| | --- |
| | license: apache-2.0 |
| | library_name: peft |
| | tags: |
| | - trl |
| | - sft |
| | - generated_from_trainer |
| | - Pytorch |
| | - Falcon |
| | base_model: tiiuae/falcon-7b |
| | model-index: |
| | - name: saqr-7b-instruct |
| | results: [] |
| | datasets: |
| | - HuggingFaceH4/ultrachat_200k |
| | - openbmb/UltraFeedback |
| | - gsm8k |
| | language: |
| | - en |
| | pipeline_tag: text-generation |
| | --- |
| | |
| | <img src="https://huggingface.co/Menouar/saqr-7b-instruct/resolve/main/saqr.jpg" alt="Saqr Logo" width="800" style="margin-left:auto; margin-right:auto; display:block;"/> |
| |
|
| |
|
| | # saqr-7b-instruct |
| |
|
| | This model is a fine-tuned version of [tiiuae/falcon-7b](https://huggingface.co/tiiuae/falcon-7b) on [ultrachat_200k](https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k), [UltraFeedback](https://huggingface.co/datasets/openbmb/UltraFeedback), and [gsm8k](https://huggingface.co/datasets/gsm8k) datasets. |
| |
|
| | ## Model description |
| |
|
| | This model is a fine-tuned version of [tiiuae/falcon-7b](https://huggingface.co/tiiuae/falcon-7b) using supervised fine-tuning on nearly the same datasets as Zephyr-7B-beta. |
| |
|
| | ## Training and evaluation data |
| |
|
| | The evaluation for training can be found [here](https://huggingface.co/Menouar/saqr-7b-instruct/tensorboard). |
| |
|
| | The evaluation can be found at the Hugging Face Leaderboard [here](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Menouar/saqr-7b-instruct/). |
| |
|
| |
|
| | ## Training procedure |
| |
|
| | Can be found [here](https://colab.research.google.com/github/menouarazib/llm/blob/main/Saqr_7B.ipynb). |
| |
|
| | ### Training hyperparameters |
| |
|
| | The following hyperparameters were used during training: |
| | - learning_rate: 0.0002 |
| | - train_batch_size: 7 |
| | - eval_batch_size: 8 |
| | - seed: 42 |
| | - gradient_accumulation_steps: 2 |
| | - total_train_batch_size: 14 |
| | - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
| | - lr_scheduler_type: constant |
| | - lr_scheduler_warmup_ratio: 0.03 |
| | - training_steps: 5000 |
| |
|
| | ### Training results |
| |
|
| |
|
| |
|
| | ### Framework versions |
| |
|
| | - PEFT 0.8.2 |
| | - Transformers 4.38.0.dev0 |
| | - Pytorch 2.1.0+cu121 |
| | - Datasets 2.17.0 |
| | - Tokenizers 0.15.1 |
