Supported self-hosted models and hardware requirements

Tier: Ultimate with GitLab Duo Enterprise - Start a trial Offering: Self-managed Status: Beta
History
  • Introduced in GitLab 17.1 with a flag named ai_custom_model. Disabled by default.
  • Enabled on self-managed in GitLab 17.6.
  • Changed to require GitLab Duo add-on in GitLab 17.6 and later.
  • Feature flag ai_custom_model removed in GitLab 17.8

The following table shows the supported models along with their specific features and hardware requirements to help you select the model that best fits your infrastructure needs for optimal performance.

Approved LLMs

Install one of the following GitLab-approved LLM models:

Model family Model Code completion Code generation GitLab Duo Chat
Mistral Codestral Codestral 22B v0.1 Yes Yes No
Mistral Mistral 7B-it v0.3 Yes Yes Yes
Mistral Mixtral 8x7B-it v0.1 Yes Yes Yes
Mistral Mixtral 8x22B-it v0.1 Yes Yes Yes
Claude 3 Claude 3.5 Sonnet Yes Yes Yes
GPT GPT-4 Turbo Yes Yes Yes
GPT GPT-4o Yes Yes Yes
GPT GPT-4o-mini Yes Yes Yes

The following models are under evaluation, and support is limited:

Model family Model Code completion Code generation GitLab Duo Chat
CodeGemma CodeGemma 2b Yes No No
CodeGemma CodeGemma 7b-it No Yes No
CodeGemma CodeGemma 7b-code Yes No No
Code Llama Code-Llama 13b-code Yes No No
Code Llama Code-Llama 13b No Yes No
DeepSeek Coder DeepSeek Coder 33b Instruct Yes Yes No
DeepSeek Coder DeepSeek Coder 33b Base Yes No No
Mistral Mistral 7B-it v0.2 Yes Yes Yes

Hardware Requirements

For optimal performance, the following hardware specifications are recommended as baselines for hosting these models. Hosting requirements may vary depending model to model, so we recommend checking model vendor documentation as well:

  • CPU: Minimum 8 cores (16 threads recommended).
  • RAM: At least 32 GB (64 GB or more recommended for larger models).
  • GPU:
    • Minimum: 2x NVIDIA A100 or equivalent for optimal inference performance.
    • Note: For running Mixtral 8x22B and Mixtral 8x22B-it, it is recommended to use 8x NVIDIA A100 GPUs.
  • Storage: SSD with sufficient space for model weights and data.