Qwen2.5-72B
Parameters
72B
Context Length
131K
Modality
Text
Architecture
Dense
License
Qwen License
Release Date
19 Sept 2024
Knowledge Cutoff
Jan 2025
System Requirements
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
8x RTX 4090
24GB VRAM
Datacenter
3x NVIDIA A100
80GB VRAM
Apple Silicon
2x Apple M3 Max
128GB VRAM
131,072 tokens
Consumer
10x RTX 4090
24GB VRAM
Datacenter
3x NVIDIA A100
80GB VRAM
Apple Silicon
2x Apple M3 Max
128GB VRAM
Architecture Diagram
Evaluation Benchmarks
Rank
#127
| Benchmark | Score | Rank |
|---|---|---|
QA Assistant | 0.935 | 12 |
Agent Arena | 0.07 | 16 |
Summarization | 0.742 | 19 |
Professional Knowledge | 0.71 | 62 |
General Text | 1303 | 128 |
Rankings
Overall Rank
#127
Coding Rank
#51
About Qwen2.5-72B
Qwen2.5-72B is a core component of the Qwen2.5 series of large language models developed by Alibaba. This model is built upon a Transformer architecture and operates as a causal language model. Its design incorporates Rotary Position Embeddings (RoPE), SwiGLU as the activation function, and RMSNorm for normalization, complemented by an attention mechanism that includes QKV bias. These architectural choices provide a robust foundation for general-purpose language processing tasks.
The Qwen2.5-72B model features advancements compared to its predecessor, Qwen2. It exhibits enhanced capabilities in handling complex knowledge, excelling in areas such as coding and mathematics. The model also demonstrates improved instruction following, making it more adaptable to diverse user prompts and conditional scenarios. Its design focuses on practical applications requiring high fidelity in output generation.
This model is engineered for extensive text processing, supporting context lengths up to 131,072 tokens and generating outputs up to 8,192 tokens. It is proficient in generating long-form content, understanding structured data formats like tables, and producing structured outputs such as JSON. Additionally, Qwen2.5-72B provides multilingual support across more than 29 languages, making it suitable for a wide array of content generation, coding assistance, and advanced artificial intelligence applications like chatbots and virtual assistants.
Technical Specifications
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
128
Key-Value Heads
8
Attention Head Dimension
-
Position Embedding
ROPE
RoPE Theta
1,000,000
Sliding Window Attention
No
Sliding Window Size
-
Sliding Window Ratio
-
Linear Attention
-
Linear Attention Ratio
-
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Hidden Dimension Size
12,288
Number of Layers
80
FFN Intermediate Size (Dense)
29,568
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
152,064
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.