|
--- |
|
license: other |
|
base_model: NousResearch/Meta-Llama-3-8B |
|
tags: |
|
- generated_from_trainer |
|
model-index: |
|
- name: out-llama8b-alpaca-data-pt-br |
|
results: [] |
|
--- |
|
|
|
<!-- 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. --> |
|
|
|
[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl) |
|
<details><summary>See axolotl config</summary> |
|
|
|
axolotl version: `0.4.0` |
|
```yaml |
|
base_model: NousResearch/Meta-Llama-3-8B |
|
model_type: LlamaForCausalLM |
|
tokenizer_type: AutoTokenizer |
|
|
|
load_in_8bit: false |
|
load_in_4bit: false |
|
strict: false |
|
|
|
datasets: |
|
- path: dominguesm/alpaca-data-pt-br |
|
type: alpaca |
|
dataset_prepared_path: last_run_prepared |
|
val_set_size: 0.05 |
|
output_dir: ./out-llama8b-alpaca-data-pt-br |
|
|
|
sequence_len: 8192 |
|
sample_packing: true |
|
pad_to_sequence_len: true |
|
|
|
wandb_project: meta-llama-8b-alpacadata-br |
|
wandb_entity: |
|
wandb_watch: |
|
wandb_name: |
|
wandb_log_model: |
|
|
|
gradient_accumulation_steps: 8 |
|
micro_batch_size: 1 |
|
num_epochs: 2 |
|
optimizer: paged_adamw_8bit |
|
lr_scheduler: cosine |
|
learning_rate: 2e-5 |
|
|
|
train_on_inputs: false |
|
group_by_length: false |
|
bf16: auto |
|
fp16: |
|
tf32: false |
|
|
|
gradient_checkpointing: true |
|
gradient_checkpointing_kwargs: |
|
use_reentrant: false |
|
early_stopping_patience: |
|
resume_from_checkpoint: |
|
logging_steps: 1 |
|
xformers_attention: |
|
flash_attention: true |
|
|
|
warmup_steps: 100 |
|
evals_per_epoch: 2 |
|
eval_table_size: |
|
saves_per_epoch: 1 |
|
debug: |
|
deepspeed: |
|
weight_decay: 0.0 |
|
fsdp: |
|
fsdp_config: |
|
special_tokens: |
|
pad_token: <|end_of_text|> |
|
|
|
``` |
|
|
|
</details><br> |
|
|
|
# LLama 3- 8B -alpaca-data-pt-br |
|
|
|
Thanks to [Redmond.ai](https://redmond.ai) for the GPU Support! |
|
|
|
This model is a fine-tuned version of [NousResearch/Meta-Llama-3-8B](https://huggingface.co/NousResearch/Meta-Llama-3-8B) on the [dominguesm/alpaca-data-pt-br](https://huggingface.co/dominguesm/alpaca-data-pt-br) dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 1.1227 |
|
|
|
## Model description |
|
|
|
The model is a Portuguese language understanding model designed to generate responses to a wide range of questions and prompts. It takes as input a natural language question or prompt and outputs a corresponding response. |
|
|
|
The model is trained on a dataset of 51k examples, which is a cleaned and translated version of the original Alpaca Dataset released by Stanford. The original dataset was translated to Portuguese (Brazil) to provide a more culturally and linguistically relevant resource for the Brazilian market. |
|
|
|
The dataset was carefully reviewed to identify and fix issues present in the original release, ensuring that the model is trained on high-quality data. The model is intended to be used in applications where a deep understanding of Portuguese language is required, such as chatbots, virtual assistants, and language translation systems. |
|
|
|
## Intended uses: |
|
|
|
Generating responses to natural language questions and prompts in Portuguese |
|
|
|
Supporting chatbots, virtual assistants, and other conversational AI applications |
|
|
|
Enhancing language translation systems and machine translation models |
|
|
|
Providing a culturally and linguistically relevant resource for the Brazilian market |
|
|
|
## Limitations |
|
|
|
The model may not generalize well to other languages or dialects |
|
|
|
The model may not perform well on out-of-domain or unseen topics |
|
|
|
The model may not be able to handle ambiguous or open-ended prompts |
|
|
|
The model may not be able to understand nuances of regional dialects or slang |
|
|
|
The model may not be able to handle prompts that require common sense or real-world knowledge |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 2e-05 |
|
- train_batch_size: 1 |
|
- eval_batch_size: 1 |
|
- seed: 42 |
|
- gradient_accumulation_steps: 8 |
|
- total_train_batch_size: 8 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: cosine |
|
- lr_scheduler_warmup_steps: 100 |
|
- num_epochs: 2 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | |
|
|:-------------:|:-----:|:----:|:---------------:| |
|
| 1.382 | 0.01 | 1 | 1.4056 | |
|
| 1.1762 | 0.5 | 45 | 1.1987 | |
|
| 1.1294 | 0.99 | 90 | 1.1493 | |
|
| 1.0028 | 1.47 | 135 | 1.1331 | |
|
| 0.9899 | 1.97 | 180 | 1.1227 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.40.0.dev0 |
|
- Pytorch 2.2.2+cu121 |
|
- Datasets 2.15.0 |
|
- Tokenizers 0.15.0 |
|
|
