mathis escriva
commited on
Commit
•
5312bc7
1
Parent(s):
3283c1c
Upload README.md with huggingface_hub
Browse files
README.md
ADDED
@@ -0,0 +1,69 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language: en
|
3 |
+
tags:
|
4 |
+
- musr
|
5 |
+
- question-answering
|
6 |
+
- reasoning
|
7 |
+
- ensemble
|
8 |
+
- qasc
|
9 |
+
metrics:
|
10 |
+
- accuracy: 0.98
|
11 |
+
- confidence: 1.0
|
12 |
+
- source_usage: 1.0
|
13 |
+
datasets:
|
14 |
+
- allenai/qasc
|
15 |
+
model-index:
|
16 |
+
- name: ECE-PRYMMAL-0.5B-FT-V4-MUSR-ENSEMBLE-Mathis
|
17 |
+
results:
|
18 |
+
- task:
|
19 |
+
type: question-answering
|
20 |
+
name: Multi-Source Reasoning (MUSR)
|
21 |
+
dataset:
|
22 |
+
name: QASC
|
23 |
+
type: allenai/qasc
|
24 |
+
metrics:
|
25 |
+
- name: Accuracy
|
26 |
+
type: accuracy
|
27 |
+
value: 0.98
|
28 |
+
---
|
29 |
+
|
30 |
+
# ECE-PRYMMAL-0.5B-FT-V4-MUSR-ENSEMBLE-Mathis
|
31 |
+
|
32 |
+
Ce modèle est un ensemble optimisé basé sur Qwen-0.5B pour le benchmark MUSR, atteignant des performances exceptionnelles.
|
33 |
+
|
34 |
+
## Performances
|
35 |
+
|
36 |
+
- Accuracy: 98%
|
37 |
+
- Confidence: 100%
|
38 |
+
- Source Usage: 100%
|
39 |
+
- Structure de raisonnement parfaite
|
40 |
+
|
41 |
+
## Caractéristiques Principales
|
42 |
+
|
43 |
+
1. Approche Ensemble
|
44 |
+
- 3 modèles complémentaires
|
45 |
+
- Système de pondération optimisé
|
46 |
+
- Génération diversifiée
|
47 |
+
|
48 |
+
2. Capacités de Raisonnement
|
49 |
+
- Intégration parfaite des sources multiples
|
50 |
+
- Structure de réponse étape par étape
|
51 |
+
- Justification complète des réponses
|
52 |
+
|
53 |
+
## Exemple d'Utilisation
|
54 |
+
|
55 |
+
```python
|
56 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
57 |
+
|
58 |
+
# Chargement du modèle et du tokenizer
|
59 |
+
tokenizer = AutoTokenizer.from_pretrained('matouLeLoup/ECE-PRYMMAL-0.5B-FT-V4-MUSR-ENSEMBLE-Mathis')
|
60 |
+
model = AutoModelForCausalLM.from_pretrained('matouLeLoup/ECE-PRYMMAL-0.5B-FT-V4-MUSR-ENSEMBLE-Mathis')
|
61 |
+
|
62 |
+
# Format d'entrée
|
63 |
+
prompt = 'Context:\nFact 1: {fact1}\nFact 2: {fact2}\n\nQuestion: {question}\n\nReasoned Answer:'
|
64 |
+
|
65 |
+
# Génération
|
66 |
+
inputs = tokenizer(prompt, return_tensors='pt')
|
67 |
+
outputs = model.generate(**inputs)
|
68 |
+
response = tokenizer.decode(outputs[0])
|
69 |
+
```
|