Qwen2.5-0.5B
Parameters
500M
Context Length
33K
Modality
Text
Architecture
Dense
License
Apache 2.0
Release Date
19 Sept 2024
Knowledge Cutoff
-
System Requirements
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
1x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
32,768 tokens
Consumer
1x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
Architecture Diagram
Evaluation Benchmarks
No evaluation benchmarks for Qwen2.5-0.5B available.
Rankings
Overall Rank
-
Coding Rank
-
About Qwen2.5-0.5B
Qwen2.5-0.5B is a foundational large language model developed by the Qwen team at Alibaba Cloud. It is part of the Qwen2.5 series, which represents an advancement in language model capabilities, featuring improvements in knowledge acquisition, coding proficiency, and mathematical reasoning. This variant, with approximately 0.49 billion parameters, serves as a robust base model, primarily designed for pretraining and subsequent fine-tuning for specialized applications. Its architecture is engineered to handle complex language tasks efficiently across multiple languages.
Architecturally, Qwen2.5-0.5B is a dense, decoder-only Transformer model. It incorporates Rotary Position Embedding (RoPE) for effective positional encoding, SwiGLU as its activation function, and RMSNorm for normalization. The attention mechanism utilizes Grouped Query Attention (GQA), specifically configured with 14 query heads and 2 key-value heads for this model size. The model is structured with 24 layers, contributing to its depth and capacity for learning intricate patterns in language data.
As a causal language model, Qwen2.5-0.5B is suitable for a range of downstream applications following post-training processes such as supervised fine-tuning or reinforcement learning from human feedback. Its capabilities include instruction following, generating extended text sequences, and processing structured data formats like JSON. The model supports a full context length of 32,768 tokens, with the broader Qwen2.5 series capable of handling contexts up to 128,000 tokens and generating outputs up to 8,000 tokens. It offers multilingual support, encompassing over 29 languages.
Technical Specifications
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
16
Key-Value Heads
8
Attention Head Dimension
-
Position Embedding
ROPE
RoPE Theta
1,000,000
Sliding Window Attention
No
Sliding Window Size
32,768
Sliding Window Ratio
-
Linear Attention
-
Linear Attention Ratio
-
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Hidden Dimension Size
768
Number of Layers
24
FFN Intermediate Size (Dense)
4,864
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
151,936
About Qwen2.5
Qwen2.5 by Alibaba is a family of dense, decoder-only language models available in various sizes, with some variants utilizing Mixture-of-Experts. These models are pretrained on large-scale datasets, supporting extended context lengths and multilingual communication. The family includes specialized models for coding, mathematics, and multimodal tasks, such as vision and audio processing.