Update README.md
Browse files
README.md
CHANGED
@@ -2603,6 +2603,70 @@ license: mit
|
|
2603 |
|
2604 |
# gte-large
|
2605 |
|
2606 |
-
Gegeral Text Embeddings (GTE) model.
|
2607 |
|
2608 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2603 |
|
2604 |
# gte-large
|
2605 |
|
2606 |
+
Gegeral Text Embeddings (GTE) model.
|
2607 |
|
2608 |
+
The GTE models are trained by Alibaba DAMO Academy. They are mainly based on the BERT framework and currently offer three different sizes of models, including [GTE-large](https://huggingface.co/thenlper/gte-large), [GTE-base](https://huggingface.co/thenlper/gte-base), and [GTE-small](https://huggingface.co/thenlper/gte-small). The GTE models are trained on a large-scale corpus of relevance text pairs, covering a wide range of domains and scenarios. This enables the GTE models to be applied to various downstream tasks of text embeddings, including **information retrieval**, **semantic textual similarity**, **text reranking**, etc.
|
2609 |
+
|
2610 |
+
## Metrics
|
2611 |
+
|
2612 |
+
We compared the performance of the GTE models with other popular text embedding models on the MTEB benchmark. For more detailed comparison results, please refer to the [MTEB leaderboard](https://huggingface.co/spaces/mteb/leaderboard).
|
2613 |
+
|
2614 |
+
|
2615 |
+
|
2616 |
+
| Model Name | Model Size (GB) | Dimension | Sequence Length | Average (56) | Clustering (11) | Pair Classification (3) | Reranking (4) | Retrieval (15) | STS (10) | Summarization (1) | Classification (12) |
|
2617 |
+
|:----:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|
2618 |
+
| [**gte-large**](https://huggingface.co/thenlper/gte-large) | 0.67 | 1024 | 512 | **63.13** | 46.84 | 85.00 | 59.13 | 52.22 | 83.35 | 31.66 | 73.33 |
|
2619 |
+
| [**gte-base**](https://huggingface.co/thenlper/gte-base) | 0.22 | 768 | 512 | **62.39** | 46.2 | 84.57 | 58.61 | 51.14 | 82.3 | 31.17 | 73.01 |
|
2620 |
+
| [e5-large-v2](https://huggingface.co/intfloat/e5-large-v2) | 1.34 | 1024| 512 | 62.25 | 44.49 | 86.03 | 56.61 | 50.56 | 82.05 | 30.19 | 75.24 |
|
2621 |
+
| [e5-base-v2](https://huggingface.co/intfloat/e5-base-v2) | 0.44 | 768 | 512 | 61.5 | 43.80 | 85.73 | 55.91 | 50.29 | 81.05 | 30.28 | 73.84 |
|
2622 |
+
| [**gte-small**](https://huggingface.co/thenlper/gte-small) | 0.07 | 384 | 512 | **61.36** | 44.89 | 83.54 | 57.7 | 49.46 | 82.07 | 30.42 | 72.31 |
|
2623 |
+
| [text-embedding-ada-002](https://platform.openai.com/docs/guides/embeddings) | - | 1536 | 8192 | 60.99 | 45.9 | 84.89 | 56.32 | 49.25 | 80.97 | 30.8 | 70.93 |
|
2624 |
+
| [e5-small-v2](https://huggingface.co/intfloat/e5-base-v2) | 0.13 | 384 | 512 | 59.93 | 39.92 | 84.67 | 54.32 | 49.04 | 80.39 | 31.16 | 72.94 |
|
2625 |
+
| [sentence-t5-xxl](https://huggingface.co/sentence-transformers/sentence-t5-xxl) | 9.73 | 768 | 512 | 59.51 | 43.72 | 85.06 | 56.42 | 42.24 | 82.63 | 30.08 | 73.42 |
|
2626 |
+
| [all-mpnet-base-v2](https://huggingface.co/sentence-transformers/all-mpnet-base-v2) | 0.44 | 768 | 514 | 57.78 | 43.69 | 83.04 | 59.36 | 43.81 | 80.28 | 27.49 | 65.07 |
|
2627 |
+
| [sgpt-bloom-7b1-msmarco](https://huggingface.co/bigscience/sgpt-bloom-7b1-msmarco) | 28.27 | 4096 | 2048 | 57.59 | 38.93 | 81.9 | 55.65 | 48.22 | 77.74 | 33.6 | 66.19 |
|
2628 |
+
| [all-MiniLM-L12-v2](https://huggingface.co/sentence-transformers/all-MiniLM-L12-v2) | 0.13 | 384 | 512 | 56.53 | 41.81 | 82.41 | 58.44 | 42.69 | 79.8 | 27.9 | 63.21 |
|
2629 |
+
| [all-MiniLM-L6-v2](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) | 0.09 | 384 | 512 | 56.26 | 42.35 | 82.37 | 58.04 | 41.95 | 78.9 | 30.81 | 63.05 |
|
2630 |
+
| [contriever-base-msmarco](https://huggingface.co/nthakur/contriever-base-msmarco) | 0.44 | 768 | 512 | 56.00 | 41.1 | 82.54 | 53.14 | 41.88 | 76.51 | 30.36 | 66.68 |
|
2631 |
+
| [sentence-t5-base](https://huggingface.co/sentence-transformers/sentence-t5-base) | 0.22 | 768 | 512 | 55.27 | 40.21 | 85.18 | 53.09 | 33.63 | 81.14 | 31.39 | 69.81 |
|
2632 |
+
|
2633 |
+
|
2634 |
+
## Usage
|
2635 |
+
|
2636 |
+
Code example
|
2637 |
+
|
2638 |
+
```
|
2639 |
+
import torch.nn.functional as F
|
2640 |
+
from torch import Tensor
|
2641 |
+
from transformers import AutoTokenizer, AutoModel
|
2642 |
+
|
2643 |
+
def average_pool(last_hidden_states: Tensor,
|
2644 |
+
attention_mask: Tensor) -> Tensor:
|
2645 |
+
last_hidden = last_hidden_states.masked_fill(~attention_mask[..., None].bool(), 0.0)
|
2646 |
+
return last_hidden.sum(dim=1) / attention_mask.sum(dim=1)[..., None]
|
2647 |
+
|
2648 |
+
input_texts = [
|
2649 |
+
"what is the capital of China?",
|
2650 |
+
"how to implement quick sort in python?",
|
2651 |
+
"Beijing",
|
2652 |
+
"sorting algorithms"
|
2653 |
+
]
|
2654 |
+
|
2655 |
+
tokenizer = AutoTokenizer.from_pretrained("thenlper/gte-large")
|
2656 |
+
model = AutoModel.from_pretrained("thenlper/gte-large")
|
2657 |
+
|
2658 |
+
# Tokenize the input texts
|
2659 |
+
batch_dict = tokenizer(input_texts, max_length=512, padding=True, truncation=True, return_tensors='pt')
|
2660 |
+
|
2661 |
+
outputs = model(**batch_dict)
|
2662 |
+
embeddings = average_pool(outputs.last_hidden_state, batch_dict['attention_mask'])
|
2663 |
+
|
2664 |
+
# (Optionally) normalize embeddings
|
2665 |
+
embeddings = F.normalize(embeddings, p=2, dim=1)
|
2666 |
+
scores = (embeddings[:1] @ embeddings[1:].T) * 100
|
2667 |
+
print(scores.tolist())
|
2668 |
+
```
|
2669 |
+
|
2670 |
+
### Limitation
|
2671 |
+
|
2672 |
+
This model exclusively caters to English texts, and any lengthy texts will be truncated to a maximum of 512 tokens.
|