README.md

---
language:
- en
license: apache-2.0
tags:
- text-generation-inference
- transformers
- unsloth
- llama
- trl
base_model: unsloth/llama-3-8b-bnb-4bit
datasets:
- cognitivecomputations/samantha-data
---

# Uploaded  model

- **Developed by:** ruslandev
- **License:** apache-2.0
- **Finetuned from model :** unsloth/llama-3-8b-bnb-4bit

This model is finetuned on the data of [Samantha](https://erichartford.com/meet-samantha).  
Prompt format is Alpaca. I used the same system prompt as the original Samantha.
```
"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
{SYSTEM_PROMPT}

### Input:
{QUESTION}

### Response:
"""
```

# Training

[gptchain](https://github.com/RuslanPeresy/gptchain) framework has been used for training.

## Training hyperparameters

- learning_rate: 2e-4
- seed: 3407
- gradient_accumulation_steps: 4
- per_device_train_batch_size: 2
- optimizer: adamw_8bit
- lr_scheduler_type: linear
- warmup_steps: 5
- num_epochs: 2
- weight_decay: 0.01

## Training results

|Training Loss | Epoch | Step |
|--------------|-------|------|
|2.0778        |0.0    |1     |
|0.6255        |0.18   |120	  |
|0.6208        |0.94   |620   |
|0.6244        |2.0    |1306  |

2 epoch finetuning from llama-3-8b took 1 hour on a single A100 with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.

[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)