File size: 2,864 Bytes
6d140c7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
88
---
base_model: Qwen/Qwen2-0.5B-Instruct
library_name: distily
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: distily_experiments_loss_kl
  results: []
---

# distily_experiments_loss_kl

This student model is distilled from the teacher model [Qwen/Qwen2-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2-0.5B-Instruct) using the dataset (unspecified).

The [Distily](https://github.com/lapp0/distily) library was used for this distillation.

It achieves the following results on the evaluation set:
- eval_enwikippl: nan
- eval_frwikippl: nan
- eval_zhwikippl: nan
- eval_loss: nan
- eval_runtime: 90.8319
- eval_samples_per_second: 11.009
- eval_steps_per_second: 2.752

<!-- 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.

## 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:
- distillation_strategy: logits_activations
- loss_fn: kl
- train_embeddings: True
- learning_rate: 4e-05
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant
- num_epochs: 1.0

### Resource Usage
Peak GPU Memory: 19.8804 GB

### Eval-Phase Metrics
| step | epoch | enwikippl | frwikippl | loss | runtime | samples_per_second | steps_per_second | zhwikippl |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| **teacher eval** |  | 13.0697 | 11.6518 |  |  |  |  | 21.6262 |
| 0 | 0 | 179854.0938 | 180181.3125 | 782.5541 | 90.5683 | 11.041 | 2.76 | 180786.8438 |
| 500 | 0.0808 | nan | nan | nan | 90.7978 | 11.013 | 2.753 | nan |
| 1000 | 0.1616 | nan | nan | nan | 91.1195 | 10.975 | 2.744 | nan |
| 1500 | 0.2424 | nan | nan | nan | 90.9106 | 11.0 | 2.75 | nan |
| 2000 | 0.3232 | nan | nan | nan | 90.7616 | 11.018 | 2.754 | nan |
| 2500 | 0.4040 | nan | nan | nan | 90.73 | 11.022 | 2.755 | nan |
| 3000 | 0.4848 | nan | nan | nan | 90.7247 | 11.022 | 2.756 | nan |
| 3500 | 0.5657 | nan | nan | nan | 90.5701 | 11.041 | 2.76 | nan |
| 4000 | 0.6465 | nan | nan | nan | 90.768 | 11.017 | 2.754 | nan |
| 4500 | 0.7273 | nan | nan | nan | 90.8206 | 11.011 | 2.753 | nan |
| 5000 | 0.8081 | nan | nan | nan | 90.7225 | 11.023 | 2.756 | nan |
| 5500 | 0.8889 | nan | nan | nan | 90.7674 | 11.017 | 2.754 | nan |
| 6000 | 0.9697 | nan | nan | nan | 90.7133 | 11.024 | 2.756 | nan |
| 6187 | 0.9999 | nan | nan | nan | 90.8319 | 11.009 | 2.752 | nan |

### Framework versions
- Distily 0.1.0
- Transformers 4.43.3
- Pytorch 2.3.0
- Datasets 2.20.0