|
|
--- |
|
|
base_model: Qwen/Qwen2-0.5B-Instruct |
|
|
library_name: distily |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- generated_from_trainer |
|
|
model-index: |
|
|
- name: distily_experiments_loss_reverse_kl |
|
|
results: [] |
|
|
--- |
|
|
|
|
|
# distily_experiments_loss_reverse_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: 2760.3779 |
|
|
- eval_frwikippl: 28158.2578 |
|
|
- eval_zhwikippl: 441247.4688 |
|
|
- eval_loss: 3.1654 |
|
|
- eval_runtime: 90.8509 |
|
|
- eval_samples_per_second: 11.007 |
|
|
- 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: reverse_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: 19.8832 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 | 180187.8438 | 182062.6875 | 131.8108 | 90.6539 | 11.031 | 2.758 | 181762.375 | |
|
|
| 500 | 0.0808 | 14699.2041 | 52797.9922 | 6.0418 | 90.8884 | 11.003 | 2.751 | 371252.0312 | |
|
|
| 1000 | 0.1616 | 8812.4561 | 47709.9297 | 4.9882 | 90.8533 | 11.007 | 2.752 | 384212.3438 | |
|
|
| 1500 | 0.2424 | 7321.3081 | 44922.375 | 4.6195 | 90.7179 | 11.023 | 2.756 | 400192.5625 | |
|
|
| 2000 | 0.3232 | 6277.4165 | 42254.6719 | 4.2012 | 90.8257 | 11.01 | 2.753 | 423631.0938 | |
|
|
| 2500 | 0.4040 | 5452.0264 | 39927.7812 | 3.9955 | 90.7803 | 11.016 | 2.754 | 445022.5938 | |
|
|
| 3000 | 0.4848 | 4708.5049 | 37660.8359 | 3.7784 | 90.8232 | 11.01 | 2.753 | 447453.4375 | |
|
|
| 3500 | 0.5657 | 4329.6147 | 35350.4805 | 3.6816 | 90.8654 | 11.005 | 2.751 | 455292.8125 | |
|
|
| 4000 | 0.6465 | 3840.0864 | 33493.6836 | 3.5800 | 90.7858 | 11.015 | 2.754 | 446474.3125 | |
|
|
| 4500 | 0.7273 | 3495.4482 | 31764.3340 | 3.4447 | 90.8083 | 11.012 | 2.753 | 447611.3438 | |
|
|
| 5000 | 0.8081 | 3245.5376 | 30812.8379 | 3.3323 | 90.7976 | 11.014 | 2.753 | 448982.8438 | |
|
|
| 5500 | 0.8889 | 3057.9595 | 29516.0742 | 3.2926 | 90.7385 | 11.021 | 2.755 | 459842.8125 | |
|
|
| 6000 | 0.9697 | 2831.3643 | 28517.0625 | 3.1956 | 90.7677 | 11.017 | 2.754 | 441979.4375 | |
|
|
| 6187 | 0.9999 | 2760.3779 | 28158.2578 | 3.1654 | 90.8509 | 11.007 | 2.752 | 441247.4688 | |
|
|
|
|
|
### Framework versions |
|
|
- Distily 0.1.0 |
|
|
- Transformers 4.43.3 |
|
|
- Pytorch 2.3.0 |
|
|
- Datasets 2.20.0 |
|
|
|
