Add first description
Browse files
README.md
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language: fr
|
| 3 |
+
license: mit
|
| 4 |
+
tags:
|
| 5 |
+
- nlp
|
| 6 |
+
- text-generation
|
| 7 |
+
- french
|
| 8 |
+
base_model: asi/gpt-fr-cased-small
|
| 9 |
+
datasets:
|
| 10 |
+
- manu/project_gutenberg
|
| 11 |
+
model_name: gpt-fr-gutenberg-clm
|
| 12 |
+
widget:
|
| 13 |
+
- text: "Une silhouette drapée de voiles noirs errait dans les couloirs du château, et"
|
| 14 |
+
metrics:
|
| 15 |
+
- perplexity
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
# GPT-FR – Gutenberg CLM
|
| 19 |
+
|
| 20 |
+
Modèle GPT-2 affiné sur un roman français
|
| 21 |
+
(“Les belles-de-nuit, tome IV” – Paul Féval, Project Gutenberg #43554).
|
| 22 |
+
Idéal pour la génération créative façon roman-feuilleton XIXᵉ siècle.
|
| 23 |
+
|
| 24 |
+
| | Valeur |
|
| 25 |
+
|-----------------------|--------|
|
| 26 |
+
| **Base** | asi/gpt-fr-cased-small |
|
| 27 |
+
| **Corpus** | 1 roman FR (~1,2 M tokens) |
|
| 28 |
+
| **Époques** | 40 |
|
| 29 |
+
| **Perplexité (val)** | 12 |
|
| 30 |
+
| **Token accuracy** | 88 % |
|
| 31 |
+
|
| 32 |
+
## Exemple
|
| 33 |
+
|
| 34 |
+
```python
|
| 35 |
+
from transformers import pipeline
|
| 36 |
+
gen = pipeline("text-generation",
|
| 37 |
+
model="mkd13mkd/gpt-fr-gutenberg-clm",
|
| 38 |
+
tokenizer="asi/gpt-fr-cased-small",
|
| 39 |
+
max_new_tokens=60, do_sample=True, temperature=0.8)
|
| 40 |
+
print(gen("Une silhouette drapée de voiles noirs errait dans le château, et")[0]["generated_text"])
|