InternVL2-1B

Original model repository: OpenGVLab/InternVL2-1B

Model Introduction

InternVL2-1B is an instruction-tuned Vision-Language Model (VLM) for understanding images and generating text responses. It uses an InternViT-300M vision encoder, an MLP projector, and Qwen2-0.5B-Instruct as its language model. The model can be used for visual question answering, image description, OCR, document and chart understanding, and general multimodal dialogue.

Deployment Metrics

Model Parameters

Metric	Value
Total model parameters	938.2M
Vision model (ViT) parameters	308.5M
Language model (LM) parameters	629.7M

Parameter counts are calculated from the tensors stored in the upstream checkpoint.

Performance Metrics

Test Configuration

Metric	Value
Platform	Matrix6P
Data type	W8A8
ViT image size	448 × 448
Sequence length	512
Maximum context length	1024
BPU cores (ViT / Prefill / Decode)	4 / 4 / 4

Performance Results

Metric	Value
ViT latency	41.451 ms
Time to first token (TTFT)	65.766 ms
Prefill throughput	24,955.227 tokens/s
Decode throughput	159.481 tokens/s

Memory Usage

Metric	Value
BPU memory	1.01 GB
CPU memory	0.68 GB

Note: TTFT includes preprocessing and ViT latency. Memory values represent the peak memory usage measured during the specified performance test.