ChatGLM3-6B
Parameters
6B
Context Length
8K
Modality
Text
Architecture
Dense
License
Apache 2.0
Release Date
27 Oct 2023
Knowledge Cutoff
Jul 2023
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
8,192 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
Rank
#163
| Benchmark | Score | Rank |
|---|---|---|
General Text | 1056 | 139 |
Rankings
Overall Rank
#163
Coding Rank
-
About ChatGLM3-6B
ChatGLM3-6B is an advanced bilingual (Chinese-English) large language model developed through a collaboration between Zhipu AI and the Knowledge Engineering Group at Tsinghua University. As the third generation in the ChatGLM series, this model implements a refined General Language Model architecture that bridges the functional divide between autoencoding and autoregressive objectives. The pre-training phase utilizes a diverse corpus comprising approximately one trillion tokens, optimized for conversational coherence and instruction following across multiple domains including mathematics, programming, and logical reasoning.
Technically, the model is built on a dense Transformer-based architecture featuring Multi-Head Attention and RoPE (Rotary Positional Embeddings) for efficient sequence handling. A significant advancement in the ChatGLM3 iteration is its native support for complex agent-centric workflows, including function calling and code execution via an integrated interpreter. This functionality is supported by a redesigned prompt format that facilitates structured interactions and multi-turn dialogue management, making it suitable for deployment in scenarios requiring autonomous task execution.
Designed for local and edge deployment, ChatGLM3-6B maintains a low computational footprint while delivering enhanced performance relative to its predecessors. It utilizes SwiGLU activation functions and RMSNorm for stable training, with a vocabulary expanded to support efficient bilingual tokenization. The model's versatility is demonstrated through its ability to handle a variety of downstream applications, from standard question-answering to sophisticated agentic behaviors, all while operating within a context window optimized for standard conversational tasks.
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.