artificialguybr's picture
Upload README.md with huggingface_hub
3235183 verified
metadata
library_name: transformers
license: apache-2.0
base_model: Qwen/Qwen2.5-0.5B
language:
  - en
pipeline_tag: text-generation
tags:
  - generated_from_trainer
  - instruction-tuning
model-index:
  - name: outputs/qwen2.5-0.5b-ft-synthia15-i
    results: []

Built with Axolotl

Qwen2.5-0.5B Fine-tuned on Synthia v1.5-I

This model is a fine-tuned version of Qwen/Qwen2.5-0.5B on the Synthia v1.5-I dataset, which contains over 20.7k instruction-following examples.

Model Description

Qwen2.5-0.5B is part of the latest Qwen2.5 series of large language models. The base model brings significant improvements in:

  • Instruction following and generating long texts
  • Understanding structured data and generating structured outputs
  • Support for over 29 languages
  • Long context support up to 32,768 tokens

This fine-tuned version enhances the base model's instruction-following capabilities through training on the Synthia v1.5-I dataset.

Model Architecture

  • Type: Causal Language Model
  • Parameters: 0.49B (0.36B non-embedding)
  • Layers: 24
  • Attention Heads: 14 for Q and 2 for KV (GQA)
  • Context Length: 32,768 tokens
  • Training Framework: Transformers 4.45.0.dev0

Intended Uses & Limitations

This model is intended for:

  • Instruction following and task completion
  • Text generation and completion
  • Conversational AI applications

The model inherits the multilingual capabilities and long context support of the base Qwen2.5-0.5B model, while being specifically tuned for instruction following.

Training Procedure

Training Data

The model was fine-tuned on the Synthia v1.5-I dataset containing 20.7k instruction-following examples.

Training Hyperparameters

The following hyperparameters were used during training:

  • Learning rate: 1e-05
  • Train batch size: 5
  • Eval batch size: 5
  • Seed: 42
  • Gradient accumulation steps: 8
  • Total train batch size: 40
  • Optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • LR scheduler type: cosine
  • LR scheduler warmup steps: 100
  • Number of epochs: 3
  • Sequence length: 4096
  • Sample packing: enabled
  • Pad to sequence length: enabled

Framework Versions

  • Transformers 4.45.0.dev0
  • Pytorch 2.3.1+cu121
  • Datasets 2.21.0
  • Tokenizers 0.19.1
See axolotl config

axolotl version: 0.4.1