Instructions to use TencentBAC/U-MARVEL-Qwen2VL-7B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use TencentBAC/U-MARVEL-Qwen2VL-7B-Instruct with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModelForSeq2SeqLM processor = AutoProcessor.from_pretrained("TencentBAC/U-MARVEL-Qwen2VL-7B-Instruct") model = AutoModelForSeq2SeqLM.from_pretrained("TencentBAC/U-MARVEL-Qwen2VL-7B-Instruct", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Update README.md
Browse files
README.md
CHANGED
|
@@ -6,4 +6,17 @@ language:
|
|
| 6 |
- en
|
| 7 |
base_model:
|
| 8 |
- Qwen/Qwen2-VL-7B-Instruct
|
| 9 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
- en
|
| 7 |
base_model:
|
| 8 |
- Qwen/Qwen2-VL-7B-Instruct
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
| **Task** | **Dataset** | **Metric** | **CLIP-L** | **SigLIP** | **UniIR-BLIPFF** | **UniIR-CLIPSF** | **LamRA-Ret** | **U-MARVEL** | | **LamRA** | **U-MARVEL+** |
|
| 12 |
+
| ------------------------------------------------------------ | ----------- | ---------- | ---------------- | ---------------- | ---------------- | ---------------- | ---------------- | ---------------- | ---- | ------------- | ------------- |
|
| 13 |
+
| | | | **single model** | **single model** | **single model** | **single model** | **single model** | **single model** | | **+reranker** | **+reranker** |
|
| 14 |
+
| qᵗ → cⁱ | VisualNews<br>MSCOCO<br>Fashion200K | R@5<br>R@5<br>R@10 | 43.3<br>61.1<br>6.6 | 30.1<br>75.7<br>36.5 | 23.4<br>79.7<br>26.1 | 42.6<br>81.1<br>18 | 41.6<br>81.5<br>28.7 | **47.3**<br>**84.4**<br>33.6 | | 48<br>85.2<br>32.9 | **49.4**<br>85.6<br>34.2 |
|
| 15 |
+
| qᵗ → cᵗ | WebQA | R@5 | 36.2 | 39.8 | 80 | 84.7 | 86 | **97.1** | | 96.7 | **98.5** |
|
| 16 |
+
| qᵗ → (cⁱ, cᵗ) | EDIS<br>WebQA | R@5<br>R@5 | 43.3<br>45.1 | 27<br>43.5 | 50.9<br>79.8 | 59.4<br>78.7 | 62.6<br>81.2 | **78.8**<br>**88.5** | | 75.8<br>87.7 | **81.4**<br>89.4 |
|
| 17 |
+
| qⁱ → cᵗ | VisualNews<br>MSCOCO<br>Fashion200K | R@5<br>R@5<br>R@10 | 41.3<br>79<br>7.7 | 30.8<br>88.2<br>34.2 | 22.8<br>89.9<br>28.9 | 43.1<br>92.3<br>18.3 | 39.6<br>90.6<br>30.4 | **47.3**<br>**93.5**<br>35.1 | | 48.6<br>92.3<br>36.1 | **50.5**<br>88.4<br>37.7 |
|
| 18 |
+
| qⁱ → cⁱ | NIGHTS | R@5 | 26.1 | 28.9 | 33 | 32 | 32.1 | **34.2** | | 33.5 | **34.7** |
|
| 19 |
+
| (qⁱ, qᵗ) → cᵗ | OVEN<br>InfoSeek | R@5<br>R@5 | 24.2<br>20.5 | 29.7<br>25.1 | 41<br>22.4 | 45.5<br>27.9 | 54.1<br>52.1 | **62.5**<br>**58.3** | | 59.2<br>64.1 | **63.7**<br>62.9 |
|
| 20 |
+
| (qⁱ, qᵗ) → cⁱ | FashionIQ<br>CIRR | R@10<br>R@5 | 7<br>13.2 | 14.4<br>22.7 | 29.2<br>52.2 | 24.4<br>44.6 | 33.2<br>53.1 | **36.4**<br>**60.7** | | 37.8<br>63.3 | **38.2**<br>63.2 |
|
| 21 |
+
| (qⁱ, qᵗ) → (cⁱ, cᵗ) | OVEN<br>InfoSeek | R@5<br>R@5 | 38.8<br>26.4 | 41.7<br>27.4 | 55.8<br>33 | 67.6<br>48.9 | 76.2<br>63.3 | **79.4**<br>**74.7** | | 79.2<br>78.3 | **80.8**<br>78.9 |
|
| 22 |
+
| | Avg. | | 32.5 | 37.2 | 46.8 | 50.6 | 56.6 | **63.2** | | 63.7 | **64.8** |
|