Implementation:OpenGVLab InternVL Internvl Custom2hf

Knowledge Sources	InternVL
Domains	Model_Deployment, Tooling
Last Updated	2026-02-07 00:00 GMT

Overview

Concrete tool for converting InternVL custom checkpoint format to HuggingFace format provided by the InternVL tools.

Description

The internvl_custom2hf.py script converts InternVL training checkpoint format to standard HuggingFace format by remapping weight keys and saving in the expected directory structure.

Usage

Run after training when the output checkpoint uses internal naming conventions that differ from HuggingFace format.

Code Reference

Source Location

• Repository: InternVL
• File: internvl_chat/tools/internvl_custom2hf.py
• Lines: L1-31

Signature

python internvl_chat/tools/internvl_custom2hf.py <input_path> <output_path>

Import

python internvl_chat/tools/internvl_custom2hf.py ./output/checkpoint ./output/hf_model

I/O Contract

Inputs

Name	Type	Required	Description
input_path	str	Yes	Path to InternVL custom format checkpoint
output_path	str	Yes	Path for HuggingFace format output

Outputs

Name	Type	Description
HF checkpoint	Directory	Standard HuggingFace model directory

Usage Examples

Convert Checkpoint Format

python internvl_chat/tools/internvl_custom2hf.py \
    ./output/training_checkpoint \
    ./output/hf_compatible_model

# Verify conversion
python -c "
from internvl.model.internvl_chat import InternVLChatModel
model = InternVLChatModel.from_pretrained('./output/hf_compatible_model')
print('Model loaded successfully')
"

Related Pages

Implements Principle

• Principle:OpenGVLab_InternVL_Weight_Format_Conversion

Requires Environment

• Environment:OpenGVLab_InternVL_PyTorch_CUDA