---
library_name: peft
license: llama3.1
base_model: meta-llama/Llama-3.1-8B
tags:
- generated_from_trainer
model-index:
- name: outputs/020
  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/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>

axolotl version: `0.5.2`
```yaml
# Название базовой модели, которая будет использоваться
base_model: meta-llama/Llama-3.1-8B

chat_template: llama3
datasets:
  - path: /workspace/dataset_200_30_repeats_by_cycles.jsonl # (A,B,C)x30
    type: chat_template
    field_messages: conversations
    message_field_role: role
    message_field_content: content
    roles:
      user: ["user"]
      assistant: ["assistant"]
      system: ["system"]
    roles_to_train: ["assistant", "user"]
    train_on_eos: turn  # Тренировать EOS на каждом конце реплики для лучшего запоминания

# Путь к директории для сохранения результатов обучения
output_dir: ./outputs/020

# Настройки обучения
gradient_accumulation_steps: 5
micro_batch_size: 2
num_epochs: 1
learning_rate: 0.000002
warmup_steps: 500
logging_steps: 10

# Использование повторного обучения через LoRA
adapter: lora
lora_r: 16  # Увеличенное значение для сохранения памяти и генерации точных ответов
lora_alpha: 32
lora_dropout: 0.1
lora_target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj

# Тип модели и токенизатора
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

# Настройки последовательности
sequence_len: 4096  # Достаточная длина для обработки 7 реплик
sample_packing: false  # Отключено для лучшего соответствия тексту
pad_to_sequence_len: true

# Оптимизация
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
weight_decay: 0.01
gradient_checkpointing: true

# Использование BF16 для экономии памяти
bf16: true

# Flash Attention для ускорения
flash_attention: true

# Доля данных для валидации
val_set_size: 0.1

# Настройки сохранения
save_safetensors: true
saves_per_epoch: 3  # Увеличено для промежуточного анализа качества модели

# Настройки метрик
evals_per_epoch: 10
eval_max_new_tokens: 128

# Специальные токены
special_tokens:
  pad_token: "<|finetune_right_pad_id|>"
  bos_token: "<|begin_of_text|>"
  eos_token: "<|end_of_text|>"

# WandB интеграция (если требуется)
wandb_project:
wandb_entity:
wandb_watch:
wandb_log_model:

```

</details><br>

# outputs/020

This model is a fine-tuned version of [meta-llama/Llama-3.1-8B](https://huggingface.co/meta-llama/Llama-3.1-8B) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 1.0477

## 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: 2e-06
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 5
- total_train_batch_size: 10
- optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 500
- num_epochs: 1

### Training results

| Training Loss | Epoch  | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| No log        | 0.0009 | 1    | 1.5900          |
| 1.4756        | 0.1007 | 106  | 1.5880          |
| 1.4972        | 0.2013 | 212  | 1.5655          |
| 1.4563        | 0.3020 | 318  | 1.5018          |
| 1.3754        | 0.4027 | 424  | 1.4127          |
| 1.271         | 0.5033 | 530  | 1.3056          |
| 1.2054        | 0.6040 | 636  | 1.2009          |
| 1.1065        | 0.7047 | 742  | 1.1182          |
| 1.0592        | 0.8053 | 848  | 1.0689          |
| 1.0322        | 0.9060 | 954  | 1.0477          |


### Framework versions

- PEFT 0.13.2
- Transformers 4.46.3
- Pytorch 2.5.1+cu124
- Datasets 3.1.0
- Tokenizers 0.20.3