densonsmith commited on
Commit
96f8fcf
·
verified ·
1 Parent(s): 7a971f7

Add new SentenceTransformer model

Browse files
Files changed (2) hide show
  1. README.md +212 -175
  2. model.safetensors +1 -1
README.md CHANGED
@@ -7,132 +7,168 @@ tags:
7
  - sentence-similarity
8
  - feature-extraction
9
  - generated_from_trainer
10
- - dataset_size:318
11
  - loss:MatryoshkaLoss
12
  - loss:MultipleNegativesRankingLoss
13
  base_model: nomic-ai/modernbert-embed-base
14
  widget:
15
- - source_sentence: What type of deal did Conan O'Brien sign with SiriusXM?
16
  sentences:
17
- - 'References:
18
-
19
- - ([Andy Richter Net Worth | Celebrity Net Worth](https://www.celebritynetworth.com))
20
-
21
 
22
- Media:
23
 
24
- - image: https://upload.wikimedia.org/wikipedia/commons/8/8f/Conan_O%27Brien_and_Andy_Richter.jpg
25
 
26
- - video: https://www.youtube.com/watch?v=WKLSAeIuK-8'
27
- - "Description: Max Weinberg led the band on 'Late Night' and 'The Tonight Show\
28
- \ with Conan O'Brien' before departing in 2010.\n- TBS-era bandleader jimmy_vivino\
29
- \ (Strength: strong)\n Description: Jimmy Vivino took over as bandleader on Conan's\
30
- \ TBS show after Max Weinberg stepped away.\n- Spouse liza_powel_obrien (Strength:\
31
- \ very strong)"
32
- - "Relationships:\n- Owner team_coco (Strength: strong)\n Description: SiriusXM\
33
- \ owns Team Coco and its digital content library.\n- On-air talent conan_obrien\
34
- \ (Strength: strong)\n Description: Conan signed a multi-year deal with SiriusXM\
35
- \ to produce content.\n\nAwards and Recognitions:\n- Leader in comedy audio"
36
- - source_sentence: What webcomic did Pierre Bernard co-create with Conan O'Brien?
37
- sentences:
38
- - 'Graph: Team Coco Knowledge Graph
39
 
40
- Node ID: pierre_bernard
41
 
42
- Category: people
43
 
44
- Name: Pierre Bernard
 
 
45
 
46
- Type: Person
47
 
 
48
 
49
- Description: Pierre Bernard is a graphic artist recognized for his illustration
50
- work on Conan O''Brien''s shows and for co-creating the ''Morrigan Webcomic''
51
- with Conan.'
52
- - "Relationships:\n- Host conan_obrien (Strength: creator)\n Description: Conan\
53
- \ developed his comedic voice on 'Late Night'.\n- Original sidekick andy_richter\
54
- \ (Strength: core cast)\n Description: Andy Richter contributed greatly to the\
55
- \ show's dynamic.\n- Network nbc (Strength: parent)\n Description: 'Late Night'\
56
- \ was produced and aired on NBC.\n- Production company conaco (Strength: partner)"
57
- - "Relationships:\n- Host & friend conan_obrien (Strength: very strong)\n Description:\
58
- \ Andy has been Conan's primary collaborator across multiple talk shows and tours.\n\
59
- - Performer team_coco (Strength: strong)\n Description: Andy is a prominent member\
60
- \ of Team Coco, hosting live events and his own podcast series."
61
- - source_sentence: What type of entity is Liza Powel O'Brien classified as?
62
  sentences:
63
  - 'Graph: Team Coco Knowledge Graph
64
 
65
- Node ID: 2010_nbc_conflict
66
 
67
  Category: events
68
 
69
- Name: 2010 Tonight Show Conflict
70
 
71
  Type: Event
72
 
73
 
74
- Description: A public and professional clash in 2010 when NBC reinstated Jay Leno
75
- as host of ''The Tonight Show'', leading Conan to leave the network.
 
76
 
77
 
78
  Relationships:
79
 
80
- - Key figure conan_obrien
81
 
82
- - Network nbc'
83
- - "Relationships:\n- Late-night host conan_obrien (Strength: strong)\n Description:\
84
- \ Conan headlined TBS's late-night lineup for 11 years.\n- Digital content partner\
85
- \ team_coco (Strength: strong)\n Description: Team Coco produced many online\
86
- \ segments during the TBS era.\n\nAwards and Recognitions:\n- Innovative blend\
87
- \ of on-air and online content\n- Consistent ratings success"
88
- - 'Graph: Team Coco Knowledge Graph
89
-
90
- Node ID: liza_powel_obrien
91
 
92
- Category: people
 
93
 
94
- Name: Liza Powel O''Brien
95
 
96
- Type: Person'
97
- - source_sentence: What is the name of the podcast production team?
 
98
  sentences:
99
- - 'Description: Smigel voices and operates the Triumph puppet in comedic sketches.'
 
 
 
100
  - 'Major Events:
101
 
102
- - 1954 ''Tonight'' Launch – Established late-night TV with ''The Tonight Show''.
103
 
104
- - 2009 Transition Briefly handed ''Tonight'' to Conan, then reinstated Leno.'
105
- - "- Producer team_coco (Strength: parent)\n Description: The podcast is produced\
106
- \ by Team Coco in partnership with Stitcher's Earwolf."
107
- - source_sentence: What role did Andy Richter have when he joined 'Late Night'?
 
 
 
108
  sentences:
 
 
 
 
 
 
 
 
 
 
 
109
  - 'Graph: Team Coco Knowledge Graph
110
 
111
- Node ID: jeff_ross_producer
112
 
113
- Category: people
114
 
115
- Name: Jeff Ross (Producer)
116
 
117
- Type: Person
118
 
119
 
120
- Description: Jeff Ross is a television producer who has served as Conan O''Brien''s
121
- executive producer since 1993. He is a key business partner in Conan''s media
122
- ventures and helped establish the Team Coco brand.'
123
- - 'Collaboration Timeline:
 
 
 
 
 
 
 
 
 
124
 
125
- - 1993: Joined Conan''s ''Late Night'' as sidekick ([Late Night with Conan O''Brien
126
- - Wikipedia](https://en.wikipedia.org/wiki/Late_Night_with_Conan_O%27Brien))
127
 
128
- - 2000: Left ''Late Night'' to pursue acting ([Andy Richter - Wikipedia](https://en.wikipedia.org/wiki/Andy_Richter))
129
 
130
- - 2009: Reunited with Conan as announcer for NBC''s ''Tonight Show'''
131
- - "Relationships:\n- Founder & figurehead conan_obrien (Strength: very strong)\n\
132
- \ Description: Conan drives the creative vision of Team Coco.\n- Parent company\
133
- \ siriusxm (Strength: strong)\n Description: SiriusXM acquired Team Coco in 2022.\n\
134
- - Production partner conaco (Strength: strong)\n Description: Team Coco collaborates\
135
- \ with Conaco to produce Conan's TV and streaming content."
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
136
  pipeline_tag: sentence-similarity
137
  library_name: sentence-transformers
138
  metrics:
@@ -168,10 +204,10 @@ model-index:
168
  value: 0.8611111111111112
169
  name: Cosine Accuracy@3
170
  - type: cosine_accuracy@5
171
- value: 0.8888888888888888
172
  name: Cosine Accuracy@5
173
  - type: cosine_accuracy@10
174
- value: 1.0
175
  name: Cosine Accuracy@10
176
  - type: cosine_precision@1
177
  value: 0.7222222222222222
@@ -180,10 +216,10 @@ model-index:
180
  value: 0.2870370370370371
181
  name: Cosine Precision@3
182
  - type: cosine_precision@5
183
- value: 0.1777777777777778
184
  name: Cosine Precision@5
185
  - type: cosine_precision@10
186
- value: 0.10000000000000002
187
  name: Cosine Precision@10
188
  - type: cosine_recall@1
189
  value: 0.7222222222222222
@@ -192,19 +228,19 @@ model-index:
192
  value: 0.8611111111111112
193
  name: Cosine Recall@3
194
  - type: cosine_recall@5
195
- value: 0.8888888888888888
196
  name: Cosine Recall@5
197
  - type: cosine_recall@10
198
- value: 1.0
199
  name: Cosine Recall@10
200
  - type: cosine_ndcg@10
201
- value: 0.8504185091821412
202
  name: Cosine Ndcg@10
203
  - type: cosine_mrr@10
204
- value: 0.8041225749559083
205
  name: Cosine Mrr@10
206
  - type: cosine_map@100
207
- value: 0.8041225749559082
208
  name: Cosine Map@100
209
  - task:
210
  type: information-retrieval
@@ -217,46 +253,46 @@ model-index:
217
  value: 0.6944444444444444
218
  name: Cosine Accuracy@1
219
  - type: cosine_accuracy@3
220
- value: 0.8333333333333334
221
  name: Cosine Accuracy@3
222
  - type: cosine_accuracy@5
223
- value: 0.8611111111111112
224
  name: Cosine Accuracy@5
225
  - type: cosine_accuracy@10
226
- value: 1.0
227
  name: Cosine Accuracy@10
228
  - type: cosine_precision@1
229
  value: 0.6944444444444444
230
  name: Cosine Precision@1
231
  - type: cosine_precision@3
232
- value: 0.27777777777777785
233
  name: Cosine Precision@3
234
  - type: cosine_precision@5
235
- value: 0.17222222222222225
236
  name: Cosine Precision@5
237
  - type: cosine_precision@10
238
- value: 0.10000000000000002
239
  name: Cosine Precision@10
240
  - type: cosine_recall@1
241
  value: 0.6944444444444444
242
  name: Cosine Recall@1
243
  - type: cosine_recall@3
244
- value: 0.8333333333333334
245
  name: Cosine Recall@3
246
  - type: cosine_recall@5
247
- value: 0.8611111111111112
248
  name: Cosine Recall@5
249
  - type: cosine_recall@10
250
- value: 1.0
251
  name: Cosine Recall@10
252
  - type: cosine_ndcg@10
253
- value: 0.8349549866777143
254
  name: Cosine Ndcg@10
255
  - type: cosine_mrr@10
256
- value: 0.7842151675485008
257
  name: Cosine Mrr@10
258
  - type: cosine_map@100
259
- value: 0.7842151675485008
260
  name: Cosine Map@100
261
  - task:
262
  type: information-retrieval
@@ -266,49 +302,49 @@ model-index:
266
  type: dim_256
267
  metrics:
268
  - type: cosine_accuracy@1
269
- value: 0.6944444444444444
270
  name: Cosine Accuracy@1
271
  - type: cosine_accuracy@3
272
- value: 0.8333333333333334
273
  name: Cosine Accuracy@3
274
  - type: cosine_accuracy@5
275
- value: 0.8611111111111112
276
  name: Cosine Accuracy@5
277
  - type: cosine_accuracy@10
278
- value: 1.0
279
  name: Cosine Accuracy@10
280
  - type: cosine_precision@1
281
- value: 0.6944444444444444
282
  name: Cosine Precision@1
283
  - type: cosine_precision@3
284
- value: 0.27777777777777785
285
  name: Cosine Precision@3
286
  - type: cosine_precision@5
287
- value: 0.17222222222222225
288
  name: Cosine Precision@5
289
  - type: cosine_precision@10
290
- value: 0.10000000000000002
291
  name: Cosine Precision@10
292
  - type: cosine_recall@1
293
- value: 0.6944444444444444
294
  name: Cosine Recall@1
295
  - type: cosine_recall@3
296
- value: 0.8333333333333334
297
  name: Cosine Recall@3
298
  - type: cosine_recall@5
299
- value: 0.8611111111111112
300
  name: Cosine Recall@5
301
  - type: cosine_recall@10
302
- value: 1.0
303
  name: Cosine Recall@10
304
  - type: cosine_ndcg@10
305
- value: 0.8363970656972288
306
  name: Cosine Ndcg@10
307
  - type: cosine_mrr@10
308
- value: 0.7866071428571428
309
  name: Cosine Mrr@10
310
  - type: cosine_map@100
311
- value: 0.7866071428571428
312
  name: Cosine Map@100
313
  - task:
314
  type: information-retrieval
@@ -318,49 +354,49 @@ model-index:
318
  type: dim_128
319
  metrics:
320
  - type: cosine_accuracy@1
321
- value: 0.6111111111111112
322
  name: Cosine Accuracy@1
323
  - type: cosine_accuracy@3
324
- value: 0.7777777777777778
325
  name: Cosine Accuracy@3
326
  - type: cosine_accuracy@5
327
- value: 0.8055555555555556
328
  name: Cosine Accuracy@5
329
  - type: cosine_accuracy@10
330
- value: 0.9722222222222222
331
  name: Cosine Accuracy@10
332
  - type: cosine_precision@1
333
- value: 0.6111111111111112
334
  name: Cosine Precision@1
335
  - type: cosine_precision@3
336
- value: 0.2592592592592593
337
  name: Cosine Precision@3
338
  - type: cosine_precision@5
339
- value: 0.16111111111111112
340
  name: Cosine Precision@5
341
  - type: cosine_precision@10
342
- value: 0.09722222222222224
343
  name: Cosine Precision@10
344
  - type: cosine_recall@1
345
- value: 0.6111111111111112
346
  name: Cosine Recall@1
347
  - type: cosine_recall@3
348
- value: 0.7777777777777778
349
  name: Cosine Recall@3
350
  - type: cosine_recall@5
351
- value: 0.8055555555555556
352
  name: Cosine Recall@5
353
  - type: cosine_recall@10
354
- value: 0.9722222222222222
355
  name: Cosine Recall@10
356
  - type: cosine_ndcg@10
357
- value: 0.7747574471542903
358
  name: Cosine Ndcg@10
359
  - type: cosine_mrr@10
360
- value: 0.7143077601410934
361
  name: Cosine Mrr@10
362
  - type: cosine_map@100
363
- value: 0.7158509700176366
364
  name: Cosine Map@100
365
  - task:
366
  type: information-retrieval
@@ -370,49 +406,49 @@ model-index:
370
  type: dim_64
371
  metrics:
372
  - type: cosine_accuracy@1
373
- value: 0.5
374
  name: Cosine Accuracy@1
375
  - type: cosine_accuracy@3
376
- value: 0.6944444444444444
377
  name: Cosine Accuracy@3
378
  - type: cosine_accuracy@5
379
- value: 0.8055555555555556
380
  name: Cosine Accuracy@5
381
  - type: cosine_accuracy@10
382
  value: 0.9166666666666666
383
  name: Cosine Accuracy@10
384
  - type: cosine_precision@1
385
- value: 0.5
386
  name: Cosine Precision@1
387
  - type: cosine_precision@3
388
- value: 0.23148148148148145
389
  name: Cosine Precision@3
390
  - type: cosine_precision@5
391
- value: 0.16111111111111112
392
  name: Cosine Precision@5
393
  - type: cosine_precision@10
394
  value: 0.09166666666666669
395
  name: Cosine Precision@10
396
  - type: cosine_recall@1
397
- value: 0.5
398
  name: Cosine Recall@1
399
  - type: cosine_recall@3
400
- value: 0.6944444444444444
401
  name: Cosine Recall@3
402
  - type: cosine_recall@5
403
- value: 0.8055555555555556
404
  name: Cosine Recall@5
405
  - type: cosine_recall@10
406
  value: 0.9166666666666666
407
  name: Cosine Recall@10
408
  - type: cosine_ndcg@10
409
- value: 0.6915040161197078
410
  name: Cosine Ndcg@10
411
  - type: cosine_mrr@10
412
- value: 0.6216049382716049
413
  name: Cosine Mrr@10
414
  - type: cosine_map@100
415
- value: 0.6252403054619593
416
  name: Cosine Map@100
417
  ---
418
 
@@ -466,9 +502,9 @@ from sentence_transformers import SentenceTransformer
466
  model = SentenceTransformer("densonsmith/modernbert-embed-quickb")
467
  # Run inference
468
  sentences = [
469
- "What role did Andy Richter have when he joined 'Late Night'?",
470
- "Collaboration Timeline:\n- 1993: Joined Conan's 'Late Night' as sidekick ([Late Night with Conan O'Brien - Wikipedia](https://en.wikipedia.org/wiki/Late_Night_with_Conan_O%27Brien))\n- 2000: Left 'Late Night' to pursue acting ([Andy Richter - Wikipedia](https://en.wikipedia.org/wiki/Andy_Richter))\n- 2009: Reunited with Conan as announcer for NBC's 'Tonight Show'",
471
- "Graph: Team Coco Knowledge Graph\nNode ID: jeff_ross_producer\nCategory: people\nName: Jeff Ross (Producer)\nType: Person\n\nDescription: Jeff Ross is a television producer who has served as Conan O'Brien's executive producer since 1993. He is a key business partner in Conan's media ventures and helped establish the Team Coco brand.",
472
  ]
473
  embeddings = model.encode(sentences)
474
  print(embeddings.shape)
@@ -515,21 +551,21 @@ You can finetune this model on your own dataset.
515
 
516
  | Metric | dim_768 | dim_512 | dim_256 | dim_128 | dim_64 |
517
  |:--------------------|:-----------|:----------|:-----------|:-----------|:-----------|
518
- | cosine_accuracy@1 | 0.7222 | 0.6944 | 0.6944 | 0.6111 | 0.5 |
519
- | cosine_accuracy@3 | 0.8611 | 0.8333 | 0.8333 | 0.7778 | 0.6944 |
520
- | cosine_accuracy@5 | 0.8889 | 0.8611 | 0.8611 | 0.8056 | 0.8056 |
521
- | cosine_accuracy@10 | 1.0 | 1.0 | 1.0 | 0.9722 | 0.9167 |
522
- | cosine_precision@1 | 0.7222 | 0.6944 | 0.6944 | 0.6111 | 0.5 |
523
- | cosine_precision@3 | 0.287 | 0.2778 | 0.2778 | 0.2593 | 0.2315 |
524
- | cosine_precision@5 | 0.1778 | 0.1722 | 0.1722 | 0.1611 | 0.1611 |
525
- | cosine_precision@10 | 0.1 | 0.1 | 0.1 | 0.0972 | 0.0917 |
526
- | cosine_recall@1 | 0.7222 | 0.6944 | 0.6944 | 0.6111 | 0.5 |
527
- | cosine_recall@3 | 0.8611 | 0.8333 | 0.8333 | 0.7778 | 0.6944 |
528
- | cosine_recall@5 | 0.8889 | 0.8611 | 0.8611 | 0.8056 | 0.8056 |
529
- | cosine_recall@10 | 1.0 | 1.0 | 1.0 | 0.9722 | 0.9167 |
530
- | **cosine_ndcg@10** | **0.8504** | **0.835** | **0.8364** | **0.7748** | **0.6915** |
531
- | cosine_mrr@10 | 0.8041 | 0.7842 | 0.7866 | 0.7143 | 0.6216 |
532
- | cosine_map@100 | 0.8041 | 0.7842 | 0.7866 | 0.7159 | 0.6252 |
533
 
534
  <!--
535
  ## Bias, Risks and Limitations
@@ -549,19 +585,19 @@ You can finetune this model on your own dataset.
549
 
550
  #### Unnamed Dataset
551
 
552
- * Size: 318 training samples
553
  * Columns: <code>anchor</code> and <code>positive</code>
554
- * Approximate statistics based on the first 318 samples:
555
- | | anchor | positive |
556
- |:--------|:----------------------------------------------------------------------------------|:-----------------------------------------------------------------------------------|
557
- | type | string | string |
558
- | details | <ul><li>min: 7 tokens</li><li>mean: 14.24 tokens</li><li>max: 28 tokens</li></ul> | <ul><li>min: 15 tokens</li><li>mean: 74.8 tokens</li><li>max: 117 tokens</li></ul> |
559
  * Samples:
560
- | anchor | positive |
561
- |:----------------------------------------------------------------------------------------------------------------|:---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
562
- | <code>What is the Node ID of TBS?</code> | <code>Graph: Team Coco Knowledge Graph<br>Node ID: tbs<br>Category: organizations<br>Name: TBS<br>Type: Organization<br><br>Description: TBS is the cable network that hosted Conan O'Brien's late-night show 'Conan' from 2010 until 2021.</code> |
563
- | <code>Where can information about 'The Legally Prohibited from Being Funny on Television Tour' be found?</code> | <code>References:<br>- ([The Legally Prohibited from Being Funny on Television Tour - Wikipedia](https://en.wikipedia.org/wiki/The_Legally_Prohibited_from_Being_Funny_on_Television_Tour))<br>- ([15 Years Ago, Conan O'Brien's 'Tonight Show'...](https://www.tvinsider.com))</code> |
564
- | <code>What is the strength classification for the creator & performer triumph_insult_comic_dog?</code> | <code>Relationships:<br>- Creative collaborator conan_obrien (Strength: strong)<br> Description: Robert helped shape Conan's early comedy, including the creation of signature characters.<br>- Instrumental in sidekick role andy_richter (Strength: moderate)<br> Description: Smigel recognized Andy's on-camera chemistry with Conan.<br>- Creator & performer triumph_insult_comic_dog (Strength: N/A)</code> |
565
  * Loss: [<code>MatryoshkaLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#matryoshkaloss) with these parameters:
566
  ```json
567
  {
@@ -724,10 +760,11 @@ You can finetune this model on your own dataset.
724
  ### Training Logs
725
  | Epoch | Step | Training Loss | dim_768_cosine_ndcg@10 | dim_512_cosine_ndcg@10 | dim_256_cosine_ndcg@10 | dim_128_cosine_ndcg@10 | dim_64_cosine_ndcg@10 |
726
  |:-------:|:------:|:-------------:|:----------------------:|:----------------------:|:----------------------:|:----------------------:|:---------------------:|
727
- | 1.0 | 5 | - | 0.7899 | 0.7843 | 0.7653 | 0.7110 | 0.6216 |
728
- | 2.0 | 10 | 18.3776 | 0.8343 | 0.8248 | 0.8061 | 0.7663 | 0.6764 |
729
- | **3.0** | **15** | **-** | **0.8504** | **0.835** | **0.8226** | **0.7748** | **0.6894** |
730
- | 4.0 | 20 | 6.1577 | 0.8504 | 0.8350 | 0.8364 | 0.7748 | 0.6915 |
 
731
 
732
  * The bold row denotes the saved checkpoint.
733
 
 
7
  - sentence-similarity
8
  - feature-extraction
9
  - generated_from_trainer
10
+ - dataset_size:321
11
  - loss:MatryoshkaLoss
12
  - loss:MultipleNegativesRankingLoss
13
  base_model: nomic-ai/modernbert-embed-base
14
  widget:
15
+ - source_sentence: Since what year have they been married?
16
  sentences:
17
+ - 'Graph: Team Coco Knowledge Graph
 
 
 
18
 
19
+ Node ID: 2015_conan_cuba
20
 
21
+ Category: events
22
 
23
+ Name: Conan in Cuba
 
 
 
 
 
 
 
 
 
 
 
 
24
 
25
+ Type: Event
26
 
 
27
 
28
+ Description: Conan O''Brien traveled to Havana to film a historic episode—the
29
+ first by an American late-night host in over 50 years—part of his ''Conan Without
30
+ Borders'' specials.
31
 
 
32
 
33
+ Relationships:
34
 
35
+ - Host conan_obrien
36
+
37
+ - Occurred during conan_tbs'
38
+ - 'Description: Liza Powel O''Brien is an American playwright and podcast host.
39
+ She met Conan O''Brien in 2000 while working at an advertising agency, and they
40
+ married in 2002. She has written numerous plays staged at theaters like the Geffen
41
+ Playhouse and Ojai Playwrights Conference, and in 2022 she launched the history
42
+ podcast "Significant Others" on Conan''s Team Coco network.'
43
+ - "Relationships:\n- Spouse conan_obrien (Strength: very strong)\n Description:\
44
+ \ Married since 2002; they have two children together.\n- Podcast host team_coco\
45
+ \ (Strength: moderate)\n Description: Hosts the \"Significant Others\" podcast\
46
+ \ under the Team Coco banner."
47
+ - source_sentence: Which team produced Conan's final late night episode?
48
  sentences:
49
  - 'Graph: Team Coco Knowledge Graph
50
 
51
+ Node ID: 2021_conan_finale
52
 
53
  Category: events
54
 
55
+ Name: Conan''s Final Late Night Episode
56
 
57
  Type: Event
58
 
59
 
60
+ Description: The final episode of ''Conan'' on TBS, marking the end of Conan O''Brien''s
61
+ 28-year run as a late-night host with heartfelt goodbyes and memorable comedy
62
+ moments.
63
 
64
 
65
  Relationships:
66
 
67
+ - Honoree conan_obrien
68
 
69
+ - Participant andy_richter
 
 
 
 
 
 
 
 
70
 
71
+ - Producer team_coco'
72
+ - 'References:
73
 
74
+ - ([Conan O''Brien - Wikipedia](https://en.wikipedia.org/wiki/Conan_O%27Brien))
75
 
76
+ - ([Andy Richter Net Worth | Celebrity Net Worth](https://www.celebritynetworth.com))'
77
+ - 'Description: Airing on SiriusXM''s Team Coco Radio channel.'
78
+ - source_sentence: What type of document is referenced for the tour?
79
  sentences:
80
+ - "Relationships:\n- Late-night host conan_obrien (Strength: core talent)\n Description:\
81
+ \ Conan's break in late night came through NBC.\n- Production partner conaco (Strength:\
82
+ \ strong)\n Description: NBC worked with Conaco on Conan's shows.\n\nAwards and\
83
+ \ Recognitions:\n- Legacy of late-night programming"
84
  - 'Major Events:
85
 
86
+ - 1993 Joined ''Late Night'' with Conan
87
 
88
+ - 2009 Transitioned to ''The Tonight Show''
89
+
90
+ - 2010 Concluded run as Conan''s bandleader'
91
+ - 'References:
92
+
93
+ - ([The Legally Prohibited from Being Funny on Television Tour - Wikipedia](https://en.wikipedia.org/wiki/The_Legally_Prohibited_from_Being_Funny_on_Television_Tour))'
94
+ - source_sentence: In what year did Triumph the Insult Comic Dog debut?
95
  sentences:
96
+ - "Relationships:\n- Host-guest (Prankster) conan_obrien (Strength: moderate)\n\
97
+ \ Description: Repeatedly played the 'Mac and Me' gag, to Conan's feigned exasperation.\n\
98
+ \nMajor Events:\n- 2004 First Mac and Me Gag on 'Late Night'\n- 2021 Final TBS\
99
+ \ Show Prank cameo"
100
+ - 'Awards and Recognitions:
101
+
102
+ - MFA in Fiction Writing from Columbia University
103
+
104
+ - Playwright with works at the Geffen Playhouse and Ojai Playwrights Conference
105
+
106
+ - Host of the "Significant Others" podcast (2022–present)'
107
  - 'Graph: Team Coco Knowledge Graph
108
 
109
+ Node ID: triumph_insult_comic_dog
110
 
111
+ Category: creative works
112
 
113
+ Name: Triumph the Insult Comic Dog
114
 
115
+ Type: Puppet character
116
 
117
 
118
+ Description: A recurring canine puppet character, voiced by Robert Smigel, that
119
+ debuted on Conan''s ''Late Night'' in 1997, known for roasting celebrities and
120
+ absurd humor.
121
+
122
+
123
+ Relationships:
124
+
125
+ - Creator/performer robert_smigel
126
+
127
+ - Host platform conan_obrien'
128
+ - source_sentence: Who are the hosts of The Conan & Jordan Show?
129
+ sentences:
130
+ - 'Awards and Recognitions:
131
 
132
+ - 7 Primetime Emmy nominations for writing on Conan''s shows
 
133
 
134
+ - 10 WGA Award nominations (with 2 wins)
135
 
136
+ - 2 Daytime Emmy nominations for Animated Program performance
137
+
138
+
139
+ Major Events:
140
+
141
+ - 1993 Late Night Debut – Joined Conan''s first show as sidekick.
142
+
143
+ - 2000 Departure – Left ''Late Night'' to pursue acting.
144
+
145
+ - 2010 Tour & TBS Move – Reunited with Conan on the live tour and TBS.'
146
+ - 'Graph: Team Coco Knowledge Graph
147
+
148
+ Node ID: the_conan_and_jordan_show
149
+
150
+ Category: shows
151
+
152
+ Name: The Conan & Jordan Show (radio program)
153
+
154
+ Type: Show
155
+
156
+
157
+ Description: A spin-off audio series on SiriusXM''s Team Coco Radio, launched
158
+ in 2023, featuring Conan O''Brien and Jordan Schlansky continuing their comedic
159
+ odd-couple dynamic.'
160
+ - 'Major Events:
161
+
162
+ - 2010 Premiere – ''Conan'' debuted on TBS.
163
+
164
+ - 2015 ''Conan Without Borders'' – International travel specials aired.
165
+
166
+ - 2021 Finale – Conan ended his TBS run.
167
+
168
+
169
+ References:
170
+
171
+ - ([Conan O''Brien - Wikipedia](https://en.wikipedia.org/wiki/Conan_O%27Brien))'
172
  pipeline_tag: sentence-similarity
173
  library_name: sentence-transformers
174
  metrics:
 
204
  value: 0.8611111111111112
205
  name: Cosine Accuracy@3
206
  - type: cosine_accuracy@5
207
+ value: 0.9166666666666666
208
  name: Cosine Accuracy@5
209
  - type: cosine_accuracy@10
210
+ value: 0.9444444444444444
211
  name: Cosine Accuracy@10
212
  - type: cosine_precision@1
213
  value: 0.7222222222222222
 
216
  value: 0.2870370370370371
217
  name: Cosine Precision@3
218
  - type: cosine_precision@5
219
+ value: 0.18333333333333338
220
  name: Cosine Precision@5
221
  - type: cosine_precision@10
222
+ value: 0.09444444444444446
223
  name: Cosine Precision@10
224
  - type: cosine_recall@1
225
  value: 0.7222222222222222
 
228
  value: 0.8611111111111112
229
  name: Cosine Recall@3
230
  - type: cosine_recall@5
231
+ value: 0.9166666666666666
232
  name: Cosine Recall@5
233
  - type: cosine_recall@10
234
+ value: 0.9444444444444444
235
  name: Cosine Recall@10
236
  - type: cosine_ndcg@10
237
+ value: 0.8363985989991439
238
  name: Cosine Ndcg@10
239
  - type: cosine_mrr@10
240
+ value: 0.800925925925926
241
  name: Cosine Mrr@10
242
  - type: cosine_map@100
243
+ value: 0.8041634291634291
244
  name: Cosine Map@100
245
  - task:
246
  type: information-retrieval
 
253
  value: 0.6944444444444444
254
  name: Cosine Accuracy@1
255
  - type: cosine_accuracy@3
256
+ value: 0.8888888888888888
257
  name: Cosine Accuracy@3
258
  - type: cosine_accuracy@5
259
+ value: 0.9166666666666666
260
  name: Cosine Accuracy@5
261
  - type: cosine_accuracy@10
262
+ value: 0.9722222222222222
263
  name: Cosine Accuracy@10
264
  - type: cosine_precision@1
265
  value: 0.6944444444444444
266
  name: Cosine Precision@1
267
  - type: cosine_precision@3
268
+ value: 0.29629629629629634
269
  name: Cosine Precision@3
270
  - type: cosine_precision@5
271
+ value: 0.18333333333333335
272
  name: Cosine Precision@5
273
  - type: cosine_precision@10
274
+ value: 0.09722222222222224
275
  name: Cosine Precision@10
276
  - type: cosine_recall@1
277
  value: 0.6944444444444444
278
  name: Cosine Recall@1
279
  - type: cosine_recall@3
280
+ value: 0.8888888888888888
281
  name: Cosine Recall@3
282
  - type: cosine_recall@5
283
+ value: 0.9166666666666666
284
  name: Cosine Recall@5
285
  - type: cosine_recall@10
286
+ value: 0.9722222222222222
287
  name: Cosine Recall@10
288
  - type: cosine_ndcg@10
289
+ value: 0.8349701465406345
290
  name: Cosine Ndcg@10
291
  - type: cosine_mrr@10
292
+ value: 0.7909722222222222
293
  name: Cosine Mrr@10
294
  - type: cosine_map@100
295
+ value: 0.791703216374269
296
  name: Cosine Map@100
297
  - task:
298
  type: information-retrieval
 
302
  type: dim_256
303
  metrics:
304
  - type: cosine_accuracy@1
305
+ value: 0.6666666666666666
306
  name: Cosine Accuracy@1
307
  - type: cosine_accuracy@3
308
+ value: 0.8611111111111112
309
  name: Cosine Accuracy@3
310
  - type: cosine_accuracy@5
311
+ value: 0.9166666666666666
312
  name: Cosine Accuracy@5
313
  - type: cosine_accuracy@10
314
+ value: 0.9444444444444444
315
  name: Cosine Accuracy@10
316
  - type: cosine_precision@1
317
+ value: 0.6666666666666666
318
  name: Cosine Precision@1
319
  - type: cosine_precision@3
320
+ value: 0.28703703703703703
321
  name: Cosine Precision@3
322
  - type: cosine_precision@5
323
+ value: 0.18333333333333335
324
  name: Cosine Precision@5
325
  - type: cosine_precision@10
326
+ value: 0.09444444444444446
327
  name: Cosine Precision@10
328
  - type: cosine_recall@1
329
+ value: 0.6666666666666666
330
  name: Cosine Recall@1
331
  - type: cosine_recall@3
332
+ value: 0.8611111111111112
333
  name: Cosine Recall@3
334
  - type: cosine_recall@5
335
+ value: 0.9166666666666666
336
  name: Cosine Recall@5
337
  - type: cosine_recall@10
338
+ value: 0.9444444444444444
339
  name: Cosine Recall@10
340
  - type: cosine_ndcg@10
341
+ value: 0.8074890903790802
342
  name: Cosine Ndcg@10
343
  - type: cosine_mrr@10
344
+ value: 0.7627314814814814
345
  name: Cosine Mrr@10
346
  - type: cosine_map@100
347
+ value: 0.7662037037037037
348
  name: Cosine Map@100
349
  - task:
350
  type: information-retrieval
 
354
  type: dim_128
355
  metrics:
356
  - type: cosine_accuracy@1
357
+ value: 0.6388888888888888
358
  name: Cosine Accuracy@1
359
  - type: cosine_accuracy@3
360
+ value: 0.8611111111111112
361
  name: Cosine Accuracy@3
362
  - type: cosine_accuracy@5
363
+ value: 0.9166666666666666
364
  name: Cosine Accuracy@5
365
  - type: cosine_accuracy@10
366
+ value: 0.9444444444444444
367
  name: Cosine Accuracy@10
368
  - type: cosine_precision@1
369
+ value: 0.6388888888888888
370
  name: Cosine Precision@1
371
  - type: cosine_precision@3
372
+ value: 0.2870370370370371
373
  name: Cosine Precision@3
374
  - type: cosine_precision@5
375
+ value: 0.18333333333333338
376
  name: Cosine Precision@5
377
  - type: cosine_precision@10
378
+ value: 0.09444444444444446
379
  name: Cosine Precision@10
380
  - type: cosine_recall@1
381
+ value: 0.6388888888888888
382
  name: Cosine Recall@1
383
  - type: cosine_recall@3
384
+ value: 0.8611111111111112
385
  name: Cosine Recall@3
386
  - type: cosine_recall@5
387
+ value: 0.9166666666666666
388
  name: Cosine Recall@5
389
  - type: cosine_recall@10
390
+ value: 0.9444444444444444
391
  name: Cosine Recall@10
392
  - type: cosine_ndcg@10
393
+ value: 0.803777679552595
394
  name: Cosine Ndcg@10
395
  - type: cosine_mrr@10
396
+ value: 0.7574074074074074
397
  name: Cosine Mrr@10
398
  - type: cosine_map@100
399
+ value: 0.7597654530591711
400
  name: Cosine Map@100
401
  - task:
402
  type: information-retrieval
 
406
  type: dim_64
407
  metrics:
408
  - type: cosine_accuracy@1
409
+ value: 0.6111111111111112
410
  name: Cosine Accuracy@1
411
  - type: cosine_accuracy@3
412
+ value: 0.7777777777777778
413
  name: Cosine Accuracy@3
414
  - type: cosine_accuracy@5
415
+ value: 0.8333333333333334
416
  name: Cosine Accuracy@5
417
  - type: cosine_accuracy@10
418
  value: 0.9166666666666666
419
  name: Cosine Accuracy@10
420
  - type: cosine_precision@1
421
+ value: 0.6111111111111112
422
  name: Cosine Precision@1
423
  - type: cosine_precision@3
424
+ value: 0.2592592592592593
425
  name: Cosine Precision@3
426
  - type: cosine_precision@5
427
+ value: 0.16666666666666669
428
  name: Cosine Precision@5
429
  - type: cosine_precision@10
430
  value: 0.09166666666666669
431
  name: Cosine Precision@10
432
  - type: cosine_recall@1
433
+ value: 0.6111111111111112
434
  name: Cosine Recall@1
435
  - type: cosine_recall@3
436
+ value: 0.7777777777777778
437
  name: Cosine Recall@3
438
  - type: cosine_recall@5
439
+ value: 0.8333333333333334
440
  name: Cosine Recall@5
441
  - type: cosine_recall@10
442
  value: 0.9166666666666666
443
  name: Cosine Recall@10
444
  - type: cosine_ndcg@10
445
+ value: 0.7608354868794361
446
  name: Cosine Ndcg@10
447
  - type: cosine_mrr@10
448
+ value: 0.7111441798941799
449
  name: Cosine Mrr@10
450
  - type: cosine_map@100
451
+ value: 0.7139831037236697
452
  name: Cosine Map@100
453
  ---
454
 
 
502
  model = SentenceTransformer("densonsmith/modernbert-embed-quickb")
503
  # Run inference
504
  sentences = [
505
+ 'Who are the hosts of The Conan & Jordan Show?',
506
+ "Graph: Team Coco Knowledge Graph\nNode ID: the_conan_and_jordan_show\nCategory: shows\nName: The Conan & Jordan Show (radio program)\nType: Show\n\nDescription: A spin-off audio series on SiriusXM's Team Coco Radio, launched in 2023, featuring Conan O'Brien and Jordan Schlansky continuing their comedic odd-couple dynamic.",
507
+ "Awards and Recognitions:\n- 7 Primetime Emmy nominations for writing on Conan's shows\n- 10 WGA Award nominations (with 2 wins)\n- 2 Daytime Emmy nominations for Animated Program performance\n\nMajor Events:\n- 1993 Late Night Debut Joined Conan's first show as sidekick.\n- 2000 Departure Left 'Late Night' to pursue acting.\n- 2010 Tour & TBS Move Reunited with Conan on the live tour and TBS.",
508
  ]
509
  embeddings = model.encode(sentences)
510
  print(embeddings.shape)
 
551
 
552
  | Metric | dim_768 | dim_512 | dim_256 | dim_128 | dim_64 |
553
  |:--------------------|:-----------|:----------|:-----------|:-----------|:-----------|
554
+ | cosine_accuracy@1 | 0.7222 | 0.6944 | 0.6667 | 0.6389 | 0.6111 |
555
+ | cosine_accuracy@3 | 0.8611 | 0.8889 | 0.8611 | 0.8611 | 0.7778 |
556
+ | cosine_accuracy@5 | 0.9167 | 0.9167 | 0.9167 | 0.9167 | 0.8333 |
557
+ | cosine_accuracy@10 | 0.9444 | 0.9722 | 0.9444 | 0.9444 | 0.9167 |
558
+ | cosine_precision@1 | 0.7222 | 0.6944 | 0.6667 | 0.6389 | 0.6111 |
559
+ | cosine_precision@3 | 0.287 | 0.2963 | 0.287 | 0.287 | 0.2593 |
560
+ | cosine_precision@5 | 0.1833 | 0.1833 | 0.1833 | 0.1833 | 0.1667 |
561
+ | cosine_precision@10 | 0.0944 | 0.0972 | 0.0944 | 0.0944 | 0.0917 |
562
+ | cosine_recall@1 | 0.7222 | 0.6944 | 0.6667 | 0.6389 | 0.6111 |
563
+ | cosine_recall@3 | 0.8611 | 0.8889 | 0.8611 | 0.8611 | 0.7778 |
564
+ | cosine_recall@5 | 0.9167 | 0.9167 | 0.9167 | 0.9167 | 0.8333 |
565
+ | cosine_recall@10 | 0.9444 | 0.9722 | 0.9444 | 0.9444 | 0.9167 |
566
+ | **cosine_ndcg@10** | **0.8364** | **0.835** | **0.8075** | **0.8038** | **0.7608** |
567
+ | cosine_mrr@10 | 0.8009 | 0.791 | 0.7627 | 0.7574 | 0.7111 |
568
+ | cosine_map@100 | 0.8042 | 0.7917 | 0.7662 | 0.7598 | 0.714 |
569
 
570
  <!--
571
  ## Bias, Risks and Limitations
 
585
 
586
  #### Unnamed Dataset
587
 
588
+ * Size: 321 training samples
589
  * Columns: <code>anchor</code> and <code>positive</code>
590
+ * Approximate statistics based on the first 321 samples:
591
+ | | anchor | positive |
592
+ |:--------|:----------------------------------------------------------------------------------|:------------------------------------------------------------------------------------|
593
+ | type | string | string |
594
+ | details | <ul><li>min: 7 tokens</li><li>mean: 14.03 tokens</li><li>max: 24 tokens</li></ul> | <ul><li>min: 15 tokens</li><li>mean: 74.79 tokens</li><li>max: 117 tokens</li></ul> |
595
  * Samples:
596
+ | anchor | positive |
597
+ |:--------------------------------------------------------------------------------------------|:--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
598
+ | <code>What brand did Jeff Ross help establish?</code> | <code>Graph: Team Coco Knowledge Graph<br>Node ID: jeff_ross_producer<br>Category: people<br>Name: Jeff Ross (Producer)<br>Type: Person<br><br>Description: Jeff Ross is a television producer who has served as Conan O'Brien's executive producer since 1993. He is a key business partner in Conan's media ventures and helped establish the Team Coco brand.</code> |
599
+ | <code>In what year did Conan O'Brien launch the travel show 'Conan O'Brien Must Go'?</code> | <code>Description: Conan O'Brien is an American television host, comedian, writer, actor, and producer, best known for hosting late-night shows including "Late Night with Conan O'Brien", "The Tonight Show with Conan O'Brien", and "Conan". He also hosts the podcast "Conan O'Brien Needs a Friend" and, in 2024, launched the travel show "Conan O'Brien Must Go" on Max.</code> |
600
+ | <code>What is the strength of the network TBS?</code> | <code>- Network tbs (Strength: parent)<br> Description: TBS provided the platform for the show.</code> |
601
  * Loss: [<code>MatryoshkaLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#matryoshkaloss) with these parameters:
602
  ```json
603
  {
 
760
  ### Training Logs
761
  | Epoch | Step | Training Loss | dim_768_cosine_ndcg@10 | dim_512_cosine_ndcg@10 | dim_256_cosine_ndcg@10 | dim_128_cosine_ndcg@10 | dim_64_cosine_ndcg@10 |
762
  |:-------:|:------:|:-------------:|:----------------------:|:----------------------:|:----------------------:|:----------------------:|:---------------------:|
763
+ | 1.0 | 6 | - | 0.7909 | 0.8034 | 0.7711 | 0.7992 | 0.6908 |
764
+ | 1.7901 | 10 | 16.3044 | - | - | - | - | - |
765
+ | **2.0** | **12** | **-** | **0.8364** | **0.8294** | **0.8022** | **0.8038** | **0.7691** |
766
+ | 3.0 | 18 | - | 0.8364 | 0.8313 | 0.8059 | 0.7938 | 0.7599 |
767
+ | 3.3951 | 20 | 5.6348 | 0.8364 | 0.8350 | 0.8075 | 0.8038 | 0.7608 |
768
 
769
  * The bold row denotes the saved checkpoint.
770
 
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:30007d809f0dc784f184bff2181f312b34b89a55bb3f690aaccf00c3aaa6bb3f
3
  size 596070136
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a0e8097896bbf1d21fed4e34fc84e2ba906a3dd91944327abe7cfb599b5adc63
3
  size 596070136