metadata
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 using the dataset (unspecified).
The 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
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