AdaptiveLayerLoss(model=model,
Browse filesloss=train_loss,
n_layers_per_step = 1,
last_layer_weight = 1,
prior_layers_weight= 1,
kl_div_weight = 1,
kl_temperature= 1,
)''')
lr = 1e-6. batch = 42, schedule = cosine
- README.md +325 -109
- pytorch_model.bin +1 -1
README.md
CHANGED
@@ -7,7 +7,7 @@ tags:
|
|
7 |
- sentence-similarity
|
8 |
- feature-extraction
|
9 |
- generated_from_trainer
|
10 |
-
- dataset_size:
|
11 |
- loss:AdaptiveLayerLoss
|
12 |
- loss:MultipleNegativesRankingLoss
|
13 |
base_model: microsoft/deberta-v3-small
|
@@ -49,34 +49,44 @@ metrics:
|
|
49 |
- max_precision
|
50 |
- max_recall
|
51 |
- max_ap
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
52 |
widget:
|
53 |
-
- source_sentence: A
|
54 |
sentences:
|
55 |
-
-
|
56 |
-
-
|
57 |
-
-
|
58 |
-
- source_sentence:
|
|
|
59 |
sentences:
|
60 |
-
-
|
61 |
-
-
|
62 |
-
-
|
63 |
-
- source_sentence:
|
|
|
64 |
sentences:
|
65 |
-
-
|
66 |
-
-
|
67 |
-
-
|
68 |
-
- source_sentence:
|
69 |
sentences:
|
70 |
-
-
|
71 |
-
-
|
72 |
-
|
73 |
-
|
74 |
-
- source_sentence: This may be overly obvious, but in American English, saying "you're
|
75 |
-
welcome" is certainly polite and standard.
|
76 |
sentences:
|
77 |
-
-
|
78 |
-
-
|
79 |
-
-
|
80 |
pipeline_tag: sentence-similarity
|
81 |
model-index:
|
82 |
- name: SentenceTransformer based on microsoft/deberta-v3-small
|
@@ -89,110 +99,147 @@ model-index:
|
|
89 |
type: unknown
|
90 |
metrics:
|
91 |
- type: cosine_accuracy
|
92 |
-
value: 0.
|
93 |
name: Cosine Accuracy
|
94 |
- type: cosine_accuracy_threshold
|
95 |
-
value: 0.
|
96 |
name: Cosine Accuracy Threshold
|
97 |
- type: cosine_f1
|
98 |
-
value: 0.
|
99 |
name: Cosine F1
|
100 |
- type: cosine_f1_threshold
|
101 |
-
value: 0.
|
102 |
name: Cosine F1 Threshold
|
103 |
- type: cosine_precision
|
104 |
-
value: 0.
|
105 |
name: Cosine Precision
|
106 |
- type: cosine_recall
|
107 |
-
value: 0.
|
108 |
name: Cosine Recall
|
109 |
- type: cosine_ap
|
110 |
-
value: 0.
|
111 |
name: Cosine Ap
|
112 |
- type: dot_accuracy
|
113 |
-
value: 0.
|
114 |
name: Dot Accuracy
|
115 |
- type: dot_accuracy_threshold
|
116 |
-
value:
|
117 |
name: Dot Accuracy Threshold
|
118 |
- type: dot_f1
|
119 |
-
value: 0.
|
120 |
name: Dot F1
|
121 |
- type: dot_f1_threshold
|
122 |
-
value:
|
123 |
name: Dot F1 Threshold
|
124 |
- type: dot_precision
|
125 |
-
value: 0.
|
126 |
name: Dot Precision
|
127 |
- type: dot_recall
|
128 |
-
value: 0.
|
129 |
name: Dot Recall
|
130 |
- type: dot_ap
|
131 |
-
value: 0.
|
132 |
name: Dot Ap
|
133 |
- type: manhattan_accuracy
|
134 |
-
value: 0.
|
135 |
name: Manhattan Accuracy
|
136 |
- type: manhattan_accuracy_threshold
|
137 |
-
value:
|
138 |
name: Manhattan Accuracy Threshold
|
139 |
- type: manhattan_f1
|
140 |
-
value: 0.
|
141 |
name: Manhattan F1
|
142 |
- type: manhattan_f1_threshold
|
143 |
-
value:
|
144 |
name: Manhattan F1 Threshold
|
145 |
- type: manhattan_precision
|
146 |
-
value: 0.
|
147 |
name: Manhattan Precision
|
148 |
- type: manhattan_recall
|
149 |
-
value: 0.
|
150 |
name: Manhattan Recall
|
151 |
- type: manhattan_ap
|
152 |
-
value: 0.
|
153 |
name: Manhattan Ap
|
154 |
- type: euclidean_accuracy
|
155 |
-
value: 0.
|
156 |
name: Euclidean Accuracy
|
157 |
- type: euclidean_accuracy_threshold
|
158 |
-
value:
|
159 |
name: Euclidean Accuracy Threshold
|
160 |
- type: euclidean_f1
|
161 |
-
value: 0.
|
162 |
name: Euclidean F1
|
163 |
- type: euclidean_f1_threshold
|
164 |
-
value:
|
165 |
name: Euclidean F1 Threshold
|
166 |
- type: euclidean_precision
|
167 |
-
value: 0.
|
168 |
name: Euclidean Precision
|
169 |
- type: euclidean_recall
|
170 |
-
value: 0.
|
171 |
name: Euclidean Recall
|
172 |
- type: euclidean_ap
|
173 |
-
value: 0.
|
174 |
name: Euclidean Ap
|
175 |
- type: max_accuracy
|
176 |
-
value: 0.
|
177 |
name: Max Accuracy
|
178 |
- type: max_accuracy_threshold
|
179 |
-
value:
|
180 |
name: Max Accuracy Threshold
|
181 |
- type: max_f1
|
182 |
-
value: 0.
|
183 |
name: Max F1
|
184 |
- type: max_f1_threshold
|
185 |
-
value:
|
186 |
name: Max F1 Threshold
|
187 |
- type: max_precision
|
188 |
-
value: 0.
|
189 |
name: Max Precision
|
190 |
- type: max_recall
|
191 |
-
value: 0.
|
192 |
name: Max Recall
|
193 |
- type: max_ap
|
194 |
-
value: 0.
|
195 |
name: Max Ap
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
196 |
---
|
197 |
|
198 |
# SentenceTransformer based on microsoft/deberta-v3-small
|
@@ -245,9 +292,9 @@ from sentence_transformers import SentenceTransformer
|
|
245 |
model = SentenceTransformer("bobox/DeBERTaV3-small-ST-AdaptiveLayers-ep2")
|
246 |
# Run inference
|
247 |
sentences = [
|
248 |
-
'
|
249 |
-
'
|
250 |
-
|
251 |
]
|
252 |
embeddings = model.encode(sentences)
|
253 |
print(embeddings.shape)
|
@@ -293,41 +340,58 @@ You can finetune this model on your own dataset.
|
|
293 |
|
294 |
| Metric | Value |
|
295 |
|:-----------------------------|:-----------|
|
296 |
-
| cosine_accuracy | 0.
|
297 |
-
| cosine_accuracy_threshold | 0.
|
298 |
-
| cosine_f1 | 0.
|
299 |
-
| cosine_f1_threshold | 0.
|
300 |
-
| cosine_precision | 0.
|
301 |
-
| cosine_recall | 0.
|
302 |
-
| cosine_ap | 0.
|
303 |
-
| dot_accuracy | 0.
|
304 |
-
| dot_accuracy_threshold |
|
305 |
-
| dot_f1 | 0.
|
306 |
-
| dot_f1_threshold |
|
307 |
-
| dot_precision | 0.
|
308 |
-
| dot_recall | 0.
|
309 |
-
| dot_ap | 0.
|
310 |
-
| manhattan_accuracy | 0.
|
311 |
-
| manhattan_accuracy_threshold |
|
312 |
-
| manhattan_f1 | 0.
|
313 |
-
| manhattan_f1_threshold |
|
314 |
-
| manhattan_precision | 0.
|
315 |
-
| manhattan_recall | 0.
|
316 |
-
| manhattan_ap | 0.
|
317 |
-
| euclidean_accuracy | 0.
|
318 |
-
| euclidean_accuracy_threshold |
|
319 |
-
| euclidean_f1 | 0.
|
320 |
-
| euclidean_f1_threshold |
|
321 |
-
| euclidean_precision | 0.
|
322 |
-
| euclidean_recall | 0.
|
323 |
-
| euclidean_ap | 0.
|
324 |
-
| max_accuracy | 0.
|
325 |
-
| max_accuracy_threshold |
|
326 |
-
| max_f1 | 0.
|
327 |
-
| max_f1_threshold |
|
328 |
-
| max_precision | 0.
|
329 |
-
| max_recall | 0.
|
330 |
-
| **max_ap** | **0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
331 |
|
332 |
<!--
|
333 |
## Bias, Risks and Limitations
|
@@ -348,19 +412,19 @@ You can finetune this model on your own dataset.
|
|
348 |
#### stanfordnlp/snli
|
349 |
|
350 |
* Dataset: [stanfordnlp/snli](https://huggingface.co/datasets/stanfordnlp/snli) at [cdb5c3d](https://huggingface.co/datasets/stanfordnlp/snli/tree/cdb5c3d5eed6ead6e5a341c8e56e669bb666725b)
|
351 |
-
* Size:
|
352 |
* Columns: <code>sentence1</code>, <code>sentence2</code>, and <code>label</code>
|
353 |
* Approximate statistics based on the first 1000 samples:
|
354 |
-
| | sentence1
|
355 |
-
|
356 |
-
| type | string
|
357 |
-
| details | <ul><li>min:
|
358 |
* Samples:
|
359 |
-
| sentence1
|
360 |
-
|
361 |
-
| <code>
|
362 |
-
| <code>
|
363 |
-
| <code>
|
364 |
* Loss: [<code>AdaptiveLayerLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#adaptivelayerloss) with these parameters:
|
365 |
```json
|
366 |
{
|
@@ -403,10 +467,162 @@ You can finetune this model on your own dataset.
|
|
403 |
}
|
404 |
```
|
405 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
406 |
### Training Logs
|
407 |
-
| Epoch | Step | loss | max_ap |
|
408 |
-
|
409 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
410 |
|
411 |
|
412 |
### Framework Versions
|
|
|
7 |
- sentence-similarity
|
8 |
- feature-extraction
|
9 |
- generated_from_trainer
|
10 |
+
- dataset_size:67190
|
11 |
- loss:AdaptiveLayerLoss
|
12 |
- loss:MultipleNegativesRankingLoss
|
13 |
base_model: microsoft/deberta-v3-small
|
|
|
49 |
- max_precision
|
50 |
- max_recall
|
51 |
- max_ap
|
52 |
+
- pearson_cosine
|
53 |
+
- spearman_cosine
|
54 |
+
- pearson_manhattan
|
55 |
+
- spearman_manhattan
|
56 |
+
- pearson_euclidean
|
57 |
+
- spearman_euclidean
|
58 |
+
- pearson_dot
|
59 |
+
- spearman_dot
|
60 |
+
- pearson_max
|
61 |
+
- spearman_max
|
62 |
widget:
|
63 |
+
- source_sentence: A worker peers out from atop a building under construction.
|
64 |
sentences:
|
65 |
+
- The man pleads for mercy.
|
66 |
+
- People and a baby crossing the street.
|
67 |
+
- A person is atop of a building.
|
68 |
+
- source_sentence: An aisle at Best Buy with an employee standing at the computer
|
69 |
+
and a Geek Squad sign in the background.
|
70 |
sentences:
|
71 |
+
- the man is watching the stars
|
72 |
+
- The employee is wearing a blue shirt.
|
73 |
+
- A person balancing.
|
74 |
+
- source_sentence: A man with a long white beard is examining a camera and another
|
75 |
+
man with a black shirt is in the background.
|
76 |
sentences:
|
77 |
+
- a man is with another man
|
78 |
+
- Children in uniforms climb a tower.
|
79 |
+
- There are five children.
|
80 |
+
- source_sentence: A black dog with a blue collar is jumping into the water.
|
81 |
sentences:
|
82 |
+
- The dog is playing tug of war with a stick.
|
83 |
+
- There is a woman painting.
|
84 |
+
- A black dog wearing a blue collar is chasing something into the water.
|
85 |
+
- source_sentence: A wet child stands in chest deep ocean water.
|
|
|
|
|
86 |
sentences:
|
87 |
+
- A woman paints a portrait of her best friend.
|
88 |
+
- A person in red is cutting the grass on a riding mower
|
89 |
+
- The child s playing on the beach.
|
90 |
pipeline_tag: sentence-similarity
|
91 |
model-index:
|
92 |
- name: SentenceTransformer based on microsoft/deberta-v3-small
|
|
|
99 |
type: unknown
|
100 |
metrics:
|
101 |
- type: cosine_accuracy
|
102 |
+
value: 0.6583157259281618
|
103 |
name: Cosine Accuracy
|
104 |
- type: cosine_accuracy_threshold
|
105 |
+
value: 0.6766541004180908
|
106 |
name: Cosine Accuracy Threshold
|
107 |
- type: cosine_f1
|
108 |
+
value: 0.7049362860324137
|
109 |
name: Cosine F1
|
110 |
- type: cosine_f1_threshold
|
111 |
+
value: 0.6017583012580872
|
112 |
name: Cosine F1 Threshold
|
113 |
- type: cosine_precision
|
114 |
+
value: 0.6115046147241897
|
115 |
name: Cosine Precision
|
116 |
- type: cosine_recall
|
117 |
+
value: 0.8320677570093458
|
118 |
name: Cosine Recall
|
119 |
- type: cosine_ap
|
120 |
+
value: 0.6995030811464378
|
121 |
name: Cosine Ap
|
122 |
- type: dot_accuracy
|
123 |
+
value: 0.6272260790824027
|
124 |
name: Dot Accuracy
|
125 |
- type: dot_accuracy_threshold
|
126 |
+
value: 163.25054931640625
|
127 |
name: Dot Accuracy Threshold
|
128 |
- type: dot_f1
|
129 |
+
value: 0.6976381461675579
|
130 |
name: Dot F1
|
131 |
- type: dot_f1_threshold
|
132 |
+
value: 119.20779418945312
|
133 |
name: Dot F1 Threshold
|
134 |
- type: dot_precision
|
135 |
+
value: 0.5639409221902018
|
136 |
name: Dot Precision
|
137 |
- type: dot_recall
|
138 |
+
value: 0.914427570093458
|
139 |
name: Dot Recall
|
140 |
- type: dot_ap
|
141 |
+
value: 0.643747511442345
|
142 |
name: Dot Ap
|
143 |
- type: manhattan_accuracy
|
144 |
+
value: 0.6571083610021129
|
145 |
name: Manhattan Accuracy
|
146 |
- type: manhattan_accuracy_threshold
|
147 |
+
value: 243.75453186035156
|
148 |
name: Manhattan Accuracy Threshold
|
149 |
- type: manhattan_f1
|
150 |
+
value: 0.7055783910745744
|
151 |
name: Manhattan F1
|
152 |
- type: manhattan_f1_threshold
|
153 |
+
value: 295.95947265625
|
154 |
name: Manhattan F1 Threshold
|
155 |
- type: manhattan_precision
|
156 |
+
value: 0.5900608917697898
|
157 |
name: Manhattan Precision
|
158 |
- type: manhattan_recall
|
159 |
+
value: 0.8773364485981309
|
160 |
name: Manhattan Recall
|
161 |
- type: manhattan_ap
|
162 |
+
value: 0.7072033306346501
|
163 |
name: Manhattan Ap
|
164 |
- type: euclidean_accuracy
|
165 |
+
value: 0.6590703290069424
|
166 |
name: Euclidean Accuracy
|
167 |
- type: euclidean_accuracy_threshold
|
168 |
+
value: 12.141830444335938
|
169 |
name: Euclidean Accuracy Threshold
|
170 |
- type: euclidean_f1
|
171 |
+
value: 0.7036813518406759
|
172 |
name: Euclidean F1
|
173 |
- type: euclidean_f1_threshold
|
174 |
+
value: 14.197540283203125
|
175 |
name: Euclidean F1 Threshold
|
176 |
- type: euclidean_precision
|
177 |
+
value: 0.5996708496194199
|
178 |
name: Euclidean Precision
|
179 |
- type: euclidean_recall
|
180 |
+
value: 0.8513434579439252
|
181 |
name: Euclidean Recall
|
182 |
- type: euclidean_ap
|
183 |
+
value: 0.7035256676322055
|
184 |
name: Euclidean Ap
|
185 |
- type: max_accuracy
|
186 |
+
value: 0.6590703290069424
|
187 |
name: Max Accuracy
|
188 |
- type: max_accuracy_threshold
|
189 |
+
value: 243.75453186035156
|
190 |
name: Max Accuracy Threshold
|
191 |
- type: max_f1
|
192 |
+
value: 0.7055783910745744
|
193 |
name: Max F1
|
194 |
- type: max_f1_threshold
|
195 |
+
value: 295.95947265625
|
196 |
name: Max F1 Threshold
|
197 |
- type: max_precision
|
198 |
+
value: 0.6115046147241897
|
199 |
name: Max Precision
|
200 |
- type: max_recall
|
201 |
+
value: 0.914427570093458
|
202 |
name: Max Recall
|
203 |
- type: max_ap
|
204 |
+
value: 0.7072033306346501
|
205 |
name: Max Ap
|
206 |
+
- task:
|
207 |
+
type: semantic-similarity
|
208 |
+
name: Semantic Similarity
|
209 |
+
dataset:
|
210 |
+
name: Unknown
|
211 |
+
type: unknown
|
212 |
+
metrics:
|
213 |
+
- type: pearson_cosine
|
214 |
+
value: 0.732169941341086
|
215 |
+
name: Pearson Cosine
|
216 |
+
- type: spearman_cosine
|
217 |
+
value: 0.7344587206087978
|
218 |
+
name: Spearman Cosine
|
219 |
+
- type: pearson_manhattan
|
220 |
+
value: 0.7537099624360986
|
221 |
+
name: Pearson Manhattan
|
222 |
+
- type: spearman_manhattan
|
223 |
+
value: 0.7550555196955944
|
224 |
+
name: Spearman Manhattan
|
225 |
+
- type: pearson_euclidean
|
226 |
+
value: 0.7468210439584286
|
227 |
+
name: Pearson Euclidean
|
228 |
+
- type: spearman_euclidean
|
229 |
+
value: 0.74849026008206
|
230 |
+
name: Spearman Euclidean
|
231 |
+
- type: pearson_dot
|
232 |
+
value: 0.6142835401925993
|
233 |
+
name: Pearson Dot
|
234 |
+
- type: spearman_dot
|
235 |
+
value: 0.6100201108417316
|
236 |
+
name: Spearman Dot
|
237 |
+
- type: pearson_max
|
238 |
+
value: 0.7537099624360986
|
239 |
+
name: Pearson Max
|
240 |
+
- type: spearman_max
|
241 |
+
value: 0.7550555196955944
|
242 |
+
name: Spearman Max
|
243 |
---
|
244 |
|
245 |
# SentenceTransformer based on microsoft/deberta-v3-small
|
|
|
292 |
model = SentenceTransformer("bobox/DeBERTaV3-small-ST-AdaptiveLayers-ep2")
|
293 |
# Run inference
|
294 |
sentences = [
|
295 |
+
'A wet child stands in chest deep ocean water.',
|
296 |
+
'The child s playing on the beach.',
|
297 |
+
'A woman paints a portrait of her best friend.',
|
298 |
]
|
299 |
embeddings = model.encode(sentences)
|
300 |
print(embeddings.shape)
|
|
|
340 |
|
341 |
| Metric | Value |
|
342 |
|:-----------------------------|:-----------|
|
343 |
+
| cosine_accuracy | 0.6583 |
|
344 |
+
| cosine_accuracy_threshold | 0.6767 |
|
345 |
+
| cosine_f1 | 0.7049 |
|
346 |
+
| cosine_f1_threshold | 0.6018 |
|
347 |
+
| cosine_precision | 0.6115 |
|
348 |
+
| cosine_recall | 0.8321 |
|
349 |
+
| cosine_ap | 0.6995 |
|
350 |
+
| dot_accuracy | 0.6272 |
|
351 |
+
| dot_accuracy_threshold | 163.2505 |
|
352 |
+
| dot_f1 | 0.6976 |
|
353 |
+
| dot_f1_threshold | 119.2078 |
|
354 |
+
| dot_precision | 0.5639 |
|
355 |
+
| dot_recall | 0.9144 |
|
356 |
+
| dot_ap | 0.6437 |
|
357 |
+
| manhattan_accuracy | 0.6571 |
|
358 |
+
| manhattan_accuracy_threshold | 243.7545 |
|
359 |
+
| manhattan_f1 | 0.7056 |
|
360 |
+
| manhattan_f1_threshold | 295.9595 |
|
361 |
+
| manhattan_precision | 0.5901 |
|
362 |
+
| manhattan_recall | 0.8773 |
|
363 |
+
| manhattan_ap | 0.7072 |
|
364 |
+
| euclidean_accuracy | 0.6591 |
|
365 |
+
| euclidean_accuracy_threshold | 12.1418 |
|
366 |
+
| euclidean_f1 | 0.7037 |
|
367 |
+
| euclidean_f1_threshold | 14.1975 |
|
368 |
+
| euclidean_precision | 0.5997 |
|
369 |
+
| euclidean_recall | 0.8513 |
|
370 |
+
| euclidean_ap | 0.7035 |
|
371 |
+
| max_accuracy | 0.6591 |
|
372 |
+
| max_accuracy_threshold | 243.7545 |
|
373 |
+
| max_f1 | 0.7056 |
|
374 |
+
| max_f1_threshold | 295.9595 |
|
375 |
+
| max_precision | 0.6115 |
|
376 |
+
| max_recall | 0.9144 |
|
377 |
+
| **max_ap** | **0.7072** |
|
378 |
+
|
379 |
+
#### Semantic Similarity
|
380 |
+
|
381 |
+
* Evaluated with [<code>EmbeddingSimilarityEvaluator</code>](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.EmbeddingSimilarityEvaluator)
|
382 |
+
|
383 |
+
| Metric | Value |
|
384 |
+
|:--------------------|:-----------|
|
385 |
+
| pearson_cosine | 0.7322 |
|
386 |
+
| **spearman_cosine** | **0.7345** |
|
387 |
+
| pearson_manhattan | 0.7537 |
|
388 |
+
| spearman_manhattan | 0.7551 |
|
389 |
+
| pearson_euclidean | 0.7468 |
|
390 |
+
| spearman_euclidean | 0.7485 |
|
391 |
+
| pearson_dot | 0.6143 |
|
392 |
+
| spearman_dot | 0.61 |
|
393 |
+
| pearson_max | 0.7537 |
|
394 |
+
| spearman_max | 0.7551 |
|
395 |
|
396 |
<!--
|
397 |
## Bias, Risks and Limitations
|
|
|
412 |
#### stanfordnlp/snli
|
413 |
|
414 |
* Dataset: [stanfordnlp/snli](https://huggingface.co/datasets/stanfordnlp/snli) at [cdb5c3d](https://huggingface.co/datasets/stanfordnlp/snli/tree/cdb5c3d5eed6ead6e5a341c8e56e669bb666725b)
|
415 |
+
* Size: 67,190 training samples
|
416 |
* Columns: <code>sentence1</code>, <code>sentence2</code>, and <code>label</code>
|
417 |
* Approximate statistics based on the first 1000 samples:
|
418 |
+
| | sentence1 | sentence2 | label |
|
419 |
+
|:--------|:-----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:-----------------------------|
|
420 |
+
| type | string | string | int |
|
421 |
+
| details | <ul><li>min: 4 tokens</li><li>mean: 21.19 tokens</li><li>max: 133 tokens</li></ul> | <ul><li>min: 4 tokens</li><li>mean: 11.77 tokens</li><li>max: 49 tokens</li></ul> | <ul><li>0: 100.00%</li></ul> |
|
422 |
* Samples:
|
423 |
+
| sentence1 | sentence2 | label |
|
424 |
+
|:---------------------------------------------------------------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------|:---------------|
|
425 |
+
| <code>Without a placebo group, we still won't know if any of the treatments are better than nothing and therefore worth giving.</code> | <code>It is necessary to use a controlled method to ensure the treatments are worthwhile.</code> | <code>0</code> |
|
426 |
+
| <code>It was conducted in silence.</code> | <code>It was done silently.</code> | <code>0</code> |
|
427 |
+
| <code>oh Lewisville any decent food in your cafeteria up there</code> | <code>Is there any decent food in your cafeteria up there in Lewisville?</code> | <code>0</code> |
|
428 |
* Loss: [<code>AdaptiveLayerLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#adaptivelayerloss) with these parameters:
|
429 |
```json
|
430 |
{
|
|
|
467 |
}
|
468 |
```
|
469 |
|
470 |
+
### Training Hyperparameters
|
471 |
+
#### Non-Default Hyperparameters
|
472 |
+
|
473 |
+
- `eval_strategy`: steps
|
474 |
+
- `per_device_train_batch_size`: 42
|
475 |
+
- `per_device_eval_batch_size`: 22
|
476 |
+
- `learning_rate`: 3e-06
|
477 |
+
- `weight_decay`: 1e-08
|
478 |
+
- `num_train_epochs`: 2
|
479 |
+
- `lr_scheduler_type`: cosine
|
480 |
+
- `warmup_ratio`: 0.5
|
481 |
+
- `save_safetensors`: False
|
482 |
+
- `fp16`: True
|
483 |
+
- `hub_model_id`: bobox/DeBERTaV3-small-ST-AdaptiveLayers-ep2-tmp
|
484 |
+
- `hub_strategy`: checkpoint
|
485 |
+
- `hub_private_repo`: True
|
486 |
+
- `batch_sampler`: no_duplicates
|
487 |
+
|
488 |
+
#### All Hyperparameters
|
489 |
+
<details><summary>Click to expand</summary>
|
490 |
+
|
491 |
+
- `overwrite_output_dir`: False
|
492 |
+
- `do_predict`: False
|
493 |
+
- `eval_strategy`: steps
|
494 |
+
- `prediction_loss_only`: True
|
495 |
+
- `per_device_train_batch_size`: 42
|
496 |
+
- `per_device_eval_batch_size`: 22
|
497 |
+
- `per_gpu_train_batch_size`: None
|
498 |
+
- `per_gpu_eval_batch_size`: None
|
499 |
+
- `gradient_accumulation_steps`: 1
|
500 |
+
- `eval_accumulation_steps`: None
|
501 |
+
- `learning_rate`: 3e-06
|
502 |
+
- `weight_decay`: 1e-08
|
503 |
+
- `adam_beta1`: 0.9
|
504 |
+
- `adam_beta2`: 0.999
|
505 |
+
- `adam_epsilon`: 1e-08
|
506 |
+
- `max_grad_norm`: 1.0
|
507 |
+
- `num_train_epochs`: 2
|
508 |
+
- `max_steps`: -1
|
509 |
+
- `lr_scheduler_type`: cosine
|
510 |
+
- `lr_scheduler_kwargs`: {}
|
511 |
+
- `warmup_ratio`: 0.5
|
512 |
+
- `warmup_steps`: 0
|
513 |
+
- `log_level`: passive
|
514 |
+
- `log_level_replica`: warning
|
515 |
+
- `log_on_each_node`: True
|
516 |
+
- `logging_nan_inf_filter`: True
|
517 |
+
- `save_safetensors`: False
|
518 |
+
- `save_on_each_node`: False
|
519 |
+
- `save_only_model`: False
|
520 |
+
- `restore_callback_states_from_checkpoint`: False
|
521 |
+
- `no_cuda`: False
|
522 |
+
- `use_cpu`: False
|
523 |
+
- `use_mps_device`: False
|
524 |
+
- `seed`: 42
|
525 |
+
- `data_seed`: None
|
526 |
+
- `jit_mode_eval`: False
|
527 |
+
- `use_ipex`: False
|
528 |
+
- `bf16`: False
|
529 |
+
- `fp16`: True
|
530 |
+
- `fp16_opt_level`: O1
|
531 |
+
- `half_precision_backend`: auto
|
532 |
+
- `bf16_full_eval`: False
|
533 |
+
- `fp16_full_eval`: False
|
534 |
+
- `tf32`: None
|
535 |
+
- `local_rank`: 0
|
536 |
+
- `ddp_backend`: None
|
537 |
+
- `tpu_num_cores`: None
|
538 |
+
- `tpu_metrics_debug`: False
|
539 |
+
- `debug`: []
|
540 |
+
- `dataloader_drop_last`: False
|
541 |
+
- `dataloader_num_workers`: 0
|
542 |
+
- `dataloader_prefetch_factor`: None
|
543 |
+
- `past_index`: -1
|
544 |
+
- `disable_tqdm`: False
|
545 |
+
- `remove_unused_columns`: True
|
546 |
+
- `label_names`: None
|
547 |
+
- `load_best_model_at_end`: False
|
548 |
+
- `ignore_data_skip`: False
|
549 |
+
- `fsdp`: []
|
550 |
+
- `fsdp_min_num_params`: 0
|
551 |
+
- `fsdp_config`: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
|
552 |
+
- `fsdp_transformer_layer_cls_to_wrap`: None
|
553 |
+
- `accelerator_config`: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
|
554 |
+
- `deepspeed`: None
|
555 |
+
- `label_smoothing_factor`: 0.0
|
556 |
+
- `optim`: adamw_torch
|
557 |
+
- `optim_args`: None
|
558 |
+
- `adafactor`: False
|
559 |
+
- `group_by_length`: False
|
560 |
+
- `length_column_name`: length
|
561 |
+
- `ddp_find_unused_parameters`: None
|
562 |
+
- `ddp_bucket_cap_mb`: None
|
563 |
+
- `ddp_broadcast_buffers`: False
|
564 |
+
- `dataloader_pin_memory`: True
|
565 |
+
- `dataloader_persistent_workers`: False
|
566 |
+
- `skip_memory_metrics`: True
|
567 |
+
- `use_legacy_prediction_loop`: False
|
568 |
+
- `push_to_hub`: False
|
569 |
+
- `resume_from_checkpoint`: None
|
570 |
+
- `hub_model_id`: bobox/DeBERTaV3-small-ST-AdaptiveLayers-ep2-tmp
|
571 |
+
- `hub_strategy`: checkpoint
|
572 |
+
- `hub_private_repo`: True
|
573 |
+
- `hub_always_push`: False
|
574 |
+
- `gradient_checkpointing`: False
|
575 |
+
- `gradient_checkpointing_kwargs`: None
|
576 |
+
- `include_inputs_for_metrics`: False
|
577 |
+
- `eval_do_concat_batches`: True
|
578 |
+
- `fp16_backend`: auto
|
579 |
+
- `push_to_hub_model_id`: None
|
580 |
+
- `push_to_hub_organization`: None
|
581 |
+
- `mp_parameters`:
|
582 |
+
- `auto_find_batch_size`: False
|
583 |
+
- `full_determinism`: False
|
584 |
+
- `torchdynamo`: None
|
585 |
+
- `ray_scope`: last
|
586 |
+
- `ddp_timeout`: 1800
|
587 |
+
- `torch_compile`: False
|
588 |
+
- `torch_compile_backend`: None
|
589 |
+
- `torch_compile_mode`: None
|
590 |
+
- `dispatch_batches`: None
|
591 |
+
- `split_batches`: None
|
592 |
+
- `include_tokens_per_second`: False
|
593 |
+
- `include_num_input_tokens_seen`: False
|
594 |
+
- `neftune_noise_alpha`: None
|
595 |
+
- `optim_target_modules`: None
|
596 |
+
- `batch_eval_metrics`: False
|
597 |
+
- `batch_sampler`: no_duplicates
|
598 |
+
- `multi_dataset_batch_sampler`: proportional
|
599 |
+
|
600 |
+
</details>
|
601 |
+
|
602 |
### Training Logs
|
603 |
+
| Epoch | Step | Training Loss | loss | max_ap | spearman_cosine |
|
604 |
+
|:-----:|:----:|:-------------:|:------:|:------:|:---------------:|
|
605 |
+
| 0.1 | 160 | 4.6003 | 4.8299 | 0.6017 | - |
|
606 |
+
| 0.2 | 320 | 4.0659 | 4.3436 | 0.6168 | - |
|
607 |
+
| 0.3 | 480 | 3.4886 | 4.0840 | 0.6339 | - |
|
608 |
+
| 0.4 | 640 | 3.0592 | 3.6422 | 0.6611 | - |
|
609 |
+
| 0.5 | 800 | 2.5728 | 3.1927 | 0.6773 | - |
|
610 |
+
| 0.6 | 960 | 2.184 | 2.8322 | 0.6893 | - |
|
611 |
+
| 0.7 | 1120 | 1.8744 | 2.4892 | 0.6954 | - |
|
612 |
+
| 0.8 | 1280 | 1.757 | 2.4453 | 0.7002 | - |
|
613 |
+
| 0.9 | 1440 | 1.5872 | 2.2565 | 0.7010 | - |
|
614 |
+
| 1.0 | 1600 | 1.446 | 2.1391 | 0.7046 | - |
|
615 |
+
| 1.1 | 1760 | 1.3892 | 2.1236 | 0.7058 | - |
|
616 |
+
| 1.2 | 1920 | 1.2567 | 1.9738 | 0.7053 | - |
|
617 |
+
| 1.3 | 2080 | 1.2233 | 1.8925 | 0.7063 | - |
|
618 |
+
| 1.4 | 2240 | 1.1954 | 1.8392 | 0.7075 | - |
|
619 |
+
| 1.5 | 2400 | 1.1395 | 1.9081 | 0.7065 | - |
|
620 |
+
| 1.6 | 2560 | 1.1211 | 1.8080 | 0.7074 | - |
|
621 |
+
| 1.7 | 2720 | 1.0825 | 1.8408 | 0.7073 | - |
|
622 |
+
| 1.8 | 2880 | 1.1358 | 1.7363 | 0.7073 | - |
|
623 |
+
| 1.9 | 3040 | 1.0628 | 1.8936 | 0.7072 | - |
|
624 |
+
| 2.0 | 3200 | 1.1412 | 1.7846 | 0.7072 | - |
|
625 |
+
| None | 0 | - | 3.0121 | 0.7072 | 0.7345 |
|
626 |
|
627 |
|
628 |
### Framework Versions
|
pytorch_model.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 565251810
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:302073fb610aae136ce3650813dfe4e09b6216dbe2b7ded3d56cad2822c48514
|
3 |
size 565251810
|