philipp-zettl commited on
Commit
c1cf91c
·
verified ·
1 Parent(s): 807149d

End of training script.

Browse files
Files changed (2) hide show
  1. README.md +23 -129
  2. model.safetensors +1 -1
README.md CHANGED
@@ -5,59 +5,11 @@ tags:
5
  - feature-extraction
6
  - dense
7
  - generated_from_trainer
8
- - dataset_size:294252
9
  - loss:ContrastiveLoss
10
  base_model: sentence-transformers/clip-ViT-B-32
11
- widget:
12
- - source_sentence: Ranger's Guile
13
- sentences:
14
- - >-
15
- https://cards.scryfall.io/normal/front/0/1/01388e24-d66f-40a8-af41-0f84dd8ed3fe.jpg?1562781768
16
- - >-
17
- https://cards.scryfall.io/normal/front/2/0/20995638-e217-49f1-bccc-d1a344694fd8.jpg?1592711262
18
- - >-
19
- https://cards.scryfall.io/normal/front/1/3/13680953-cf05-4e38-a3cf-22900c02fab7.jpg?1706240764
20
- - source_sentence: |-
21
- {2}, {T}: Choose one that hasn't been chosen —
22
- • This artifact deals 2 damage to target creature.
23
- • Tap target creature.
24
- • Sacrifice this artifact. You gain 3 life.
25
- sentences:
26
- - Haunted Ridge
27
- - >-
28
- https://cards.scryfall.io/normal/front/1/e/1ebb7ade-ebbb-49fb-b64f-fa247f9a9af6.jpg?1562544721
29
- - Three Bowls of Porridge
30
- - source_sentence: Soratami Mindsweeper
31
- sentences:
32
- - >-
33
- https://cards.scryfall.io/normal/front/c/1/c13eacff-0fd1-4e2e-8213-5bf5fff973c8.jpg?1593095724
34
- - >-
35
- https://cards.scryfall.io/normal/front/6/7/67f55ef7-8479-4d56-9801-cb4fd5526e73.jpg?1675644776
36
- - >-
37
- https://cards.scryfall.io/normal/front/0/e/0ef3b83a-777c-45b6-97a1-25be237df79b.jpg?1561932561
38
- - source_sentence: |-
39
- Flying
40
- Ward—Sacrifice a creature.
41
- Whenever a creature dies, target opponent loses 2 life and you gain 2 life.
42
- sentences:
43
- - >-
44
- https://cards.scryfall.io/normal/front/e/4/e4c7919a-de62-46df-a937-593647473ef5.jpg?1654567436
45
- - >-
46
- https://cards.scryfall.io/normal/front/9/7/9712ecaa-4059-44ba-98b7-07bfe7411b5b.jpg?1562447340
47
- - >-
48
- https://cards.scryfall.io/normal/front/0/7/078933b3-6d82-45f2-94e8-addf54cf1704.jpg?1706241798
49
- - source_sentence: Glimpse of Freedom
50
- sentences:
51
- - >-
52
- https://cards.scryfall.io/normal/front/c/1/c1e9c025-1cdb-4da8-8d28-14ad5efb512d.jpg?1581479361
53
- - >-
54
- https://cards.scryfall.io/normal/front/0/7/0770eb6a-4f01-4677-a401-14c1b30692c9.jpg?1673306796
55
- - >-
56
- https://cards.scryfall.io/normal/front/3/3/333039ef-b328-4082-85ae-162caed5612e.jpg?1738358111
57
  pipeline_tag: sentence-similarity
58
  library_name: sentence-transformers
59
- datasets:
60
- - philipp-zettl/mtg_cards-2025-04-04
61
  ---
62
 
63
  # SentenceTransformer based on sentence-transformers/clip-ViT-B-32
@@ -108,9 +60,9 @@ from sentence_transformers import SentenceTransformer
108
  model = SentenceTransformer("philipp-zettl/MTGEmb-small")
109
  # Run inference
110
  sentences = [
111
- 'Glimpse of Freedom',
112
- 'https://cards.scryfall.io/normal/front/c/1/c1e9c025-1cdb-4da8-8d28-14ad5efb512d.jpg?1581479361',
113
- 'https://cards.scryfall.io/normal/front/3/3/333039ef-b328-4082-85ae-162caed5612e.jpg?1738358111',
114
  ]
115
  embeddings = model.encode(sentences)
116
  print(embeddings.shape)
@@ -119,9 +71,9 @@ print(embeddings.shape)
119
  # Get the similarity scores for the embeddings
120
  similarities = model.similarity(embeddings, embeddings)
121
  print(similarities)
122
- # tensor([[1.0000, 0.7515, 0.7515],
123
- # [0.7515, 1.0000, 1.0000],
124
- # [0.7515, 1.0000, 1.0000]])
125
  ```
126
 
127
  <!--
@@ -166,19 +118,19 @@ You can finetune this model on your own dataset.
166
 
167
  #### Unnamed Dataset
168
 
169
- * Size: 294,252 training samples
170
  * Columns: <code>sentence_0</code>, <code>sentence_1</code>, and <code>label</code>
171
- * Approximate statistics based on the first 1000 samples:
172
- | | sentence_0 | sentence_1 | label |
173
- |:--------|:----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:------------------------------------------------|
174
- | type | string | string | int |
175
- | details | <ul><li>min: 3 tokens</li><li>mean: 25.44 tokens</li><li>max: 77 tokens</li></ul> | <ul><li>min: 3 tokens</li><li>mean: 54.71 tokens</li><li>max: 69 tokens</li></ul> | <ul><li>0: ~38.90%</li><li>1: ~61.10%</li></ul> |
176
  * Samples:
177
- | sentence_0 | sentence_1 | label |
178
- |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:------------------------------------------------------------------------------------------------------------|:---------------|
179
- | <code>Flying<br>Ascend (If you control ten or more permanents, you get the city's blessing for the rest of the game.)<br>At the beginning of your upkeep, if you have the city's blessing, reveal the top card of your library and put it into your hand. Each opponent loses X life and you gain X life, where X is that card's mana value.</code> | <code>Twilight Prophet</code> | <code>1</code> |
180
- | <code>Target creature gains deathtouch until end of turn and must be blocked this turn if able.<br>Flashback {G} (You may cast this card from your graveyard for its flashback cost. Then exile it.)</code> | <code>https://cards.scryfall.io/normal/front/f/2/f2c97632-3cf1-4b79-9d18-d8991654dcca.jpg?1562796910</code> | <code>0</code> |
181
- | <code>Pay 1 life, Sacrifice another creature: Create a Treasure token. (It's an artifact with "{T}, Sacrifice this token: Add one mana of any color.")</code> | <code>https://cards.scryfall.io/normal/front/7/7/77c0929f-bee6-416d-8571-6540ae4f6e4f.jpg?1562879392</code> | <code>0</code> |
182
  * Loss: [<code>ContrastiveLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#contrastiveloss) with these parameters:
183
  ```json
184
  {
@@ -191,9 +143,8 @@ You can finetune this model on your own dataset.
191
  ### Training Hyperparameters
192
  #### Non-Default Hyperparameters
193
 
194
- - `per_device_train_batch_size`: 56
195
- - `per_device_eval_batch_size`: 56
196
- - `num_train_epochs`: 5
197
  - `multi_dataset_batch_sampler`: round_robin
198
 
199
  #### All Hyperparameters
@@ -203,8 +154,8 @@ You can finetune this model on your own dataset.
203
  - `do_predict`: False
204
  - `eval_strategy`: no
205
  - `prediction_loss_only`: True
206
- - `per_device_train_batch_size`: 56
207
- - `per_device_eval_batch_size`: 56
208
  - `per_gpu_train_batch_size`: None
209
  - `per_gpu_eval_batch_size`: None
210
  - `gradient_accumulation_steps`: 1
@@ -216,7 +167,7 @@ You can finetune this model on your own dataset.
216
  - `adam_beta2`: 0.999
217
  - `adam_epsilon`: 1e-08
218
  - `max_grad_norm`: 1
219
- - `num_train_epochs`: 5
220
  - `max_steps`: -1
221
  - `lr_scheduler_type`: linear
222
  - `lr_scheduler_kwargs`: {}
@@ -319,63 +270,6 @@ You can finetune this model on your own dataset.
319
 
320
  </details>
321
 
322
- ### Training Logs
323
- | Epoch | Step | Training Loss |
324
- |:------:|:-----:|:-------------:|
325
- | 0.0951 | 500 | 0.0269 |
326
- | 0.1903 | 1000 | 0.0253 |
327
- | 0.2854 | 1500 | 0.0253 |
328
- | 0.3806 | 2000 | 0.0252 |
329
- | 0.4757 | 2500 | 0.0252 |
330
- | 0.5709 | 3000 | 0.0253 |
331
- | 0.6660 | 3500 | 0.0252 |
332
- | 0.7612 | 4000 | 0.0251 |
333
- | 0.8563 | 4500 | 0.0251 |
334
- | 0.9515 | 5000 | 0.0251 |
335
- | 1.0466 | 5500 | 0.0251 |
336
- | 1.1418 | 6000 | 0.0251 |
337
- | 1.2369 | 6500 | 0.0252 |
338
- | 1.3321 | 7000 | 0.0251 |
339
- | 1.4272 | 7500 | 0.0252 |
340
- | 1.5224 | 8000 | 0.0252 |
341
- | 1.6175 | 8500 | 0.0251 |
342
- | 1.7127 | 9000 | 0.0249 |
343
- | 1.8078 | 9500 | 0.0252 |
344
- | 1.9029 | 10000 | 0.025 |
345
- | 1.9981 | 10500 | 0.0252 |
346
- | 2.0932 | 11000 | 0.0251 |
347
- | 2.1884 | 11500 | 0.0251 |
348
- | 2.2835 | 12000 | 0.0252 |
349
- | 2.3787 | 12500 | 0.0252 |
350
- | 2.4738 | 13000 | 0.0251 |
351
- | 2.5690 | 13500 | 0.025 |
352
- | 2.6641 | 14000 | 0.025 |
353
- | 2.7593 | 14500 | 0.0251 |
354
- | 2.8544 | 15000 | 0.0251 |
355
- | 2.9496 | 15500 | 0.0251 |
356
- | 3.0447 | 16000 | 0.0251 |
357
- | 3.1399 | 16500 | 0.0252 |
358
- | 3.2350 | 17000 | 0.0251 |
359
- | 3.3302 | 17500 | 0.025 |
360
- | 3.4253 | 18000 | 0.0252 |
361
- | 3.5205 | 18500 | 0.0251 |
362
- | 3.6156 | 19000 | 0.025 |
363
- | 3.7108 | 19500 | 0.0251 |
364
- | 3.8059 | 20000 | 0.0252 |
365
- | 3.9010 | 20500 | 0.025 |
366
- | 3.9962 | 21000 | 0.025 |
367
- | 4.0913 | 21500 | 0.0251 |
368
- | 4.1865 | 22000 | 0.0251 |
369
- | 4.2816 | 22500 | 0.025 |
370
- | 4.3768 | 23000 | 0.025 |
371
- | 4.4719 | 23500 | 0.025 |
372
- | 4.5671 | 24000 | 0.0252 |
373
- | 4.6622 | 24500 | 0.025 |
374
- | 4.7574 | 25000 | 0.0252 |
375
- | 4.8525 | 25500 | 0.025 |
376
- | 4.9477 | 26000 | 0.025 |
377
-
378
-
379
  ### Framework Versions
380
  - Python: 3.13.7
381
  - Sentence Transformers: 5.1.2
 
5
  - feature-extraction
6
  - dense
7
  - generated_from_trainer
8
+ - dataset_size:5
9
  - loss:ContrastiveLoss
10
  base_model: sentence-transformers/clip-ViT-B-32
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11
  pipeline_tag: sentence-similarity
12
  library_name: sentence-transformers
 
 
13
  ---
14
 
15
  # SentenceTransformer based on sentence-transformers/clip-ViT-B-32
 
60
  model = SentenceTransformer("philipp-zettl/MTGEmb-small")
61
  # Run inference
62
  sentences = [
63
+ 'The weather is lovely today.',
64
+ "It's so sunny outside!",
65
+ 'He drove to the stadium.',
66
  ]
67
  embeddings = model.encode(sentences)
68
  print(embeddings.shape)
 
71
  # Get the similarity scores for the embeddings
72
  similarities = model.similarity(embeddings, embeddings)
73
  print(similarities)
74
+ # tensor([[1.0000, 0.9425, 0.8177],
75
+ # [0.9425, 1.0000, 0.8015],
76
+ # [0.8177, 0.8015, 1.0000]])
77
  ```
78
 
79
  <!--
 
118
 
119
  #### Unnamed Dataset
120
 
121
+ * Size: 5 training samples
122
  * Columns: <code>sentence_0</code>, <code>sentence_1</code>, and <code>label</code>
123
+ * Approximate statistics based on the first 5 samples:
124
+ | | sentence_0 | sentence_1 | label |
125
+ |:--------|:---------------------------------------------------------------------------------|:---------------------------------------------------------------------------------|:------------------------------------------------|
126
+ | type | string | string | int |
127
+ | details | <ul><li>min: 3 tokens</li><li>mean: 16.8 tokens</li><li>max: 66 tokens</li></ul> | <ul><li>min: 6 tokens</li><li>mean: 53.6 tokens</li><li>max: 66 tokens</li></ul> | <ul><li>0: ~60.00%</li><li>1: ~40.00%</li></ul> |
128
  * Samples:
129
+ | sentence_0 | sentence_1 | label |
130
+ |:-----------------------------|:------------------------------------------------------------------------------------------------------------|:---------------|
131
+ | <code>Veteran Armorer</code> | <code>https://cards.scryfall.io/normal/front/0/0/0000419b-0bba-4488-8f7a-6194544ce91e.jpg?1721427487</code> | <code>0</code> |
132
+ | <code>Forest</code> | <code>https://cards.scryfall.io/normal/front/0/0/0000419b-0bba-4488-8f7a-6194544ce91e.jpg?1721427487</code> | <code>1</code> |
133
+ | <code>Forest</code> | <code>Veteran Armorer</code> | <code>0</code> |
134
  * Loss: [<code>ContrastiveLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#contrastiveloss) with these parameters:
135
  ```json
136
  {
 
143
  ### Training Hyperparameters
144
  #### Non-Default Hyperparameters
145
 
146
+ - `per_device_train_batch_size`: 64
147
+ - `per_device_eval_batch_size`: 64
 
148
  - `multi_dataset_batch_sampler`: round_robin
149
 
150
  #### All Hyperparameters
 
154
  - `do_predict`: False
155
  - `eval_strategy`: no
156
  - `prediction_loss_only`: True
157
+ - `per_device_train_batch_size`: 64
158
+ - `per_device_eval_batch_size`: 64
159
  - `per_gpu_train_batch_size`: None
160
  - `per_gpu_eval_batch_size`: None
161
  - `gradient_accumulation_steps`: 1
 
167
  - `adam_beta2`: 0.999
168
  - `adam_epsilon`: 1e-08
169
  - `max_grad_norm`: 1
170
+ - `num_train_epochs`: 3
171
  - `max_steps`: -1
172
  - `lr_scheduler_type`: linear
173
  - `lr_scheduler_kwargs`: {}
 
270
 
271
  </details>
272
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
273
  ### Framework Versions
274
  - Python: 3.13.7
275
  - Sentence Transformers: 5.1.2
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:341fc80e886d58cc50059de8b9b063bd1b9a39dc246dd10f13db826617cc9d66
3
  size 605156676
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3b41f0699e0483f5792e303ee4230710736646a005a402e19017010acb4607ad
3
  size 605156676