Commit
•
61a0c91
1
Parent(s):
bcc98c7
Model save
Browse files- README.md +58 -261
- metrics.json +13 -13
- trainer_state.json +7 -7
- training_args.bin +1 -1
README.md
CHANGED
@@ -1,281 +1,78 @@
|
|
1 |
---
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
\ in building conversational AI using recent advances in natural language processing.\
|
7 |
-
\ It utilizes a BERT model fine-tuned for extractive question answering.\n\n \
|
8 |
-
\ ## Data Collection and Preprocessing\n The model was trained on the\
|
9 |
-
\ Stanford Question Answering Dataset (SQuAD), which contains over 100,000 question-answer\
|
10 |
-
\ pairs based on Wikipedia articles. The data preprocessing involved tokenizing\
|
11 |
-
\ context paragraphs and questions, truncating sequences to fit BERT's max length,\
|
12 |
-
\ and adding special tokens to mark question and paragraph segments.\n\n \
|
13 |
-
\ ## Model Architecture and Training\n The architecture is based on the BERT\
|
14 |
-
\ transformer model, which was pretrained on large unlabeled text corpora. For this\
|
15 |
-
\ project, the BERT base model was fine-tuned on SQuAD for extractive question answering,\
|
16 |
-
\ with additional output layers for predicting the start and end indices of the\
|
17 |
-
\ answer span.\n\n ## SQuAD 2.0 Dataset\n SQuAD 2.0 combines the existing\
|
18 |
-
\ SQuAD data with over 50,000 unanswerable questions written adversarially by crowdworkers\
|
19 |
-
\ to look similar to answerable ones. This version of the dataset challenges models\
|
20 |
-
\ to not only produce answers when possible but also determine when no answer is\
|
21 |
-
\ supported by the paragraph and abstain from answering.\n "
|
22 |
-
intended_use: "\n - Answering questions from the squad_v2 dataset.\n \
|
23 |
-
\ - Developing question-answering systems within the scope of the aai520-project.\n\
|
24 |
-
\ - Research and experimentation in the NLP question-answering domain.\n\
|
25 |
-
\ "
|
26 |
-
limitations_and_bias: "\n The model inherits limitations and biases from the\
|
27 |
-
\ 'roberta-base' model, as it was trained on the same foundational data. \n \
|
28 |
-
\ It may underperform on questions that are ambiguous or too far outside the\
|
29 |
-
\ scope of the topics covered in the squad_v2 dataset. \n Additionally, the\
|
30 |
-
\ model may reflect societal biases present in its training data.\n "
|
31 |
-
ethical_considerations: "\n This model should not be used for making critical\
|
32 |
-
\ decisions without human oversight, \n as it can generate incorrect or biased\
|
33 |
-
\ answers, especially for topics not covered in the training data. \n Users\
|
34 |
-
\ should also consider the ethical implications of using AI in decision-making processes\
|
35 |
-
\ and the potential for perpetuating biases.\n "
|
36 |
-
evaluation: "\n The model was evaluated on the squad_v2 dataset using various\
|
37 |
-
\ metrics. These metrics, along with their corresponding scores, \n are detailed\
|
38 |
-
\ in the 'eval_results' section. The evaluation process ensured a comprehensive\
|
39 |
-
\ assessment of the model's performance \n in question-answering scenarios.\n\
|
40 |
-
\ "
|
41 |
-
training: "\n The model was trained over 4 epochs with a learning rate of 2e-05,\
|
42 |
-
\ using a batch size of 64. \n The training utilized a cross-entropy loss\
|
43 |
-
\ function and the AdamW optimizer, with gradient accumulation over 4 steps.\n \
|
44 |
-
\ "
|
45 |
-
tips_and_tricks: "\n For optimal performance, questions should be clear, concise,\
|
46 |
-
\ and grammatically correct. \n The model performs best on questions related\
|
47 |
-
\ to topics covered in the squad_v2 dataset. \n It is advisable to pre-process\
|
48 |
-
\ text for consistency in encoding and punctuation, and to manage expectations for\
|
49 |
-
\ questions on topics outside the training data.\n "
|
50 |
model-index:
|
51 |
-
- name: roberta-finetuned
|
52 |
-
results:
|
53 |
-
- task:
|
54 |
-
type: question-answering
|
55 |
-
dataset:
|
56 |
-
name: SQuAD v2
|
57 |
-
type: squad_v2
|
58 |
-
metrics:
|
59 |
-
- type: Exact
|
60 |
-
value: 100.0
|
61 |
-
- type: F1
|
62 |
-
value: 100.0
|
63 |
-
- type: Total
|
64 |
-
value: 2
|
65 |
-
- type: Hasans Exact
|
66 |
-
value: 100.0
|
67 |
-
- type: Hasans F1
|
68 |
-
value: 100.0
|
69 |
-
- type: Hasans Total
|
70 |
-
value: 2
|
71 |
-
- type: Best Exact
|
72 |
-
value: 100.0
|
73 |
-
- type: Best Exact Thresh
|
74 |
-
value: 0.9603068232536316
|
75 |
-
- type: Best F1
|
76 |
-
value: 100.0
|
77 |
-
- type: Best F1 Thresh
|
78 |
-
value: 0.9603068232536316
|
79 |
-
- type: Total Time In Seconds
|
80 |
-
value: 0.034005987999989884
|
81 |
-
- type: Samples Per Second
|
82 |
-
value: 58.813171374423675
|
83 |
-
- type: Latency In Seconds
|
84 |
-
value: 0.017002993999994942
|
85 |
---
|
86 |
|
87 |
-
|
|
|
88 |
|
89 |
-
|
90 |
|
|
|
|
|
|
|
91 |
|
|
|
92 |
|
93 |
-
|
94 |
|
95 |
-
|
96 |
|
97 |
-
|
98 |
|
|
|
99 |
|
|
|
100 |
|
101 |
-
|
102 |
-
- **Shared by [optional]:** [More Information Needed]
|
103 |
-
- **Model type:** [More Information Needed]
|
104 |
-
- **Language(s) (NLP):** en
|
105 |
-
- **License:** mit
|
106 |
-
- **Finetuned from model [optional]:** [More Information Needed]
|
107 |
|
108 |
-
###
|
109 |
|
110 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
111 |
|
112 |
-
|
113 |
-
- **Paper [optional]:** [More Information Needed]
|
114 |
-
- **Demo [optional]:** [More Information Needed]
|
115 |
|
116 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
|
118 |
-
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
|
119 |
-
|
120 |
-
### Direct Use
|
121 |
-
|
122 |
-
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
|
123 |
-
|
124 |
-
[More Information Needed]
|
125 |
-
|
126 |
-
### Downstream Use [optional]
|
127 |
-
|
128 |
-
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
|
129 |
-
|
130 |
-
[More Information Needed]
|
131 |
-
|
132 |
-
### Out-of-Scope Use
|
133 |
-
|
134 |
-
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
|
135 |
-
|
136 |
-
[More Information Needed]
|
137 |
-
|
138 |
-
## Bias, Risks, and Limitations
|
139 |
-
|
140 |
-
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
|
141 |
-
|
142 |
-
[More Information Needed]
|
143 |
-
|
144 |
-
### Recommendations
|
145 |
-
|
146 |
-
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
|
147 |
-
|
148 |
-
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
|
149 |
-
|
150 |
-
## How to Get Started with the Model
|
151 |
-
|
152 |
-
Use the code below to get started with the model.
|
153 |
-
|
154 |
-
[More Information Needed]
|
155 |
-
|
156 |
-
## Training Details
|
157 |
-
|
158 |
-
### Training Data
|
159 |
-
|
160 |
-
<!-- This should link to a Data Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
|
161 |
-
|
162 |
-
[More Information Needed]
|
163 |
-
|
164 |
-
### Training Procedure
|
165 |
-
|
166 |
-
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
|
167 |
-
|
168 |
-
#### Preprocessing [optional]
|
169 |
-
|
170 |
-
[More Information Needed]
|
171 |
-
|
172 |
-
|
173 |
-
#### Training Hyperparameters
|
174 |
-
|
175 |
-
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
|
176 |
-
|
177 |
-
#### Speeds, Sizes, Times [optional]
|
178 |
-
|
179 |
-
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
|
180 |
-
|
181 |
-
[More Information Needed]
|
182 |
-
|
183 |
-
## Evaluation
|
184 |
-
|
185 |
-
<!-- This section describes the evaluation protocols and provides the results. -->
|
186 |
-
|
187 |
-
### Testing Data, Factors & Metrics
|
188 |
-
|
189 |
-
#### Testing Data
|
190 |
-
|
191 |
-
<!-- This should link to a Data Card if possible. -->
|
192 |
-
|
193 |
-
[More Information Needed]
|
194 |
-
|
195 |
-
#### Factors
|
196 |
-
|
197 |
-
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
|
198 |
-
|
199 |
-
[More Information Needed]
|
200 |
-
|
201 |
-
#### Metrics
|
202 |
-
|
203 |
-
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
204 |
-
|
205 |
-
[More Information Needed]
|
206 |
-
|
207 |
-
### Results
|
208 |
-
|
209 |
-
[More Information Needed]
|
210 |
-
|
211 |
-
#### Summary
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
## Model Examination [optional]
|
216 |
-
|
217 |
-
<!-- Relevant interpretability work for the model goes here -->
|
218 |
-
|
219 |
-
[More Information Needed]
|
220 |
-
|
221 |
-
## Environmental Impact
|
222 |
-
|
223 |
-
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
|
224 |
-
|
225 |
-
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
226 |
-
|
227 |
-
- **Hardware Type:** [More Information Needed]
|
228 |
-
- **Hours used:** [More Information Needed]
|
229 |
-
- **Cloud Provider:** [More Information Needed]
|
230 |
-
- **Compute Region:** [More Information Needed]
|
231 |
-
- **Carbon Emitted:** [More Information Needed]
|
232 |
-
|
233 |
-
## Technical Specifications [optional]
|
234 |
-
|
235 |
-
### Model Architecture and Objective
|
236 |
-
|
237 |
-
[More Information Needed]
|
238 |
-
|
239 |
-
### Compute Infrastructure
|
240 |
-
|
241 |
-
[More Information Needed]
|
242 |
-
|
243 |
-
#### Hardware
|
244 |
-
|
245 |
-
[More Information Needed]
|
246 |
-
|
247 |
-
#### Software
|
248 |
-
|
249 |
-
[More Information Needed]
|
250 |
-
|
251 |
-
## Citation [optional]
|
252 |
-
|
253 |
-
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
254 |
-
|
255 |
-
**BibTeX:**
|
256 |
-
|
257 |
-
[More Information Needed]
|
258 |
-
|
259 |
-
**APA:**
|
260 |
-
|
261 |
-
[More Information Needed]
|
262 |
-
|
263 |
-
## Glossary [optional]
|
264 |
-
|
265 |
-
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
266 |
-
|
267 |
-
[More Information Needed]
|
268 |
-
|
269 |
-
## More Information [optional]
|
270 |
-
|
271 |
-
[More Information Needed]
|
272 |
-
|
273 |
-
## Model Card Authors [optional]
|
274 |
-
|
275 |
-
[More Information Needed]
|
276 |
-
|
277 |
-
## Model Card Contact
|
278 |
-
|
279 |
-
[More Information Needed]
|
280 |
|
|
|
281 |
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
tags:
|
3 |
+
- generated_from_trainer
|
4 |
+
datasets:
|
5 |
+
- squad_v2
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
model-index:
|
7 |
+
- name: roberta-finetuned-squad_v2
|
8 |
+
results: []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
---
|
10 |
|
11 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
12 |
+
should probably proofread and complete it, then remove this comment. -->
|
13 |
|
14 |
+
# roberta-finetuned-squad_v2
|
15 |
|
16 |
+
This model was trained from scratch on the squad_v2 dataset.
|
17 |
+
It achieves the following results on the evaluation set:
|
18 |
+
- Loss: 0.8582
|
19 |
|
20 |
+
## Model description
|
21 |
|
22 |
+
More information needed
|
23 |
|
24 |
+
## Intended uses & limitations
|
25 |
|
26 |
+
More information needed
|
27 |
|
28 |
+
## Training and evaluation data
|
29 |
|
30 |
+
More information needed
|
31 |
|
32 |
+
## Training procedure
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
+
### Training hyperparameters
|
35 |
|
36 |
+
The following hyperparameters were used during training:
|
37 |
+
- learning_rate: 2e-05
|
38 |
+
- train_batch_size: 128
|
39 |
+
- eval_batch_size: 128
|
40 |
+
- seed: 42
|
41 |
+
- gradient_accumulation_steps: 4
|
42 |
+
- total_train_batch_size: 512
|
43 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
44 |
+
- lr_scheduler_type: linear
|
45 |
+
- num_epochs: 4
|
46 |
|
47 |
+
### Training results
|
|
|
|
|
48 |
|
49 |
+
| Training Loss | Epoch | Step | Validation Loss |
|
50 |
+
|:-------------:|:-----:|:----:|:---------------:|
|
51 |
+
| 2.9129 | 0.2 | 100 | 1.4700 |
|
52 |
+
| 1.4395 | 0.39 | 200 | 1.2407 |
|
53 |
+
| 1.2356 | 0.59 | 300 | 1.0325 |
|
54 |
+
| 1.1284 | 0.78 | 400 | 0.9750 |
|
55 |
+
| 1.0821 | 0.98 | 500 | 0.9345 |
|
56 |
+
| 0.9978 | 1.18 | 600 | 0.9893 |
|
57 |
+
| 0.9697 | 1.37 | 700 | 0.9300 |
|
58 |
+
| 0.9455 | 1.57 | 800 | 0.9351 |
|
59 |
+
| 0.9322 | 1.76 | 900 | 0.9451 |
|
60 |
+
| 0.9269 | 1.96 | 1000 | 0.9064 |
|
61 |
+
| 0.9105 | 2.16 | 1100 | 0.8837 |
|
62 |
+
| 0.8805 | 2.35 | 1200 | 0.8876 |
|
63 |
+
| 0.8703 | 2.55 | 1300 | 0.9853 |
|
64 |
+
| 0.8699 | 2.75 | 1400 | 0.9235 |
|
65 |
+
| 0.8633 | 2.94 | 1500 | 0.8930 |
|
66 |
+
| 0.828 | 3.14 | 1600 | 0.8582 |
|
67 |
+
| 0.8284 | 3.33 | 1700 | 0.9203 |
|
68 |
+
| 0.8076 | 3.53 | 1800 | 0.8866 |
|
69 |
+
| 0.7805 | 3.73 | 1900 | 0.9099 |
|
70 |
+
| 0.7974 | 3.92 | 2000 | 0.8746 |
|
71 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
72 |
|
73 |
+
### Framework versions
|
74 |
|
75 |
+
- Transformers 4.34.1
|
76 |
+
- Pytorch 2.1.0+cu118
|
77 |
+
- Datasets 2.14.5
|
78 |
+
- Tokenizers 0.14.1
|
metrics.json
CHANGED
@@ -1,15 +1,15 @@
|
|
1 |
{
|
2 |
-
"exact":
|
3 |
-
"f1":
|
4 |
-
"total":
|
5 |
-
"HasAns_exact":
|
6 |
-
"HasAns_f1":
|
7 |
-
"HasAns_total":
|
8 |
-
"
|
9 |
-
"
|
10 |
-
"
|
11 |
-
"
|
12 |
-
"
|
13 |
-
"
|
14 |
-
"
|
15 |
}
|
|
|
1 |
{
|
2 |
+
"exact": 100.0,
|
3 |
+
"f1": 100.0,
|
4 |
+
"total": 2,
|
5 |
+
"HasAns_exact": 100.0,
|
6 |
+
"HasAns_f1": 100.0,
|
7 |
+
"HasAns_total": 2,
|
8 |
+
"best_exact": 100.0,
|
9 |
+
"best_exact_thresh": 0.9603068232536316,
|
10 |
+
"best_f1": 100.0,
|
11 |
+
"best_f1_thresh": 0.9603068232536316,
|
12 |
+
"total_time_in_seconds": 0.034005987999989884,
|
13 |
+
"samples_per_second": 58.813171374423675,
|
14 |
+
"latency_in_seconds": 0.017002993999994942
|
15 |
}
|
trainer_state.json
CHANGED
@@ -292,17 +292,17 @@
|
|
292 |
"epoch": 4.0,
|
293 |
"step": 2040,
|
294 |
"total_flos": 1.3645021155456614e+17,
|
295 |
-
"train_loss": 0.
|
296 |
-
"train_runtime":
|
297 |
-
"train_samples_per_second":
|
298 |
-
"train_steps_per_second":
|
299 |
},
|
300 |
{
|
301 |
"epoch": 4.0,
|
302 |
"eval_loss": 0.8582048416137695,
|
303 |
-
"eval_runtime": 17.
|
304 |
-
"eval_samples_per_second":
|
305 |
-
"eval_steps_per_second": 10.
|
306 |
"step": 2040
|
307 |
}
|
308 |
],
|
|
|
292 |
"epoch": 4.0,
|
293 |
"step": 2040,
|
294 |
"total_flos": 1.3645021155456614e+17,
|
295 |
+
"train_loss": 0.015366485072117226,
|
296 |
+
"train_runtime": 57.0805,
|
297 |
+
"train_samples_per_second": 9148.547,
|
298 |
+
"train_steps_per_second": 35.739
|
299 |
},
|
300 |
{
|
301 |
"epoch": 4.0,
|
302 |
"eval_loss": 0.8582048416137695,
|
303 |
+
"eval_runtime": 17.613,
|
304 |
+
"eval_samples_per_second": 678.761,
|
305 |
+
"eval_steps_per_second": 10.617,
|
306 |
"step": 2040
|
307 |
}
|
308 |
],
|
training_args.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 4664
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:79779cf4e7d8a5fdc99b0dc402459aacd75bad5cb5b42f73d24b20e7d7034ed4
|
3 |
size 4664
|