jmwri commited on
Commit
48bd507
·
verified ·
1 Parent(s): cb59483

Upload model

Browse files
Files changed (6) hide show
  1. README.md +148 -0
  2. config.json +44 -0
  3. decide.json +654 -0
  4. eval.json +416 -0
  5. model.safetensors +3 -0
  6. tokenizer.json +0 -0
README.md ADDED
@@ -0,0 +1,148 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ language: en
4
+ base_model: answerdotai/ModernBERT-base
5
+ library_name: decide
6
+ pipeline_tag: zero-shot-classification
7
+ tags:
8
+ - modernbert
9
+ - decision-model
10
+ - zero-shot-classification
11
+ - calibrated
12
+ ---
13
+
14
+ # decide-0.1.0
15
+
16
+ A small, fast, **non-autoregressive decision model**. Given some state, an instruction and a set of candidate
17
+ options, it scores every option in a single forward pass and returns calibrated probabilities. It answers
18
+ three kinds of question with the same weights:
19
+
20
+ - **choice**: a distribution over K mutually exclusive options
21
+ - **noul**: the probability that a condition holds (a yes/no question)
22
+ - **score**: an expected level on an ordered scale
23
+
24
+ It is fine-tuned from [ModernBERT-base](https://huggingface.co/answerdotai/ModernBERT-base) (149M parameters).
25
+ The packed input is `<instructions> <state> [SEP] [MASK] option0 [MASK] option1 ...`; a small MLP head scores the
26
+ encoder state at each `[MASK]`. Each option attends only to the shared prefix and to itself and its position ids restart
27
+ after the prefix, so an option's score depends on *(premise, that option)* alone: **permuting the options permutes the
28
+ scores and changes nothing else** (measured max logit change under permutation: 1.4e-06).
29
+
30
+ The model, tokenizer, training loop, evaluation and this card were all produced with the pure-Go toolchain in
31
+ [jmwri/decide](https://github.com/jmwri/decide); no Python or PyTorch was used to train it.
32
+
33
+ ## Usage
34
+
35
+ ```go
36
+ import "github.com/jmwri/decide"
37
+
38
+ decide.SetDefault(decide.NewLocal("/path/to/this/model")) // or: decide pull --repo jmwri/decide
39
+ ans, _ := decide.Decide(ctx, "Replication lag exceeded 45 seconds.", decide.Options{
40
+ {Key: "infrastructure", Description: "Database, hardware, network or server failures"},
41
+ {Key: "billing", Description: "Invoices, payments, refunds"},
42
+ }, "Classify the root cause domain.")
43
+ ```
44
+
45
+ ## Evaluation
46
+
47
+ Accuracy is top-option accuracy. ECE is the expected calibration error (15 bins) of the top-option confidence,
48
+ after the per-kind temperature scaling shipped in `decide.json`.
49
+
50
+ ### Held-out validation splits of the training tasks
51
+
52
+ | task | n | accuracy | NLL | ECE |
53
+ | :-- | --: | --: | --: | --: |
54
+ | **all** | 6307 | **78.7%** | 0.512 | 0.014 |
55
+ | kind: choice | 3782 | 77.5% | 0.549 | 0.016 |
56
+ | kind: noul | 2146 | 84.6% | 0.352 | 0.012 |
57
+ | kind: score | 379 | 57.0% | 1.034 | 0.068 |
58
+ | ag_news | 300 | 91.7% | 0.257 | 0.047 |
59
+ | amazon_polarity | 300 | 94.7% | 0.172 | 0.027 |
60
+ | arc | 200 | 38.5% | 1.331 | 0.101 |
61
+ | arc_easy | 200 | 62.0% | 0.926 | 0.150 |
62
+ | banking77 | 300 | 96.3% | 0.106 | 0.022 |
63
+ | bitext_support | 300 | 99.7% | 0.015 | 0.011 |
64
+ | boolq | 300 | 75.7% | 0.493 | 0.050 |
65
+ | civil_comments | 156 | 73.1% (MAE 0.47 levels) | 0.606 | 0.065 |
66
+ | clinc150 | 300 | 98.0% | 0.059 | 0.019 |
67
+ | cola | 200 | 76.5% | 0.524 | 0.055 |
68
+ | commonsense_qa | 300 | 62.3% | 0.891 | 0.092 |
69
+ | dbpedia14 | 300 | 73.3% | 0.918 | 0.162 |
70
+ | go_emotions | 256 | 83.2% | 0.463 | 0.059 |
71
+ | hellaswag | 300 | 62.3% | 0.927 | 0.069 |
72
+ | mnli | 300 | 85.3% | 0.416 | 0.084 |
73
+ | mrpc | 200 | 84.0% | 0.367 | 0.078 |
74
+ | qnli | 200 | 90.5% | 0.234 | 0.057 |
75
+ | qqp | 200 | 83.5% | 0.410 | 0.051 |
76
+ | rte | 200 | 79.0% | 0.436 | 0.050 |
77
+ | snli | 295 | 85.4% | 0.369 | 0.045 |
78
+ | sst2 | 300 | 93.0% | 0.193 | 0.027 |
79
+ | sst5 | 300 | 57.0% (MAE 0.56 levels) | 1.068 | 0.059 |
80
+ | wanli | 300 | 73.0% | 0.591 | 0.050 |
81
+ | winogrande | 300 | 56.3% | 0.757 | 0.134 |
82
+
83
+ Tasks in this second table were **never trained on** (zero-shot to the model):
84
+
85
+ ### Held-out tasks
86
+
87
+ | task | n | accuracy | NLL | ECE |
88
+ | :-- | --: | --: | --: | --: |
89
+ | **all** | 4693 | **47.9%** | 1.242 | 0.094 |
90
+ | kind: choice | 2500 | 40.7% | 1.414 | 0.121 |
91
+ | kind: noul | 1193 | 82.6% | 0.530 | 0.085 |
92
+ | kind: score | 1000 | 24.4% | 1.662 | 0.108 |
93
+ | emotion | 1000 | 51.6% | 1.454 | 0.168 |
94
+ | mmlu | 1500 | 33.5% | 1.387 | 0.089 |
95
+ | sms_spam | 1193 | 82.6% | 0.530 | 0.085 |
96
+ | stsb | 1000 | 24.4% (MAE 0.93 levels) | 1.662 | 0.108 |
97
+
98
+ ## Training data
99
+
100
+ Fine-tuned on 208762 examples built from the public datasets below (options are re-sampled and verbalised in several
101
+ styles; classification tasks are also posed as yes/no verification questions). Licenses are as declared on the
102
+ Hugging Face hub; **review them before redistributing or commercialising the weights**, since several datasets
103
+ carry share-alike or unspecified terms.
104
+
105
+ | dataset | train examples | license (as declared) |
106
+ | :-- | --: | :-- |
107
+ | ag_news | 6000 | unknown |
108
+ | amazon_polarity | 8000 | apache-2.0 |
109
+ | arc | 1119 | cc-by-sa-4.0 |
110
+ | arc_easy | 2251 | cc-by-sa-4.0 |
111
+ | banking77 | 9993 | cc-by-4.0 |
112
+ | bitext_support | 8990 | cdla-sharing-1.0 |
113
+ | boolq | 9000 | cc-by-sa-3.0 |
114
+ | civil_comments | 9391 | cc0-1.0 |
115
+ | clinc150 | 12000 | cc-by-3.0 |
116
+ | cola | 6000 | other (GLUE) |
117
+ | commonsense_qa | 9000 | mit |
118
+ | dbpedia14 | 8000 | cc-by-sa-3.0 |
119
+ | emotion (held out, evaluation only) | 0 | other (dair-ai) |
120
+ | go_emotions | 7520 | apache-2.0 |
121
+ | hellaswag | 9000 | mit |
122
+ | mmlu (held out, evaluation only) | 0 | mit |
123
+ | mnli | 24000 | cc-by-3.0/cc-by-sa-3.0/mit/other |
124
+ | mrpc | 3668 | other (GLUE) |
125
+ | qnli | 8000 | other (GLUE) |
126
+ | qqp | 8000 | other (GLUE) |
127
+ | rte | 2490 | other (GLUE) |
128
+ | sms_spam (held out, evaluation only) | 0 | unknown (UCI) |
129
+ | snli | 15976 | cc-by-sa-4.0 |
130
+ | sst2 | 8000 | other (GLUE) |
131
+ | sst5 | 8500 | unspecified (SetFit/sst5) |
132
+ | stsb (held out, evaluation only) | 0 | other (GLUE) |
133
+ | wanli | 16000 | cc-by-4.0 |
134
+ | winogrande | 8000 | cc-by-4.0 |
135
+
136
+ ## Limitations
137
+
138
+ - English only. Inputs are limited to 8192 tokens; option sets are best kept under ~10 (use a two-stage taxonomy beyond that).
139
+ - Trained on public classification, inference and multiple-choice datasets, so it is strongest on those styles of
140
+ question. Explicit, descriptive option texts work better than bare labels. Domain-specific jargon or rubrics with
141
+ no descriptive criteria fall back to general language priors.
142
+ - Probabilities are calibrated on held-out data from the training tasks; on very different distributions they will
143
+ be over-confident. Use a confidence gate and route uncertain cases to review.
144
+ - Not for decisions that affect people's rights, safety or livelihood without human oversight.
145
+
146
+ ## License
147
+
148
+ Weights: Apache-2.0 (ModernBERT-base is Apache-2.0), subject to the training-data terms above.
config.json ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "ModernBERT-base",
3
+ "architectures": [
4
+ "ModernBertForMaskedLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 50281,
9
+ "classifier_activation": "gelu",
10
+ "classifier_bias": false,
11
+ "classifier_dropout": 0.0,
12
+ "classifier_pooling": "mean",
13
+ "cls_token_id": 50281,
14
+ "decoder_bias": true,
15
+ "deterministic_flash_attn": false,
16
+ "embedding_dropout": 0.0,
17
+ "eos_token_id": 50282,
18
+ "global_attn_every_n_layers": 3,
19
+ "global_rope_theta": 160000.0,
20
+ "gradient_checkpointing": false,
21
+ "hidden_activation": "gelu",
22
+ "hidden_size": 768,
23
+ "initializer_cutoff_factor": 2.0,
24
+ "initializer_range": 0.02,
25
+ "intermediate_size": 1152,
26
+ "layer_norm_eps": 1e-05,
27
+ "local_attention": 128,
28
+ "local_rope_theta": 10000.0,
29
+ "max_position_embeddings": 8192,
30
+ "mlp_bias": false,
31
+ "mlp_dropout": 0.0,
32
+ "model_type": "modernbert",
33
+ "norm_bias": false,
34
+ "norm_eps": 1e-05,
35
+ "num_attention_heads": 12,
36
+ "num_hidden_layers": 22,
37
+ "pad_token_id": 50283,
38
+ "position_embedding_type": "absolute",
39
+ "sep_token_id": 50282,
40
+ "tie_word_embeddings": true,
41
+ "torch_dtype": "float32",
42
+ "transformers_version": "4.47.0.dev0",
43
+ "vocab_size": 50368
44
+ }
decide.json ADDED
@@ -0,0 +1,654 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "decide",
3
+ "model_id": "decide-0.1.0",
4
+ "format": 1,
5
+ "independent_options": true,
6
+ "max_tokens": 8192,
7
+ "calibration": {
8
+ "temperature": {
9
+ "choice": 1.2391321764132588,
10
+ "noul": 1.0490489190539387,
11
+ "score": 0.9080149508177681
12
+ },
13
+ "default": 1.163017508628048,
14
+ "examples": {
15
+ "choice": 3782,
16
+ "noul": 2146,
17
+ "score": 379
18
+ },
19
+ "ece_before": {
20
+ "choice": 0.029260942121753057,
21
+ "noul": 0.013277131839831471,
22
+ "score": 0.07500635509985569
23
+ },
24
+ "ece_after": {
25
+ "choice": 0.01618419731483734,
26
+ "noul": 0.01213385330242672,
27
+ "score": 0.06755450718609643
28
+ }
29
+ },
30
+ "base_model": "answerdotai/ModernBERT-base",
31
+ "training": {
32
+ "corpus": {
33
+ "created": "2026-09-25T13:00:33Z",
34
+ "ood": 4693,
35
+ "seed": 1,
36
+ "tasks": [
37
+ {
38
+ "license": "unknown",
39
+ "name": "ag_news",
40
+ "ood": 0,
41
+ "train": 6000,
42
+ "val": 300
43
+ },
44
+ {
45
+ "license": "apache-2.0",
46
+ "name": "amazon_polarity",
47
+ "ood": 0,
48
+ "train": 8000,
49
+ "val": 300
50
+ },
51
+ {
52
+ "license": "cc-by-sa-4.0",
53
+ "name": "arc",
54
+ "ood": 0,
55
+ "train": 1119,
56
+ "val": 200
57
+ },
58
+ {
59
+ "license": "cc-by-sa-4.0",
60
+ "name": "arc_easy",
61
+ "ood": 0,
62
+ "train": 2251,
63
+ "val": 200
64
+ },
65
+ {
66
+ "license": "cc-by-4.0",
67
+ "name": "banking77",
68
+ "ood": 0,
69
+ "train": 9993,
70
+ "val": 300
71
+ },
72
+ {
73
+ "license": "cdla-sharing-1.0",
74
+ "name": "bitext_support",
75
+ "ood": 0,
76
+ "train": 8990,
77
+ "val": 300
78
+ },
79
+ {
80
+ "license": "cc-by-sa-3.0",
81
+ "name": "boolq",
82
+ "ood": 0,
83
+ "train": 9000,
84
+ "val": 300
85
+ },
86
+ {
87
+ "license": "cc0-1.0",
88
+ "name": "civil_comments",
89
+ "ood": 0,
90
+ "train": 9391,
91
+ "val": 156
92
+ },
93
+ {
94
+ "license": "cc-by-3.0",
95
+ "name": "clinc150",
96
+ "ood": 0,
97
+ "train": 12000,
98
+ "val": 300
99
+ },
100
+ {
101
+ "license": "other (GLUE)",
102
+ "name": "cola",
103
+ "ood": 0,
104
+ "train": 6000,
105
+ "val": 200
106
+ },
107
+ {
108
+ "license": "mit",
109
+ "name": "commonsense_qa",
110
+ "ood": 0,
111
+ "train": 9000,
112
+ "val": 300
113
+ },
114
+ {
115
+ "license": "cc-by-sa-3.0",
116
+ "name": "dbpedia14",
117
+ "ood": 0,
118
+ "train": 8000,
119
+ "val": 300
120
+ },
121
+ {
122
+ "license": "other (dair-ai)",
123
+ "name": "emotion",
124
+ "ood": 1000,
125
+ "train": 0,
126
+ "val": 0
127
+ },
128
+ {
129
+ "license": "apache-2.0",
130
+ "name": "go_emotions",
131
+ "ood": 0,
132
+ "train": 7520,
133
+ "val": 256
134
+ },
135
+ {
136
+ "license": "mit",
137
+ "name": "hellaswag",
138
+ "ood": 0,
139
+ "train": 9000,
140
+ "val": 300
141
+ },
142
+ {
143
+ "license": "mit",
144
+ "name": "mmlu",
145
+ "ood": 1500,
146
+ "train": 0,
147
+ "val": 0
148
+ },
149
+ {
150
+ "license": "cc-by-3.0/cc-by-sa-3.0/mit/other",
151
+ "name": "mnli",
152
+ "ood": 0,
153
+ "train": 24000,
154
+ "val": 300
155
+ },
156
+ {
157
+ "license": "other (GLUE)",
158
+ "name": "mrpc",
159
+ "ood": 0,
160
+ "train": 3668,
161
+ "val": 200
162
+ },
163
+ {
164
+ "license": "other (GLUE)",
165
+ "name": "qnli",
166
+ "ood": 0,
167
+ "train": 8000,
168
+ "val": 200
169
+ },
170
+ {
171
+ "license": "other (GLUE)",
172
+ "name": "qqp",
173
+ "ood": 0,
174
+ "train": 8000,
175
+ "val": 200
176
+ },
177
+ {
178
+ "license": "other (GLUE)",
179
+ "name": "rte",
180
+ "ood": 0,
181
+ "train": 2490,
182
+ "val": 200
183
+ },
184
+ {
185
+ "license": "unknown (UCI)",
186
+ "name": "sms_spam",
187
+ "ood": 1193,
188
+ "train": 0,
189
+ "val": 0
190
+ },
191
+ {
192
+ "license": "cc-by-sa-4.0",
193
+ "name": "snli",
194
+ "ood": 0,
195
+ "train": 15976,
196
+ "val": 295
197
+ },
198
+ {
199
+ "license": "other (GLUE)",
200
+ "name": "sst2",
201
+ "ood": 0,
202
+ "train": 8000,
203
+ "val": 300
204
+ },
205
+ {
206
+ "license": "unspecified (SetFit/sst5)",
207
+ "name": "sst5",
208
+ "ood": 0,
209
+ "train": 8500,
210
+ "val": 300
211
+ },
212
+ {
213
+ "license": "other (GLUE)",
214
+ "name": "stsb",
215
+ "ood": 1000,
216
+ "train": 0,
217
+ "val": 0
218
+ },
219
+ {
220
+ "license": "cc-by-4.0",
221
+ "name": "wanli",
222
+ "ood": 0,
223
+ "train": 16000,
224
+ "val": 300
225
+ },
226
+ {
227
+ "license": "cc-by-4.0",
228
+ "name": "winogrande",
229
+ "ood": 0,
230
+ "train": 8000,
231
+ "val": 300
232
+ }
233
+ ],
234
+ "train": 208762,
235
+ "val": 6307
236
+ },
237
+ "reports": {
238
+ "ood": {
239
+ "overall": {
240
+ "n": 4693,
241
+ "acc": 0.47879821010014917,
242
+ "nll": 1.2421905566020706,
243
+ "ece": 0.09417245522448311,
244
+ "mae": 0.9346012027336152
245
+ },
246
+ "by_kind": {
247
+ "choice": {
248
+ "n": 2500,
249
+ "acc": 0.4072,
250
+ "nll": 1.413836112602486,
251
+ "ece": 0.12063916455782075
252
+ },
253
+ "noul": {
254
+ "n": 1193,
255
+ "acc": 0.8256496227996647,
256
+ "nll": 0.5302233131509403,
257
+ "ece": 0.08536369278712
258
+ },
259
+ "score": {
260
+ "n": 1000,
261
+ "acc": 0.244,
262
+ "nll": 1.6624535880382436,
263
+ "ece": 0.10795363782931346,
264
+ "mae": 0.9346012027336152
265
+ }
266
+ },
267
+ "by_task": {
268
+ "emotion": {
269
+ "n": 1000,
270
+ "acc": 0.516,
271
+ "nll": 1.4540259847281438,
272
+ "ece": 0.16794626130579743
273
+ },
274
+ "mmlu": {
275
+ "n": 1500,
276
+ "acc": 0.33466666666666667,
277
+ "nll": 1.3870428645187136,
278
+ "ece": 0.08935882375098397
279
+ },
280
+ "sms_spam": {
281
+ "n": 1193,
282
+ "acc": 0.8256496227996647,
283
+ "nll": 0.5302233131509403,
284
+ "ece": 0.08536369278712
285
+ },
286
+ "stsb": {
287
+ "n": 1000,
288
+ "acc": 0.244,
289
+ "nll": 1.6624535880382436,
290
+ "ece": 0.10795363782931346,
291
+ "mae": 0.9346012027336152
292
+ }
293
+ }
294
+ },
295
+ "order_invariance_max_logit_delta": 0.000001430511474609375,
296
+ "val": {
297
+ "overall": {
298
+ "n": 6307,
299
+ "acc": 0.7869034406215316,
300
+ "nll": 0.5115170661565527,
301
+ "ece": 0.014218897518411378,
302
+ "mae": 0.5410988474145996
303
+ },
304
+ "by_kind": {
305
+ "choice": {
306
+ "n": 3782,
307
+ "acc": 0.7749867794817556,
308
+ "nll": 0.5493717640572233,
309
+ "ece": 0.01618419731483734
310
+ },
311
+ "noul": {
312
+ "n": 2146,
313
+ "acc": 0.8462255358807083,
314
+ "nll": 0.3524608601159457,
315
+ "ece": 0.01213385330242672
316
+ },
317
+ "score": {
318
+ "n": 379,
319
+ "acc": 0.5699208443271768,
320
+ "nll": 1.0343881761903733,
321
+ "ece": 0.06755450718609643,
322
+ "mae": 0.5410988474145996
323
+ }
324
+ },
325
+ "by_task": {
326
+ "ag_news": {
327
+ "n": 300,
328
+ "acc": 0.9166666666666666,
329
+ "nll": 0.2565925165832986,
330
+ "ece": 0.04654975703455169
331
+ },
332
+ "amazon_polarity": {
333
+ "n": 300,
334
+ "acc": 0.9466666666666667,
335
+ "nll": 0.17173071513378904,
336
+ "ece": 0.026833555150243663
337
+ },
338
+ "arc": {
339
+ "n": 200,
340
+ "acc": 0.385,
341
+ "nll": 1.3313207350684448,
342
+ "ece": 0.10139976354793183
343
+ },
344
+ "arc_easy": {
345
+ "n": 200,
346
+ "acc": 0.62,
347
+ "nll": 0.9261051654169462,
348
+ "ece": 0.14976230505505717
349
+ },
350
+ "banking77": {
351
+ "n": 300,
352
+ "acc": 0.9633333333333334,
353
+ "nll": 0.10632949024583604,
354
+ "ece": 0.022141319592502345
355
+ },
356
+ "bitext_support": {
357
+ "n": 300,
358
+ "acc": 0.9966666666666667,
359
+ "nll": 0.015218504805951343,
360
+ "ece": 0.010681751338801459
361
+ },
362
+ "boolq": {
363
+ "n": 300,
364
+ "acc": 0.7566666666666667,
365
+ "nll": 0.4932730046513006,
366
+ "ece": 0.049510831435021566
367
+ },
368
+ "civil_comments": {
369
+ "n": 156,
370
+ "acc": 0.7307692307692307,
371
+ "nll": 0.6057626789607697,
372
+ "ece": 0.0650423358988158,
373
+ "mae": 0.47329407998848366
374
+ },
375
+ "clinc150": {
376
+ "n": 300,
377
+ "acc": 0.98,
378
+ "nll": 0.05891124724717898,
379
+ "ece": 0.018629386263961198
380
+ },
381
+ "cola": {
382
+ "n": 200,
383
+ "acc": 0.765,
384
+ "nll": 0.5240199833276187,
385
+ "ece": 0.05509032579590858
386
+ },
387
+ "commonsense_qa": {
388
+ "n": 300,
389
+ "acc": 0.6233333333333333,
390
+ "nll": 0.8910569891825245,
391
+ "ece": 0.09193036205349639
392
+ },
393
+ "dbpedia14": {
394
+ "n": 300,
395
+ "acc": 0.7333333333333333,
396
+ "nll": 0.9184878257796362,
397
+ "ece": 0.16239219449252368
398
+ },
399
+ "go_emotions": {
400
+ "n": 256,
401
+ "acc": 0.83203125,
402
+ "nll": 0.4630004163802363,
403
+ "ece": 0.05899007829345532
404
+ },
405
+ "hellaswag": {
406
+ "n": 300,
407
+ "acc": 0.6233333333333333,
408
+ "nll": 0.9266911671531523,
409
+ "ece": 0.06857599903037215
410
+ },
411
+ "mnli": {
412
+ "n": 300,
413
+ "acc": 0.8533333333333334,
414
+ "nll": 0.41573105275184746,
415
+ "ece": 0.08377938464455348
416
+ },
417
+ "mrpc": {
418
+ "n": 200,
419
+ "acc": 0.84,
420
+ "nll": 0.3668160199781415,
421
+ "ece": 0.0776006643850251
422
+ },
423
+ "qnli": {
424
+ "n": 200,
425
+ "acc": 0.905,
426
+ "nll": 0.23383141787490153,
427
+ "ece": 0.057082406729685335
428
+ },
429
+ "qqp": {
430
+ "n": 200,
431
+ "acc": 0.835,
432
+ "nll": 0.40977572855048544,
433
+ "ece": 0.051345911983505854
434
+ },
435
+ "rte": {
436
+ "n": 200,
437
+ "acc": 0.79,
438
+ "nll": 0.43553678641342425,
439
+ "ece": 0.04971707095558052
440
+ },
441
+ "snli": {
442
+ "n": 295,
443
+ "acc": 0.8542372881355932,
444
+ "nll": 0.36925722531464455,
445
+ "ece": 0.04481972014506814
446
+ },
447
+ "sst2": {
448
+ "n": 300,
449
+ "acc": 0.93,
450
+ "nll": 0.19309734208899607,
451
+ "ece": 0.02714302958058557
452
+ },
453
+ "sst5": {
454
+ "n": 300,
455
+ "acc": 0.57,
456
+ "nll": 1.0681591164761377,
457
+ "ece": 0.059482954822956034,
458
+ "mae": 0.5589541028368104
459
+ },
460
+ "wanli": {
461
+ "n": 300,
462
+ "acc": 0.73,
463
+ "nll": 0.5905371330672343,
464
+ "ece": 0.049899155103776516
465
+ },
466
+ "winogrande": {
467
+ "n": 300,
468
+ "acc": 0.5633333333333334,
469
+ "nll": 0.7565139046476452,
470
+ "ece": 0.13382848952510093
471
+ }
472
+ }
473
+ },
474
+ "val_uncalibrated": {
475
+ "overall": {
476
+ "n": 6307,
477
+ "acc": 0.7869034406215316,
478
+ "nll": 0.5171425470109821,
479
+ "ece": 0.01914388866821656,
480
+ "mae": 0.5422975095408504
481
+ },
482
+ "by_kind": {
483
+ "choice": {
484
+ "n": 3782,
485
+ "acc": 0.7749867794817556,
486
+ "nll": 0.5583776189372751,
487
+ "ece": 0.029260942121753057
488
+ },
489
+ "noul": {
490
+ "n": 2146,
491
+ "acc": 0.8462255358807083,
492
+ "nll": 0.3527691499120017,
493
+ "ece": 0.013277131839831471
494
+ },
495
+ "score": {
496
+ "n": 379,
497
+ "acc": 0.5699208443271768,
498
+ "nll": 1.036388637114314,
499
+ "ece": 0.07500635509985569,
500
+ "mae": 0.5422975095408504
501
+ }
502
+ },
503
+ "by_task": {
504
+ "ag_news": {
505
+ "n": 300,
506
+ "acc": 0.9166666666666666,
507
+ "nll": 0.2431063015392516,
508
+ "ece": 0.043809829443710105
509
+ },
510
+ "amazon_polarity": {
511
+ "n": 300,
512
+ "acc": 0.9466666666666667,
513
+ "nll": 0.17587279831232996,
514
+ "ece": 0.03079354287919299
515
+ },
516
+ "arc": {
517
+ "n": 200,
518
+ "acc": 0.385,
519
+ "nll": 1.3690203129845806,
520
+ "ece": 0.1356287165567225
521
+ },
522
+ "arc_easy": {
523
+ "n": 200,
524
+ "acc": 0.62,
525
+ "nll": 0.8985914737968268,
526
+ "ece": 0.09592350247168285
527
+ },
528
+ "banking77": {
529
+ "n": 300,
530
+ "acc": 0.9633333333333334,
531
+ "nll": 0.09935893217292358,
532
+ "ece": 0.02434530582050962
533
+ },
534
+ "bitext_support": {
535
+ "n": 300,
536
+ "acc": 0.9966666666666667,
537
+ "nll": 0.010764701168010605,
538
+ "ece": 0.00891246353771034
539
+ },
540
+ "boolq": {
541
+ "n": 300,
542
+ "acc": 0.7566666666666667,
543
+ "nll": 0.4945305966907107,
544
+ "ece": 0.04332081718926268
545
+ },
546
+ "civil_comments": {
547
+ "n": 156,
548
+ "acc": 0.7307692307692307,
549
+ "nll": 0.5983461675739167,
550
+ "ece": 0.06130789466766199,
551
+ "mae": 0.4766595808692595
552
+ },
553
+ "clinc150": {
554
+ "n": 300,
555
+ "acc": 0.98,
556
+ "nll": 0.059757341261055086,
557
+ "ece": 0.008367785360717087
558
+ },
559
+ "cola": {
560
+ "n": 200,
561
+ "acc": 0.765,
562
+ "nll": 0.5234010849629517,
563
+ "ece": 0.06565588708550057
564
+ },
565
+ "commonsense_qa": {
566
+ "n": 300,
567
+ "acc": 0.6233333333333333,
568
+ "nll": 0.8602408510951066,
569
+ "ece": 0.06533702481354695
570
+ },
571
+ "dbpedia14": {
572
+ "n": 300,
573
+ "acc": 0.7333333333333333,
574
+ "nll": 1.0497484668809647,
575
+ "ece": 0.16842288562778376
576
+ },
577
+ "go_emotions": {
578
+ "n": 256,
579
+ "acc": 0.83203125,
580
+ "nll": 0.45814552259224806,
581
+ "ece": 0.0605668146014693
582
+ },
583
+ "hellaswag": {
584
+ "n": 300,
585
+ "acc": 0.6233333333333333,
586
+ "nll": 0.932426472460099,
587
+ "ece": 0.0458109846140827
588
+ },
589
+ "mnli": {
590
+ "n": 300,
591
+ "acc": 0.8533333333333334,
592
+ "nll": 0.4009607570003527,
593
+ "ece": 0.059436789332712854
594
+ },
595
+ "mrpc": {
596
+ "n": 200,
597
+ "acc": 0.84,
598
+ "nll": 0.36585517325426864,
599
+ "ece": 0.07343962385844653
600
+ },
601
+ "qnli": {
602
+ "n": 200,
603
+ "acc": 0.905,
604
+ "nll": 0.23095983012652038,
605
+ "ece": 0.0515247439228464
606
+ },
607
+ "qqp": {
608
+ "n": 200,
609
+ "acc": 0.835,
610
+ "nll": 0.40995812519282077,
611
+ "ece": 0.04288788216036697
612
+ },
613
+ "rte": {
614
+ "n": 200,
615
+ "acc": 0.79,
616
+ "nll": 0.436457330409717,
617
+ "ece": 0.05793648216390617
618
+ },
619
+ "snli": {
620
+ "n": 295,
621
+ "acc": 0.8542372881355932,
622
+ "nll": 0.3656500950165819,
623
+ "ece": 0.04518458806007379
624
+ },
625
+ "sst2": {
626
+ "n": 300,
627
+ "acc": 0.93,
628
+ "nll": 0.1905902561595418,
629
+ "ece": 0.022910879929958477
630
+ },
631
+ "sst5": {
632
+ "n": 300,
633
+ "acc": 0.57,
634
+ "nll": 1.0672356593957126,
635
+ "ece": 0.0724796112296,
636
+ "mae": 0.5595821640910361
637
+ },
638
+ "wanli": {
639
+ "n": 300,
640
+ "acc": 0.73,
641
+ "nll": 0.595219385664306,
642
+ "ece": 0.054819962020094384
643
+ },
644
+ "winogrande": {
645
+ "n": 300,
646
+ "acc": 0.5633333333333334,
647
+ "nll": 0.8077719598585643,
648
+ "ece": 0.16230751096146373
649
+ }
650
+ }
651
+ }
652
+ }
653
+ }
654
+ }
eval.json ADDED
@@ -0,0 +1,416 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "ood": {
3
+ "overall": {
4
+ "n": 4693,
5
+ "acc": 0.47879821010014917,
6
+ "nll": 1.2421905566020706,
7
+ "ece": 0.09417245522448311,
8
+ "mae": 0.9346012027336152
9
+ },
10
+ "by_kind": {
11
+ "choice": {
12
+ "n": 2500,
13
+ "acc": 0.4072,
14
+ "nll": 1.413836112602486,
15
+ "ece": 0.12063916455782075
16
+ },
17
+ "noul": {
18
+ "n": 1193,
19
+ "acc": 0.8256496227996647,
20
+ "nll": 0.5302233131509403,
21
+ "ece": 0.08536369278712
22
+ },
23
+ "score": {
24
+ "n": 1000,
25
+ "acc": 0.244,
26
+ "nll": 1.6624535880382436,
27
+ "ece": 0.10795363782931346,
28
+ "mae": 0.9346012027336152
29
+ }
30
+ },
31
+ "by_task": {
32
+ "emotion": {
33
+ "n": 1000,
34
+ "acc": 0.516,
35
+ "nll": 1.4540259847281438,
36
+ "ece": 0.16794626130579743
37
+ },
38
+ "mmlu": {
39
+ "n": 1500,
40
+ "acc": 0.33466666666666667,
41
+ "nll": 1.3870428645187136,
42
+ "ece": 0.08935882375098397
43
+ },
44
+ "sms_spam": {
45
+ "n": 1193,
46
+ "acc": 0.8256496227996647,
47
+ "nll": 0.5302233131509403,
48
+ "ece": 0.08536369278712
49
+ },
50
+ "stsb": {
51
+ "n": 1000,
52
+ "acc": 0.244,
53
+ "nll": 1.6624535880382436,
54
+ "ece": 0.10795363782931346,
55
+ "mae": 0.9346012027336152
56
+ }
57
+ }
58
+ },
59
+ "order_invariance_max_logit_delta": 0.000001430511474609375,
60
+ "val": {
61
+ "overall": {
62
+ "n": 6307,
63
+ "acc": 0.7869034406215316,
64
+ "nll": 0.5115170661565527,
65
+ "ece": 0.014218897518411378,
66
+ "mae": 0.5410988474145996
67
+ },
68
+ "by_kind": {
69
+ "choice": {
70
+ "n": 3782,
71
+ "acc": 0.7749867794817556,
72
+ "nll": 0.5493717640572233,
73
+ "ece": 0.01618419731483734
74
+ },
75
+ "noul": {
76
+ "n": 2146,
77
+ "acc": 0.8462255358807083,
78
+ "nll": 0.3524608601159457,
79
+ "ece": 0.01213385330242672
80
+ },
81
+ "score": {
82
+ "n": 379,
83
+ "acc": 0.5699208443271768,
84
+ "nll": 1.0343881761903733,
85
+ "ece": 0.06755450718609643,
86
+ "mae": 0.5410988474145996
87
+ }
88
+ },
89
+ "by_task": {
90
+ "ag_news": {
91
+ "n": 300,
92
+ "acc": 0.9166666666666666,
93
+ "nll": 0.2565925165832986,
94
+ "ece": 0.04654975703455169
95
+ },
96
+ "amazon_polarity": {
97
+ "n": 300,
98
+ "acc": 0.9466666666666667,
99
+ "nll": 0.17173071513378904,
100
+ "ece": 0.026833555150243663
101
+ },
102
+ "arc": {
103
+ "n": 200,
104
+ "acc": 0.385,
105
+ "nll": 1.3313207350684448,
106
+ "ece": 0.10139976354793183
107
+ },
108
+ "arc_easy": {
109
+ "n": 200,
110
+ "acc": 0.62,
111
+ "nll": 0.9261051654169462,
112
+ "ece": 0.14976230505505717
113
+ },
114
+ "banking77": {
115
+ "n": 300,
116
+ "acc": 0.9633333333333334,
117
+ "nll": 0.10632949024583604,
118
+ "ece": 0.022141319592502345
119
+ },
120
+ "bitext_support": {
121
+ "n": 300,
122
+ "acc": 0.9966666666666667,
123
+ "nll": 0.015218504805951343,
124
+ "ece": 0.010681751338801459
125
+ },
126
+ "boolq": {
127
+ "n": 300,
128
+ "acc": 0.7566666666666667,
129
+ "nll": 0.4932730046513006,
130
+ "ece": 0.049510831435021566
131
+ },
132
+ "civil_comments": {
133
+ "n": 156,
134
+ "acc": 0.7307692307692307,
135
+ "nll": 0.6057626789607697,
136
+ "ece": 0.0650423358988158,
137
+ "mae": 0.47329407998848366
138
+ },
139
+ "clinc150": {
140
+ "n": 300,
141
+ "acc": 0.98,
142
+ "nll": 0.05891124724717898,
143
+ "ece": 0.018629386263961198
144
+ },
145
+ "cola": {
146
+ "n": 200,
147
+ "acc": 0.765,
148
+ "nll": 0.5240199833276187,
149
+ "ece": 0.05509032579590858
150
+ },
151
+ "commonsense_qa": {
152
+ "n": 300,
153
+ "acc": 0.6233333333333333,
154
+ "nll": 0.8910569891825245,
155
+ "ece": 0.09193036205349639
156
+ },
157
+ "dbpedia14": {
158
+ "n": 300,
159
+ "acc": 0.7333333333333333,
160
+ "nll": 0.9184878257796362,
161
+ "ece": 0.16239219449252368
162
+ },
163
+ "go_emotions": {
164
+ "n": 256,
165
+ "acc": 0.83203125,
166
+ "nll": 0.4630004163802363,
167
+ "ece": 0.05899007829345532
168
+ },
169
+ "hellaswag": {
170
+ "n": 300,
171
+ "acc": 0.6233333333333333,
172
+ "nll": 0.9266911671531523,
173
+ "ece": 0.06857599903037215
174
+ },
175
+ "mnli": {
176
+ "n": 300,
177
+ "acc": 0.8533333333333334,
178
+ "nll": 0.41573105275184746,
179
+ "ece": 0.08377938464455348
180
+ },
181
+ "mrpc": {
182
+ "n": 200,
183
+ "acc": 0.84,
184
+ "nll": 0.3668160199781415,
185
+ "ece": 0.0776006643850251
186
+ },
187
+ "qnli": {
188
+ "n": 200,
189
+ "acc": 0.905,
190
+ "nll": 0.23383141787490153,
191
+ "ece": 0.057082406729685335
192
+ },
193
+ "qqp": {
194
+ "n": 200,
195
+ "acc": 0.835,
196
+ "nll": 0.40977572855048544,
197
+ "ece": 0.051345911983505854
198
+ },
199
+ "rte": {
200
+ "n": 200,
201
+ "acc": 0.79,
202
+ "nll": 0.43553678641342425,
203
+ "ece": 0.04971707095558052
204
+ },
205
+ "snli": {
206
+ "n": 295,
207
+ "acc": 0.8542372881355932,
208
+ "nll": 0.36925722531464455,
209
+ "ece": 0.04481972014506814
210
+ },
211
+ "sst2": {
212
+ "n": 300,
213
+ "acc": 0.93,
214
+ "nll": 0.19309734208899607,
215
+ "ece": 0.02714302958058557
216
+ },
217
+ "sst5": {
218
+ "n": 300,
219
+ "acc": 0.57,
220
+ "nll": 1.0681591164761377,
221
+ "ece": 0.059482954822956034,
222
+ "mae": 0.5589541028368104
223
+ },
224
+ "wanli": {
225
+ "n": 300,
226
+ "acc": 0.73,
227
+ "nll": 0.5905371330672343,
228
+ "ece": 0.049899155103776516
229
+ },
230
+ "winogrande": {
231
+ "n": 300,
232
+ "acc": 0.5633333333333334,
233
+ "nll": 0.7565139046476452,
234
+ "ece": 0.13382848952510093
235
+ }
236
+ }
237
+ },
238
+ "val_uncalibrated": {
239
+ "overall": {
240
+ "n": 6307,
241
+ "acc": 0.7869034406215316,
242
+ "nll": 0.5171425470109821,
243
+ "ece": 0.01914388866821656,
244
+ "mae": 0.5422975095408504
245
+ },
246
+ "by_kind": {
247
+ "choice": {
248
+ "n": 3782,
249
+ "acc": 0.7749867794817556,
250
+ "nll": 0.5583776189372751,
251
+ "ece": 0.029260942121753057
252
+ },
253
+ "noul": {
254
+ "n": 2146,
255
+ "acc": 0.8462255358807083,
256
+ "nll": 0.3527691499120017,
257
+ "ece": 0.013277131839831471
258
+ },
259
+ "score": {
260
+ "n": 379,
261
+ "acc": 0.5699208443271768,
262
+ "nll": 1.036388637114314,
263
+ "ece": 0.07500635509985569,
264
+ "mae": 0.5422975095408504
265
+ }
266
+ },
267
+ "by_task": {
268
+ "ag_news": {
269
+ "n": 300,
270
+ "acc": 0.9166666666666666,
271
+ "nll": 0.2431063015392516,
272
+ "ece": 0.043809829443710105
273
+ },
274
+ "amazon_polarity": {
275
+ "n": 300,
276
+ "acc": 0.9466666666666667,
277
+ "nll": 0.17587279831232996,
278
+ "ece": 0.03079354287919299
279
+ },
280
+ "arc": {
281
+ "n": 200,
282
+ "acc": 0.385,
283
+ "nll": 1.3690203129845806,
284
+ "ece": 0.1356287165567225
285
+ },
286
+ "arc_easy": {
287
+ "n": 200,
288
+ "acc": 0.62,
289
+ "nll": 0.8985914737968268,
290
+ "ece": 0.09592350247168285
291
+ },
292
+ "banking77": {
293
+ "n": 300,
294
+ "acc": 0.9633333333333334,
295
+ "nll": 0.09935893217292358,
296
+ "ece": 0.02434530582050962
297
+ },
298
+ "bitext_support": {
299
+ "n": 300,
300
+ "acc": 0.9966666666666667,
301
+ "nll": 0.010764701168010605,
302
+ "ece": 0.00891246353771034
303
+ },
304
+ "boolq": {
305
+ "n": 300,
306
+ "acc": 0.7566666666666667,
307
+ "nll": 0.4945305966907107,
308
+ "ece": 0.04332081718926268
309
+ },
310
+ "civil_comments": {
311
+ "n": 156,
312
+ "acc": 0.7307692307692307,
313
+ "nll": 0.5983461675739167,
314
+ "ece": 0.06130789466766199,
315
+ "mae": 0.4766595808692595
316
+ },
317
+ "clinc150": {
318
+ "n": 300,
319
+ "acc": 0.98,
320
+ "nll": 0.059757341261055086,
321
+ "ece": 0.008367785360717087
322
+ },
323
+ "cola": {
324
+ "n": 200,
325
+ "acc": 0.765,
326
+ "nll": 0.5234010849629517,
327
+ "ece": 0.06565588708550057
328
+ },
329
+ "commonsense_qa": {
330
+ "n": 300,
331
+ "acc": 0.6233333333333333,
332
+ "nll": 0.8602408510951066,
333
+ "ece": 0.06533702481354695
334
+ },
335
+ "dbpedia14": {
336
+ "n": 300,
337
+ "acc": 0.7333333333333333,
338
+ "nll": 1.0497484668809647,
339
+ "ece": 0.16842288562778376
340
+ },
341
+ "go_emotions": {
342
+ "n": 256,
343
+ "acc": 0.83203125,
344
+ "nll": 0.45814552259224806,
345
+ "ece": 0.0605668146014693
346
+ },
347
+ "hellaswag": {
348
+ "n": 300,
349
+ "acc": 0.6233333333333333,
350
+ "nll": 0.932426472460099,
351
+ "ece": 0.0458109846140827
352
+ },
353
+ "mnli": {
354
+ "n": 300,
355
+ "acc": 0.8533333333333334,
356
+ "nll": 0.4009607570003527,
357
+ "ece": 0.059436789332712854
358
+ },
359
+ "mrpc": {
360
+ "n": 200,
361
+ "acc": 0.84,
362
+ "nll": 0.36585517325426864,
363
+ "ece": 0.07343962385844653
364
+ },
365
+ "qnli": {
366
+ "n": 200,
367
+ "acc": 0.905,
368
+ "nll": 0.23095983012652038,
369
+ "ece": 0.0515247439228464
370
+ },
371
+ "qqp": {
372
+ "n": 200,
373
+ "acc": 0.835,
374
+ "nll": 0.40995812519282077,
375
+ "ece": 0.04288788216036697
376
+ },
377
+ "rte": {
378
+ "n": 200,
379
+ "acc": 0.79,
380
+ "nll": 0.436457330409717,
381
+ "ece": 0.05793648216390617
382
+ },
383
+ "snli": {
384
+ "n": 295,
385
+ "acc": 0.8542372881355932,
386
+ "nll": 0.3656500950165819,
387
+ "ece": 0.04518458806007379
388
+ },
389
+ "sst2": {
390
+ "n": 300,
391
+ "acc": 0.93,
392
+ "nll": 0.1905902561595418,
393
+ "ece": 0.022910879929958477
394
+ },
395
+ "sst5": {
396
+ "n": 300,
397
+ "acc": 0.57,
398
+ "nll": 1.0672356593957126,
399
+ "ece": 0.0724796112296,
400
+ "mae": 0.5595821640910361
401
+ },
402
+ "wanli": {
403
+ "n": 300,
404
+ "acc": 0.73,
405
+ "nll": 0.595219385664306,
406
+ "ece": 0.054819962020094384
407
+ },
408
+ "winogrande": {
409
+ "n": 300,
410
+ "acc": 0.5633333333333334,
411
+ "nll": 0.8077719598585643,
412
+ "ece": 0.16230751096146373
413
+ }
414
+ }
415
+ }
416
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7635aef6ea1d4119675e0835a480e2bdfad19edfab7159eecad51ee7a216db53
3
+ size 597263604
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff