YAML Metadata
Warning:
empty or missing yaml metadata in repo card
(https://huggingface.co/docs/hub/model-cards#model-card-metadata)
Quantization made by Richard Erkhov.
phi3.5-pro-10-08 - bnb 8bits
- Model creator: https://huggingface.co/win10/
- Original model: https://huggingface.co/win10/phi3.5-pro-10-08/
Original model description:
library_name: transformers license: other tags: - freeze model-index: - name: phi3.5-pro-10-08 results: []
phi3.5-pro-10-08
This model is a fine-tuned version of E:\mergekit\phi3.5-pro on the Magpie_Qwen2_Pro_300k and the Magpie_Llama_3.1_Pro_MT_300k datasets. It achieves the following results on the evaluation set:
- Loss: 0.8538
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: 2
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- training_steps: 1000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.7784 | 0.0080 | 100 | 0.8538 |
| 0.8605 | 0.0160 | 200 | 0.8538 |
| 0.8386 | 0.0240 | 300 | 0.8538 |
| 0.8201 | 0.0320 | 400 | 0.8538 |
| 0.8661 | 0.0400 | 500 | 0.8538 |
| 0.776 | 0.0480 | 600 | 0.8538 |
| 0.8377 | 0.0561 | 700 | 0.8538 |
| 0.8541 | 0.0641 | 800 | 0.8538 |
| 0.799 | 0.0721 | 900 | 0.8538 |
| 0.8146 | 0.0801 | 1000 | 0.8538 |
Framework versions
- Transformers 4.45.0
- Pytorch 2.4.0+cu124
- Datasets 2.21.0
- Tokenizers 0.20.0
- Downloads last month
- 0