Deployment and upgradesOfficial documentation5 min readDeployment and upgrades

Determine Hardware Requirements for ChatGLM2 on FastGPT

VRAM Consumption Overview Per official ChatGLM2 technical specifications, generating 8192 tokens requires a specific amount of video RAM (VRAM) based on t…

VRAM Consumption Overview

Per official ChatGLM2 technical specifications, generating 8192 tokens requires a specific amount of video RAM (VRAM) based on the selected quantization precision. FP16 precision mode uses 12.8GB of VRAM, int8 quantization reduces this to 8.1GB, and int4 quantization further lowers VRAM usage to 5.1GB. Quantization slightly impacts model performance, but this effect is not significant.

Recommended Hardware Configurations

The following table outlines validated hardware and startup command parameters for deploying ChatGLM2 with FastGPT:

Quantization TypeMinimum System RAMMinimum VRAMMinimum Disk SpaceStartup Command
fp16>=16GB>=16GB>=25GBpython openai_api.py 16
int8>=16GB>=9GB>=25GBpython openai_api.py 8
int4>=16GB>=6GB>=25GBpython openai_api.py 4

All configuration profiles share a minimum 16GB system RAM requirement and 25GB of persistent disk space for model files and supporting dependencies.

Startup Command Reference

The provided startup command python openai_api.py requires a single integer argument that defines the model's quantization bit depth. This argument directly maps to the quantization type in the configuration table: passing 16 enables FP16 precision, 8 enables int8 quantization, and 4 enables int4 quantization. Omitting this parameter may cause improper model loading, so the specified integer must be included for successful deployment.

Source: FastGPT official source

Applicability and version scope

Use this page for the documented Deployment and upgrades scenario. Confirm the FastGPT, dependency, API, and deployment versions in the official source before applying a change.

Safety guardrails

Use [REDACTED_CREDENTIAL] for credentials and private data. Confirm the documented environment and version before review.

Rollback guidance

Restore the prior technical-content authority snapshot. Restore saved configuration and data snapshots, then repeat the smallest verification scenario.