File size: 3,480 Bytes
91f21c0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
---
license: mit
base_model: sheepy928/default
tags:
- generated_from_trainer
metrics:
- accuracy
- f1
- recall
- precision
model-index:
- name: FT_3
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# FT_3

This model is a fine-tuned version of [sheepy928/default](https://huggingface.co/sheepy928/default) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.7570
- Accuracy: 0.7393
- F1: 0.6292
- Recall: 0.7393
- Precision: 0.7147
- Combined Score: 0.7056

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.0005
- train_batch_size: 8
- eval_batch_size: 32
- seed: 42
- distributed_type: multi-GPU
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 10000
- num_epochs: 30

### Training results

| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1     | Recall | Precision | Combined Score |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:------:|:------:|:---------:|:--------------:|
| 0.6785        | 1.6   | 300  | 0.7930          | 0.7387   | 0.6276 | 0.7387 | 0.5456    | 0.6627         |
| 0.5583        | 3.19  | 600  | 0.6910          | 0.7613   | 0.7316 | 0.7613 | 0.7072    | 0.7404         |
| 0.7857        | 4.79  | 900  | 0.6515          | 0.7387   | 0.6276 | 0.7387 | 0.5456    | 0.6627         |
| 0.6309        | 6.38  | 1200 | 0.5592          | 0.848    | 0.8270 | 0.848  | 0.8382    | 0.8403         |
| 0.2216        | 7.98  | 1500 | 0.5708          | 0.8773   | 0.8432 | 0.8773 | 0.8496    | 0.8619         |
| 0.3214        | 9.57  | 1800 | 0.4550          | 0.896    | 0.8584 | 0.896  | 0.8776    | 0.8820         |
| 0.7521        | 11.17 | 2100 | 0.3819          | 0.884    | 0.8423 | 0.884  | 0.8059    | 0.8541         |
| 0.5048        | 12.77 | 2400 | 0.6582          | 0.7387   | 0.6276 | 0.7387 | 0.5456    | 0.6627         |
| 0.6435        | 14.36 | 2700 | 0.5365          | 0.8467   | 0.8092 | 0.8467 | 0.7798    | 0.8206         |
| 0.9304        | 15.96 | 3000 | 0.7577          | 0.7387   | 0.6289 | 0.7387 | 0.6302    | 0.6841         |
| 0.7902        | 17.55 | 3300 | 0.7684          | 0.7387   | 0.6289 | 0.7387 | 0.6302    | 0.6841         |
| 0.6364        | 19.15 | 3600 | 0.7638          | 0.7387   | 0.6289 | 0.7387 | 0.6302    | 0.6841         |
| 0.6738        | 20.74 | 3900 | 0.7769          | 0.7393   | 0.6292 | 0.7393 | 0.7147    | 0.7056         |
| 0.8142        | 22.34 | 4200 | 0.7443          | 0.7393   | 0.6292 | 0.7393 | 0.7147    | 0.7056         |
| 0.8184        | 23.94 | 4500 | 0.7635          | 0.7393   | 0.6292 | 0.7393 | 0.7147    | 0.7056         |
| 0.7562        | 25.53 | 4800 | 0.7467          | 0.7393   | 0.6292 | 0.7393 | 0.7147    | 0.7056         |
| 0.5699        | 27.13 | 5100 | 0.7867          | 0.7393   | 0.6292 | 0.7393 | 0.7147    | 0.7056         |
| 0.761         | 28.72 | 5400 | 0.7570          | 0.7393   | 0.6292 | 0.7393 | 0.7147    | 0.7056         |


### Framework versions

- Transformers 4.34.1
- Pytorch 2.1.0
- Datasets 2.14.5
- Tokenizers 0.14.1