugging Face Space:
|
| 30 |
-
> https://huggingface.co/spaces/VAGOsolutions/
|
| 31 |
|
| 32 |
---
|
| 33 |
|
|
@@ -99,6 +100,9 @@ We compare **SauerkrautLM-GLiNER** against two baselines:
|
|
| 99 |
| science_spanish.jsonl | 555 | 32 | 54.95 / 55.95 / **55.45** | 23.89 / 5.88 / 9.43 |
|
| 100 |
| **AVERAGE** | **20,359** | – | **52.72 / 59.13 / 55.34** | **48.10 / 25.65 / 32.32** |
|
| 101 |
|
|
|
|
|
|
|
|
|
|
| 102 |
**Key Takeaway:** SauerkrautLM-GLiNER achieves **+23.02 F1 points** over gliner_multi-v2.1 on average, with particularly strong improvements in recall across all non-English languages.
|
| 103 |
|
| 104 |
---
|
|
@@ -114,6 +118,9 @@ We compare **SauerkrautLM-GLiNER** against two baselines:
|
|
| 114 |
| pii_spanish.jsonl | 500 | 20 | 55.05 / 36.13 / **43.62** | 39.86 / 20.24 / 26.85 | 66.84 / 33.25 / 44.41 |
|
| 115 |
| **AVERAGE** | **2,500** | – | **56.05 / 37.54 / 44.94** | **36.49 / 18.22 / 24.31** | **68.52 / 32.58 / 44.12** |
|
| 116 |
|
|
|
|
|
|
|
|
|
|
| 117 |
**Key Takeaway:** SauerkrautLM-GLiNER performs competitively on PII detection despite being a general-purpose NER model (not PII-specialized), achieving **+20.63 F1** over gliner_multi-v2.1 and matching the specialized gliner_multi_pii-v1 model.
|
| 118 |
|
| 119 |
---
|
|
|
|
| 26 |
|
| 27 |
As a baseline, we compare against **`urchade/gliner_multi-v2.1`**, which is built on **`microsoft/mdeberta-v3-base`**, a multilingual DeBERTa v3 model that extends BERT/RoBERTa with **disentangled attention** and an **enhanced mask decoder**, architectural features that improve token-level representations and benefit sequence-labeling tasks such as NER.
|
| 28 |
|
| 29 |
+
# Demo
|
| 30 |
> 🔍 **Demo:** You can try the model directly in the browser via our Hugging Face Space:
|
| 31 |
+
> **https://huggingface.co/spaces/VAGOsolutions/SauerkrautLM-GLiNER-Demo**
|
| 32 |
|
| 33 |
---
|
| 34 |
|
|
|
|
| 100 |
| science_spanish.jsonl | 555 | 32 | 54.95 / 55.95 / **55.45** | 23.89 / 5.88 / 9.43 |
|
| 101 |
| **AVERAGE** | **20,359** | – | **52.72 / 59.13 / 55.34** | **48.10 / 25.65 / 32.32** |
|
| 102 |
|
| 103 |
+
|
| 104 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/64b999a40b24527e9c25583a/ZmIvgb5cFJENCRVtw-1qt.png" width="600" height="auto">
|
| 105 |
+
|
| 106 |
**Key Takeaway:** SauerkrautLM-GLiNER achieves **+23.02 F1 points** over gliner_multi-v2.1 on average, with particularly strong improvements in recall across all non-English languages.
|
| 107 |
|
| 108 |
---
|
|
|
|
| 118 |
| pii_spanish.jsonl | 500 | 20 | 55.05 / 36.13 / **43.62** | 39.86 / 20.24 / 26.85 | 66.84 / 33.25 / 44.41 |
|
| 119 |
| **AVERAGE** | **2,500** | – | **56.05 / 37.54 / 44.94** | **36.49 / 18.22 / 24.31** | **68.52 / 32.58 / 44.12** |
|
| 120 |
|
| 121 |
+
|
| 122 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/64b999a40b24527e9c25583a/0I_t3mQVdt-91gaeUhi9p.png" width="600" height="auto">
|
| 123 |
+
|
| 124 |
**Key Takeaway:** SauerkrautLM-GLiNER performs competitively on PII detection despite being a general-purpose NER model (not PII-specialized), achieving **+20.63 F1** over gliner_multi-v2.1 and matching the specialized gliner_multi_pii-v1 model.
|
| 125 |
|
| 126 |
---
|