ChatGLM3-6B-32K
Parameters
6B
Context Length
33K
Modality
Text
Architecture
Dense
License
ChatGLM3-6B Model License
Release Date
27 Oct 2023
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 ChatGLM3-6B-32K available.
Rankings
Overall Rank
-
Coding Rank
-
About ChatGLM3-6B-32K
ChatGLM3-6B-32K is an advanced large language model optimized for long-context understanding and generation. Developed through a collaboration between Zhipu AI and Tsinghua University's KEG Lab, this model serves as a specialized variant of the ChatGLM3-6B architecture, specifically engineered to extend the effective context window to 32,768 tokens. This expansion allows for the processing of comprehensive documents, long-form dialogues, and complex technical texts that exceed the limits of standard transformer-based models.
The model's architecture is built upon a 28-layer dense transformer framework. It incorporates several technical refinements to maintain stability and performance across its extended context, including the use of RMSNorm for normalization and Multi-Query Attention (MQA) to optimize inference efficiency. A significant innovation in this variant is the updated Rotary Position Embedding (RoPE) mechanism, which utilizes a modified base frequency (rope_ratio) to ensure precise positional resolution over 32K tokens. Furthermore, the model is trained with a specialized methodology that emphasizes long-text coherence during the conversation stage.
Designed for technical versatility, ChatGLM3-6B-32K natively supports tool invocation through function calling, code execution via an integrated code interpreter, and complex agent-based tasks. These features make it highly suitable for building sophisticated AI agents capable of deep text analysis and multi-step reasoning. The model's weights are open for academic research and available for free commercial use following a formal registration process, reflecting a commitment to accessible high-performance natural language processing.
Technical Specifications
Attention
Attention Structure
Multi-Head Attention
Attention Heads
32
Key-Value Heads
2
Attention Head Dimension
128
Position Embedding
Absolute Position Embedding
RoPE Theta
-
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
4,096
Number of Layers
28
FFN Intermediate Size (Dense)
13,696
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
65,024
About ChatGLM
ChatGLM series models from Z.ai, based on GLM architecture.