1

14B Qwen2.5 Freya X1

Developed by Sao10K
A multi-stage training model based on Qwen2.5-14B and Qwen2.5-14B-Instruct, focusing on text generation and instruction understanding tasks.
Downloads 252
Release Time : 12/31/2024

Model Overview

This model adopts a two-stage training approach, first conducting LoRA training on literature and raw text, followed by further fine-tuning on instruction data, aiming to improve text generation quality and instruction-following capabilities.

Model Features

Multi-stage training
Adopts a two-stage training method, first basic training then instruction fine-tuning, to enhance model performance
Efficient fine-tuning
Uses LoRA adapters for parameter-efficient fine-tuning, reducing training costs
Long context support
Supports context lengths of up to 16384 tokens
Optimized training
Employs various optimization techniques such as flash attention and gradient checkpointing to improve training efficiency

Model Capabilities

Text generation
Instruction understanding
Literary creation
Dialogue systems

Use Cases

Content creation
Literary creation
Generates literary works such as novels and essays
Trained on cleaned literary datasets, capable of producing relatively high-quality literary content
Dialogue systems
Intelligent assistant
Builds instruction-following dialogue assistants
Fine-tuned on instruction data to improve instruction understanding and execution capabilities
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase