MEGA Hub
New
bluemoon-sol

qwen3.5-0.8b-mega

Models0 followersupdated Jul 30, 2026Image Text To TextTransformerslicense:apache-2.0base_model:Qwen/Qwen3.5-0.8B-BaseLicense: apache-2.0
Request DOI
Use this model

Qwen3.5-0.8B-MEGA#

A compact multimodal foundation model based on Qwen3.5-0.8B, optimized for efficient deployment, research experimentation, and edge-scale AI applications.

Overview#

Qwen3.5-0.8B combines language understanding, visual perception, and reasoning capabilities in a lightweight architecture. This repository provides model weights and configuration files compatible with the modern AI inference ecosystem.

Supported frameworks:

  • Hugging Face Transformers
  • vLLM
  • SGLang
  • KTransformers
  • Other Transformers-compatible runtimes

Key Features#

  • Multimodal Understanding
    Unified vision-language modeling for image understanding, document analysis, and visual reasoning tasks.

  • Efficient Hybrid Architecture
    Combines Gated Delta Networks and sparse modeling techniques to provide strong capability with low inference cost.

  • Long Context Support
    Native context length up to 262K tokens for long-document processing and complex reasoning.

  • Agent-Oriented Capability
    Supports instruction following, tool usage, planning, and multi-step reasoning workflows.

  • Research Friendly
    Suitable for fine-tuning, benchmarking, inference optimization, and AI system research.

Model Details#

Attribute Value
Model Type Causal Language Model with Vision Encoder
Parameters 0.8B
Context Length 262,144 tokens
Architecture Hybrid Gated DeltaNet + Attention
Language Support Multilingual
License Apache 2.0

Usage#

Transformers#

PYTHON
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "lexing/qwen3.5-0.8b-mega"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

Inference Optimization#

For production deployment, the model can be served with high-performance inference engines such as vLLM or SGLang.

Recommended scenarios:

  • Local AI assistants
  • Multimodal agents
  • Research prototypes
  • Lightweight inference services

Limitations#

This model is intended for research and development. Performance may vary across domains, and users should evaluate suitability for safety-critical or production applications.

Citation#

If you use this model in research, please cite the original Qwen3.5 release and acknowledge this MEGA repository.

License#

This repository follows the Apache-2.0 license. Please review the original model license and usage requirements before deployment.