Fine-tune of Yi-34B with Spicyboros-3.1
Three epochs of fine tuning with @jondurbin's SpicyBoros-3.1 dataset. 4.65bpw should fit on a single 3090/4090, 5.0bpw, 6.0bpw, and 8.0bpw will require more than one GPU 24 GB VRAM GPU.
Please note: you may have to turn down repetition penalty to 1.0. The model seems to get into "thesaurus" mode sometimes without this change.
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: 0.0001
- train_batch_size: 6
- eval_batch_size: 6
- seed: 42
- distributed_type: multi-GPU
- num_devices: 2
- gradient_accumulation_steps: 2
- total_train_batch_size: 24
- total_eval_batch_size: 12
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant
- num_epochs: 3
Training results
Framework versions
- Transformers 4.34.1
- Pytorch 2.0.1+cu118
- Datasets 2.14.6
- Tokenizers 0.14.1
- Downloads last month
- 2