intfloat commited on
Commit
21903a7
1 Parent(s): 3bd6250

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +40 -2
README.md CHANGED
@@ -2641,7 +2641,7 @@ batch_dict = tokenizer(input_texts, max_length=512, padding=True, truncation=Tru
2641
  outputs = model(**batch_dict)
2642
  embeddings = average_pool(outputs.last_hidden_state, batch_dict['attention_mask'])
2643
 
2644
- # (Optionally) normalize embeddings
2645
  embeddings = F.normalize(embeddings, p=2, dim=1)
2646
  scores = (embeddings[:2] @ embeddings[2:].T) * 100
2647
  print(scores.tolist())
@@ -2656,6 +2656,44 @@ Please refer to our paper at [https://arxiv.org/pdf/2212.03533.pdf](https://arxi
2656
  Check out [unilm/e5](https://github.com/microsoft/unilm/tree/master/e5) to reproduce evaluation results
2657
  on the [BEIR](https://arxiv.org/abs/2104.08663) and [MTEB benchmark](https://arxiv.org/abs/2210.07316).
2658
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2659
  ## Citation
2660
 
2661
  If you find our paper or models helpful, please consider cite as follows:
@@ -2671,4 +2709,4 @@ If you find our paper or models helpful, please consider cite as follows:
2671
 
2672
  ## Limitations
2673
 
2674
- This model only works for English texts. Long texts will be truncated to at most 512 tokens.
 
2641
  outputs = model(**batch_dict)
2642
  embeddings = average_pool(outputs.last_hidden_state, batch_dict['attention_mask'])
2643
 
2644
+ # normalize embeddings
2645
  embeddings = F.normalize(embeddings, p=2, dim=1)
2646
  scores = (embeddings[:2] @ embeddings[2:].T) * 100
2647
  print(scores.tolist())
 
2656
  Check out [unilm/e5](https://github.com/microsoft/unilm/tree/master/e5) to reproduce evaluation results
2657
  on the [BEIR](https://arxiv.org/abs/2104.08663) and [MTEB benchmark](https://arxiv.org/abs/2210.07316).
2658
 
2659
+ ## Support for Sentence Transformers
2660
+
2661
+ Below is an example for usage with sentence_transformers.
2662
+ ```python
2663
+ from sentence_transformers import SentenceTransformer
2664
+ model = SentenceTransformer('intfloat/e5-large-v2')
2665
+ input_texts = [
2666
+ 'query: how much protein should a female eat',
2667
+ 'query: summit define',
2668
+ "passage: As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.",
2669
+ "passage: Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments."
2670
+ ]
2671
+ embeddings = model.encode(input_texts, normalize_embeddings=True)
2672
+ ```
2673
+
2674
+ Package requirements
2675
+
2676
+ `pip install sentence_transformers~=2.2.2`
2677
+
2678
+ Contributors: [michaelfeil](https://huggingface.co/michaelfeil)
2679
+
2680
+ ## FAQ
2681
+
2682
+ **1. Do I need to add the prefix "query: " and "passage: " to input texts?**
2683
+
2684
+ Yes, this is how the model is trained, otherwise you will see a performance degradation.
2685
+
2686
+ Here are some rules of thumb:
2687
+ - Use "query: " and "passage: " correspondingly for asymmetric tasks such as passage retrieval in open QA, ad-hoc information retrieval.
2688
+
2689
+ - Use "query: " prefix for symmetric tasks such as semantic similarity, paraphrase retrieval.
2690
+
2691
+ - Use "query: " prefix if you want to use embeddings as features, such as linear probing classification, clustering.
2692
+
2693
+ **2. Why are my reproduced results slightly different from reported in the model card?**
2694
+
2695
+ Different versions of `transformers` and `pytorch` could cause negligible but non-zero performance differences.
2696
+
2697
  ## Citation
2698
 
2699
  If you find our paper or models helpful, please consider cite as follows:
 
2709
 
2710
  ## Limitations
2711
 
2712
+ This model only works for English texts. Long texts will be truncated to at most 512 tokens.