|
|
--- |
|
|
library_name: peft |
|
|
license: apache-2.0 |
|
|
base_model: mistralai/Mistral-7B-Instruct-v0.3 |
|
|
tags: |
|
|
- generated_from_trainer |
|
|
model-index: |
|
|
- name: resultados |
|
|
results: [] |
|
|
language: |
|
|
- es |
|
|
--- |
|
|
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
|
|
# resultados |
|
|
|
|
|
Este es un modelo afinado sobre [mistralai/Mistral-7B-Instruct-v0.3](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3) |
|
|
sobre un dataset de preguntas-respuestas del acuerdo 009 con un batch_size=1 y 10 épocas |
|
|
con un uso total de VRAM de la GPU de 24 Gz, |
|
|
logrando una pérdida de: |
|
|
- Loss: 0.2677 |
|
|
|
|
|
## Model description |
|
|
|
|
|
Este modelo se está usando para un sistema de chatbot que responde a preguntas sobre el acuerdo 009,por ahora se está usando el modelo para probarlo al estilo [Arena chatbot](https://lmarena.ai/) |
|
|
con el fin de medir el rendimiento como un chat directo y se está probando el uso de RAG para responder a preguntas |
|
|
con dcocumentos actuales. |
|
|
|
|
|
## Intended uses & limitations |
|
|
|
|
|
More information needed |
|
|
|
|
|
## Training and evaluation data |
|
|
|
|
|
Este modelo se genera del entrenamiento del modelo "mistralai/Mistral-7B-Instruct-v0.3" sobre un dataset de preguntas respuestas sobre el acuerdo 009 de la Universidad |
|
|
del valle con 1700 ejemplos. El dataset fue realizado por estudiantes del programa de Ingeniería de Sistemas con la |
|
|
ayuda de la vicedecanatura académica de ingeniería. |
|
|
|
|
|
## Training procedure |
|
|
|
|
|
### Training hyperparameters |
|
|
|
|
|
The following hyperparameters were used during training: |
|
|
- learning_rate: 2e-05 |
|
|
- train_batch_size: 1 |
|
|
- eval_batch_size: 8 |
|
|
- seed: 42 |
|
|
- gradient_accumulation_steps: 4 |
|
|
- total_train_batch_size: 4 |
|
|
- optimizer: Use paged_adamw_8bit with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments |
|
|
- lr_scheduler_type: linear |
|
|
- num_epochs: 10 |
|
|
- mixed_precision_training: Native AMP |
|
|
|
|
|
### Training results |
|
|
|
|
|
| Training Loss | Epoch | Step | Validation Loss | |
|
|
|:-------------:|:------:|:----:|:---------------:| |
|
|
| 1.2627 | 0.2694 | 100 | 1.2111 | |
|
|
| 1.0079 | 0.5387 | 200 | 1.0255 | |
|
|
| 0.8729 | 0.8081 | 300 | 0.8972 | |
|
|
| 0.7103 | 1.0754 | 400 | 0.8024 | |
|
|
| 0.6555 | 1.3448 | 500 | 0.7070 | |
|
|
| 0.5711 | 1.6141 | 600 | 0.6281 | |
|
|
| 0.6438 | 1.8835 | 700 | 0.5783 | |
|
|
| 0.5111 | 2.1508 | 800 | 0.5160 | |
|
|
| 0.4312 | 2.4202 | 900 | 0.4764 | |
|
|
| 0.4467 | 2.6896 | 1000 | 0.4446 | |
|
|
| 0.4222 | 2.9589 | 1100 | 0.4124 | |
|
|
| 0.3802 | 3.2263 | 1200 | 0.3931 | |
|
|
| 0.2767 | 3.4956 | 1300 | 0.3718 | |
|
|
| 0.3598 | 3.7650 | 1400 | 0.3577 | |
|
|
| 0.2838 | 4.0323 | 1500 | 0.3447 | |
|
|
| 0.3169 | 4.3017 | 1600 | 0.3349 | |
|
|
| 0.2737 | 4.5710 | 1700 | 0.3273 | |
|
|
| 0.2425 | 4.8404 | 1800 | 0.3138 | |
|
|
| 0.1814 | 5.1077 | 1900 | 0.3092 | |
|
|
| 0.2372 | 5.3771 | 2000 | 0.3004 | |
|
|
| 0.258 | 5.6465 | 2100 | 0.2953 | |
|
|
| 0.2488 | 5.9158 | 2200 | 0.2911 | |
|
|
| 0.2052 | 6.1832 | 2300 | 0.2926 | |
|
|
| 0.1973 | 6.4525 | 2400 | 0.2929 | |
|
|
| 0.2595 | 6.7219 | 2500 | 0.2828 | |
|
|
| 0.1843 | 6.9912 | 2600 | 0.2771 | |
|
|
| 0.1912 | 7.2586 | 2700 | 0.2784 | |
|
|
| 0.2303 | 7.5279 | 2800 | 0.2777 | |
|
|
| 0.2396 | 7.7973 | 2900 | 0.2697 | |
|
|
| 0.2031 | 8.0646 | 3000 | 0.2708 | |
|
|
| 0.1567 | 8.3340 | 3100 | 0.2730 | |
|
|
| 0.1605 | 8.6034 | 3200 | 0.2690 | |
|
|
| 0.1741 | 8.8727 | 3300 | 0.2674 | |
|
|
| 0.1727 | 9.1401 | 3400 | 0.2709 | |
|
|
| 0.1779 | 9.4094 | 3500 | 0.2666 | |
|
|
| 0.1469 | 9.6788 | 3600 | 0.2687 | |
|
|
| 0.1967 | 9.9481 | 3700 | 0.2677 | |
|
|
|
|
|
|
|
|
### Framework versions |
|
|
|
|
|
- PEFT 0.15.1 |
|
|
- Transformers 4.51.0 |
|
|
- Pytorch 2.6.0+cu126 |
|
|
- Datasets 3.5.0 |
|
|
- Tokenizers 0.21.1 |
