What are the hardware requirements for Codegeex2 6B?

Codegeex2 6B requires approximately 12.5 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Codegeex2 6B and how large is it?

Codegeex2 6B is developed by THUDM, a model with 6b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Codegeex2 6B perform on standard benchmarks?

Codegeex2 6B has the following published scores: HumanEval (Python) 33.49. Compare against reference models on this page or on the LLM Explorer leaderboards.

Codegeex2 6B by THUDM — VRAM 12.5GB

Name: Codegeex2 6B
Author: THUDM

Codegeex2 6B is an open-source language model by THUDM. Features: 6b LLM, VRAM: 12.5GB, LLM Explorer Score: 0.09, HumanEval: 33.5.

Arxiv:2303.17568 Chatglm Codegeex Custom code En Endpoints compatible Glm Pytorch Region:us Sharded Thudm Zh

Model Card on HF 🤗: https://huggingface.co/THUDM/codegeex2-6b

Codegeex2 6B Parameters and Internals

Model Type

Multilingual Code Generation

Use Cases

Areas:

Research, Commercial Applications

Applications:

Coding assistants, Code generation, Code translation

Primary Use Cases:

Code generation in various languages, Automated code completion, AI-assisted programming

Limitations:

Advised against use in critical code deployments without review

Additional Notes

CodeGeeX2 significantly improves over the first generation, adding more powerful features and multilingual support.

Supported Languages

zh (Chinese), en (English)

Training Details

Data Sources:

Various code sources for pre-training

Data Volume:

600B code tokens

Methodology:

Based on ChatGLM2 architecture.

Context Length:

8192

Model Architecture:

ChatGLM2 with additional code pre-training

Input Output

Input Format:

Language tagged code prompts

Accepted Modalities:

Textual code inputs

Output Format:

Generated code

Performance Tips:

Adding language tags to prompts can improve performance.

LLM Name	Codegeex2 6B
Repository 🤗	https://huggingface.co/THUDM/codegeex2-6b
Model Size	6b
Required VRAM	12.5 GB
Updated	2025-07-28
Maintainer	THUDM
Model Type	chatglm
Model Files	1.8 GB: 1-of-7 2.0 GB: 2-of-7 1.9 GB: 3-of-7 1.8 GB: 4-of-7 2.0 GB: 5-of-7 1.9 GB: 6-of-7 1.1 GB: 7-of-7
Supported Languages	zh en
Model Architecture	ChatGLMModel
Transformers Version	4.30.2
Tokenizer Class	ChatGLMTokenizer
Torch Data Type	bfloat16

Rank the Codegeex2 6B capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to Codegeex2 6B

Best Alternatives	Context / RAM	Downloads	Likes
Chatglm3 6B	0K / 12.5 GB	659154	1138
Chatglm2 6B	0K / 12.5 GB	617325	2050
Chatglm2 6B	0K / 12.5 GB	447665	2057
Chatglm3 6B	0K / 12.5 GB	87714	1166
Chatglm 6B	0K / 13.5 GB	1297	2903
Chatglm 6B	0K / 13.5 GB	2816	2862
Codegeex2 6B	0K / 12.5 GB	435	258
Octogeex	0K / 12.5 GB	156	22

Note: green Score (e.g. "73.2") means that the model is better than THUDM/codegeex2-6b.