metadata
license: apache-2.0
base_model: ondevicellm/tinyllama_moe
tags:
- alignment-handbook
- generated_from_trainer
- trl
- sft
- generated_from_trainer
datasets:
- HuggingFaceH4/ultrachat_200k
model-index:
- name: tinyllama_moe_sft_ultrachat200k_v2
results: []
tinyllama_moe_sft_ultrachat200k_v2
This model is a fine-tuned version of ondevicellm/tinyllama_moe on the HuggingFaceH4/ultrachat_200k dataset. It achieves the following results on the evaluation set:
- Loss: 1.1593
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: 2e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- total_eval_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 1
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 1.336 | 0.09 | 100 | 1.3140 |
| 1.2426 | 0.18 | 200 | 1.2376 |
| 1.2083 | 0.26 | 300 | 1.2100 |
| 1.1862 | 0.35 | 400 | 1.1934 |
| 1.1567 | 0.44 | 500 | 1.1820 |
| 1.1777 | 0.53 | 600 | 1.1737 |
| 1.1666 | 0.61 | 700 | 1.1677 |
| 1.1531 | 0.7 | 800 | 1.1636 |
| 1.1525 | 0.79 | 900 | 1.1610 |
| 1.1396 | 0.88 | 1000 | 1.1596 |
| 1.1681 | 0.96 | 1100 | 1.1593 |
Framework versions
- Transformers 4.36.2
- Pytorch 2.1.2+cu118
- Datasets 2.14.6
- Tokenizers 0.15.0