Parameters
1.7B
Context Length
32.768K
Modality
Text
Architecture
Dense
License
Apache 2.0
Release Date
29 Apr 2025
Knowledge Cutoff
-
Attention Structure
Grouped-Query Attention
Hidden Dimension Size
-
Number of Layers
32
Attention Heads
32
Key-Value Heads
8
Activation Function
SwigLU
Normalization
RMS Normalization
Position Embedding
ROPE
VRAM requirements for different quantization methods and context sizes
Qwen3-1.7B is a dense causal language model developed by Alibaba's Qwen Team, introduced as part of the Qwen3 series on April 29, 2025. This model is designed for general-purpose language tasks and is characterized by its compact 1.7 billion parameter count. Its architecture is optimized for efficient operation across various hardware configurations, including environments with limited resources and edge devices. The model supports a context length of 32,768 tokens, enabling it to process extensive documents and conversations.
A distinguishing architectural feature within the Qwen3 series, including the 1.7B variant, is its dual operational modes: "Thinking Mode" and "Non-Thinking Mode." The Thinking Mode facilitates complex logical reasoning, such as mathematical problem-solving and code generation, through a step-by-step reasoning process. In contrast, the Non-Thinking Mode provides rapid, direct responses suitable for general conversational applications. This hybrid approach enables dynamic switching between modes, optimizing performance based on task complexity and efficiency requirements.
The model's architecture consists of 28 transformer layers, employing Grouped Query Attention (GQA) with 16 query heads and 8 key-value heads. It integrates Rotary Positional Embeddings (RoPE), specifically enhanced with ABF-RoPE, to maintain positional information accuracy across its extended context length. Further architectural refinements include the implementation of qk layernorm and RMSNorm with pre-normalization for stable training. Qwen3-1.7B demonstrates robust multilingual support, processing over 100 languages and dialects, and features advanced agent capabilities for tool integration.
The Alibaba Qwen 3 model family comprises dense and Mixture-of-Experts (MoE) architectures, with parameter counts from 0.6B to 235B. Key innovations include a hybrid reasoning system, offering 'thinking' and 'non-thinking' modes for adaptive processing, and support for extensive context windows, enhancing efficiency and scalability.
Ranking is for Local LLMs.
No evaluation benchmarks for Qwen3-1.7B available.
Overall Rank
-
Coding Rank
-
Full Calculator
Choose the quantization method for model weights
Context Size: 1,024 tokens