InternVL2_5-1B-MPO

Introduction

InternVL 2.5 is a series of Multimodal Large Language Models (MLLMs) built on InternVL 2.0. While retaining the core model architecture, it introduces significant optimizations in training strategies, evaluation methods, and data quality.

Available NPU Models

Base Model

internvl2.5-1B-364-ax630c

The Base Model provides a context window of 256 and supports a maximum output of 1,024 tokens.
Supported Platforms: LLM630 Compute Kit, Module LLM, and Module LLM Kit

• Context Window: 256
• Maximum Output Tokens: 1,024
• Time to First Token (TTFT): 1117.27 ms
• Average Generation Speed: 10.56 token/s
• Image Encoding Resolution: 364×364
• Image Encoding Time: 1164.61 ms

Installation

apt install llm-model-internvl2.5-1b-364-ax630c

• Download llm-model-internvl2.5-1b-364-ax630c

internvl2.5-1B-448-ax650

The Base Model provides a context window of 320 and supports a maximum output of 1,024 tokens.
Supported Platforms: AI Pyramid

• Context Window: 320
• Maximum Output Tokens: 1,024
• Time to First Token (TTFT): 433.87 ms
• Average Generation Speed: 29.48 token/s
• Image Encoding Resolution: 448×448
• Image Encoding Time: 362.22 ms

Installation

apt install llm-model-internvl2.5-1b-448-ax650

• Download llm-model-internvl2.5-1b-448-ax650