---
library_name: peft
license: llama3.1
base_model: barc0/Llama-3.1-ARC-Heavy-Transduction-8B
tags:
- alignment-handbook
- trl
- sft
- generated_from_trainer
datasets:
- tttx/test-ttft
- barc0/transduction_formatted_rearc_dataset_100k
- barc0/transduction_heavy_100k_jsonl
model-index:
- name: arc-heavy-llama3.1-8b-lora64-testtime-finetuning
  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. -->

# arc-heavy-llama3.1-8b-lora64-testtime-finetuning

This model is a fine-tuned version of [barc0/Llama-3.1-ARC-Heavy-Transduction-8B](https://huggingface.co/barc0/Llama-3.1-ARC-Heavy-Transduction-8B) on the tttx/test-ttft, the barc0/transduction_formatted_rearc_dataset_100k and the barc0/transduction_heavy_100k_jsonl datasets.
It achieves the following results on the evaluation set:
- Loss: 0.0401

## 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: 0.0002
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- total_eval_batch_size: 16
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 3

### Training results

| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 0.0373        | 1.0   | 667  | 0.0589          |
| 0.0344        | 2.0   | 1334 | 0.0550          |
| 0.009         | 3.0   | 2001 | 0.0401          |


### Framework versions

- PEFT 0.13.2
- Transformers 4.46.2
- Pytorch 2.4.0+cu121
- Datasets 3.1.0
- Tokenizers 0.20.3