mixed-nlp commited on
Commit
82918cd
1 Parent(s): 9a24e4f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +84 -14
README.md CHANGED
@@ -1195,7 +1195,7 @@ model-index:
1195
  - type: map_at_5
1196
  value: 15.271
1197
  - type: mrr_at_1
1198
- value: 69.0
1199
  - type: mrr_at_10
1200
  value: 75.304
1201
  - type: mrr_at_100
@@ -1219,9 +1219,9 @@ model-index:
1219
  - type: ndcg_at_5
1220
  value: 42.104
1221
  - type: precision_at_1
1222
- value: 69.0
1223
  - type: precision_at_10
1224
- value: 33.0
1225
  - type: precision_at_100
1226
  value: 10.75
1227
  - type: precision_at_1000
@@ -1815,7 +1815,7 @@ model-index:
1815
  - type: ndcg_at_3
1816
  value: 85.435
1817
  - type: ndcg_at_5
1818
- value: 87.0
1819
  - type: precision_at_1
1820
  value: 81.24
1821
  - type: precision_at_10
@@ -1910,13 +1910,13 @@ model-index:
1910
  - type: precision_at_1
1911
  value: 24.8
1912
  - type: precision_at_10
1913
- value: 12.0
1914
  - type: precision_at_100
1915
  value: 2.5420000000000003
1916
  - type: precision_at_1000
1917
  value: 0.39899999999999997
1918
  - type: precision_at_3
1919
- value: 20.0
1920
  - type: precision_at_5
1921
  value: 17.4
1922
  - type: recall_at_1
@@ -2197,7 +2197,7 @@ model-index:
2197
  - type: recall_at_100
2198
  value: 96.167
2199
  - type: recall_at_1000
2200
- value: 100.0
2201
  - type: recall_at_3
2202
  value: 74.117
2203
  - type: recall_at_5
@@ -2250,7 +2250,7 @@ model-index:
2250
  - type: manhattan_precision
2251
  value: 91.72482552342971
2252
  - type: manhattan_recall
2253
- value: 92.0
2254
  - type: max_accuracy
2255
  value: 99.83861386138614
2256
  - type: max_ap
@@ -2331,7 +2331,7 @@ model-index:
2331
  - type: map_at_5
2332
  value: 1.001
2333
  - type: mrr_at_1
2334
- value: 76.0
2335
  - type: mrr_at_10
2336
  value: 85.667
2337
  - type: mrr_at_100
@@ -2343,7 +2343,7 @@ model-index:
2343
  - type: mrr_at_5
2344
  value: 85.667
2345
  - type: ndcg_at_1
2346
- value: 72.0
2347
  - type: ndcg_at_10
2348
  value: 68.637
2349
  - type: ndcg_at_100
@@ -2355,7 +2355,7 @@ model-index:
2355
  - type: ndcg_at_5
2356
  value: 71.808
2357
  - type: precision_at_1
2358
- value: 76.0
2359
  - type: precision_at_10
2360
  value: 73.8
2361
  - type: precision_at_100
@@ -2365,7 +2365,7 @@ model-index:
2365
  - type: precision_at_3
2366
  value: 74.667
2367
  - type: precision_at_5
2368
- value: 78.0
2369
  - type: recall_at_1
2370
  value: 0.22100000000000003
2371
  - type: recall_at_10
@@ -2596,6 +2596,10 @@ model-index:
2596
  value: 85.53503846009764
2597
  - type: max_f1
2598
  value: 77.68167368965773
 
 
 
 
2599
  ---
2600
 
2601
  <br><br>
@@ -2605,7 +2609,7 @@ model-index:
2605
  </p>
2606
 
2607
  <p align="center">
2608
- <b>The crispy rerank family from <a href="https://mixedbread.ai"><b>mixedbread ai</b></a>.</b>
2609
  </p>
2610
 
2611
  # mxbai-embed-2d-large-v1
@@ -2617,7 +2621,73 @@ model-index:
2617
 
2618
  Currently, the best way to use our models is with the most recent version of sentence-transformers.
2619
 
2620
- `pip install -U sentence-transformers`
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2621
 
2622
  ### angle-emb
2623
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1195
  - type: map_at_5
1196
  value: 15.271
1197
  - type: mrr_at_1
1198
+ value: 69
1199
  - type: mrr_at_10
1200
  value: 75.304
1201
  - type: mrr_at_100
 
1219
  - type: ndcg_at_5
1220
  value: 42.104
1221
  - type: precision_at_1
1222
+ value: 69
1223
  - type: precision_at_10
1224
+ value: 33
1225
  - type: precision_at_100
1226
  value: 10.75
1227
  - type: precision_at_1000
 
1815
  - type: ndcg_at_3
1816
  value: 85.435
1817
  - type: ndcg_at_5
1818
+ value: 87
1819
  - type: precision_at_1
1820
  value: 81.24
1821
  - type: precision_at_10
 
1910
  - type: precision_at_1
1911
  value: 24.8
1912
  - type: precision_at_10
1913
+ value: 12
1914
  - type: precision_at_100
1915
  value: 2.5420000000000003
1916
  - type: precision_at_1000
1917
  value: 0.39899999999999997
1918
  - type: precision_at_3
1919
+ value: 20
1920
  - type: precision_at_5
1921
  value: 17.4
1922
  - type: recall_at_1
 
2197
  - type: recall_at_100
2198
  value: 96.167
2199
  - type: recall_at_1000
2200
+ value: 100
2201
  - type: recall_at_3
2202
  value: 74.117
2203
  - type: recall_at_5
 
2250
  - type: manhattan_precision
2251
  value: 91.72482552342971
2252
  - type: manhattan_recall
2253
+ value: 92
2254
  - type: max_accuracy
2255
  value: 99.83861386138614
2256
  - type: max_ap
 
2331
  - type: map_at_5
2332
  value: 1.001
2333
  - type: mrr_at_1
2334
+ value: 76
2335
  - type: mrr_at_10
2336
  value: 85.667
2337
  - type: mrr_at_100
 
2343
  - type: mrr_at_5
2344
  value: 85.667
2345
  - type: ndcg_at_1
2346
+ value: 72
2347
  - type: ndcg_at_10
2348
  value: 68.637
2349
  - type: ndcg_at_100
 
2355
  - type: ndcg_at_5
2356
  value: 71.808
2357
  - type: precision_at_1
2358
+ value: 76
2359
  - type: precision_at_10
2360
  value: 73.8
2361
  - type: precision_at_100
 
2365
  - type: precision_at_3
2366
  value: 74.667
2367
  - type: precision_at_5
2368
+ value: 78
2369
  - type: recall_at_1
2370
  value: 0.22100000000000003
2371
  - type: recall_at_10
 
2596
  value: 85.53503846009764
2597
  - type: max_f1
2598
  value: 77.68167368965773
2599
+ license: apache-2.0
2600
+ language:
2601
+ - en
2602
+ library_name: transformers
2603
  ---
2604
 
2605
  <br><br>
 
2609
  </p>
2610
 
2611
  <p align="center">
2612
+ <b>The crispy sentence embedding family from <a href="https://mixedbread.ai"><b>mixedbread ai</b></a>.</b>
2613
  </p>
2614
 
2615
  # mxbai-embed-2d-large-v1
 
2621
 
2622
  Currently, the best way to use our models is with the most recent version of sentence-transformers.
2623
 
2624
+ ```bash
2625
+ python -m pip install -U sentence-transformers
2626
+ ```
2627
+
2628
+
2629
+ ```python
2630
+ from sentence_transformers import models, SentenceTransformer
2631
+ from sentence_transformers.util import cos_sim
2632
+
2633
+
2634
+ # 1. load model with `cls` pooling
2635
+ word_embedding_model = models.Transformer("mixedbread-ai/mxbai-embed-2d-large-v1")
2636
+ pooling_model = models.Pooling(word_embedding_model.get_word_embedding_dimension(), pooling_mode="cls")
2637
+ model = SentenceTransformer(modules=[word_embedding_model, pooling_model])
2638
+
2639
+ # 2. set adaptive layer and embedding size.
2640
+ # it is recommended to set layers from 20 to 24.
2641
+ new_num_layers = 22 # 1d: layer
2642
+ model[0].auto_model.encoder.layer = model[0].auto_model.encoder.layer[:new_num_layers]
2643
+ new_embedding_size = 768 # 2d: embedding size
2644
+
2645
+
2646
+ # 3. encode
2647
+ embeddings = model.encode(
2648
+ [
2649
+ 'Who is german and likes bread?',
2650
+ 'Everybody in German.'
2651
+ ]
2652
+ )
2653
+
2654
+ # Similarity of the first sentence with the other two
2655
+ similarities = cos_sim(embeddings[0, :new_embedding_size], embeddings[1, :new_embedding_size])
2656
+
2657
+ print('similarities:', similarities)
2658
+ ```
2659
 
2660
  ### angle-emb
2661
 
2662
+ You can also use the lastest `angle-emb` for inference, as follows:
2663
+
2664
+ ```bash
2665
+ python -m pip install -U angle-emb
2666
+ ```
2667
+
2668
+ ```python
2669
+ from angle_emb import AnglE
2670
+ from sentence_transformers.util import cos_sim
2671
+
2672
+ # 1. load model
2673
+ model = AnglE.from_pretrained("mixedbread-ai/mxbai-embed-2d-large-v1", pooling_strategy='cls').cuda()
2674
+
2675
+
2676
+ # 2. set adaptive layer and embedding size.
2677
+ # it is recommended to set layers from 20 to 24.
2678
+ layer_index = 22 # 1d: layer
2679
+ embedding_size = 768 # 2d: embedding size
2680
+
2681
+ # 3. encode
2682
+ embeddings = model.encode([
2683
+ 'Who is german and likes bread?',
2684
+ 'Everybody in German.'
2685
+ ], layer_index=layer_index, embedding_size=embedding_size)
2686
+
2687
+ similarities = cos_sim(embeddings[0], embeddings[1:])
2688
+ print('similarities:', similarities)
2689
+ ```
2690
+
2691
+ ### Using API
2692
+ You’ll be able to use the models through our API as well. The API is coming soon and will have some exciting features. Stay tuned!
2693
+