FlagRelease/Qwen3.8-27B-BF16-sunrise-FlagOS

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 14, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

FlagRelease/Qwen3.8-27B-BF16-sunrise-FlagOS is a 27 billion parameter vision-language model based on the Qwen architecture, developed by Alibaba and adapted by the Zhongzhi (众智) FlagOS community. This model is specifically optimized for multi-chip deployment and precision alignment across 11 different AI chips, including NVIDIA and Sunrise, with support for BF16 precision. It provides out-of-the-box solutions for developers, enabling efficient and automated model migration and deployment across diverse hardware platforms.

Loading preview...

Model Overview

FlagRelease/Qwen3.8-27B-BF16-sunrise-FlagOS is a 27 billion parameter vision-language model, originating from Alibaba's Qwen3.8-27B, that has been extensively adapted by the Zhongzhi (众智) FlagOS community. Its primary distinction lies in its comprehensive multi-chip adaptation and precision alignment, supporting 11 different AI chips such as NVIDIA, Moore Threads, Ascend, and Sunrise. This model is designed to facilitate a "develop once, run anywhere" workflow, significantly reducing the complexity and cost of deploying AI workloads across varied hardware.

Key Capabilities & Features

  • Broad Hardware Compatibility: Adapted for 11 AI chips, with NVIDIA and Moore Threads supporting FP8 precision, and others running on BF16.
  • Edge-side Deployment: Includes a W4A8 low-bit version for ARM edge-side platforms.
  • Integrated Deployment: Offers out-of-the-box inference scripts and a dedicated FlagOS-sunrise container image for rapid setup.
  • Consistency Validation: Rigorously benchmarked against native stacks to ensure performance and result consistency.
  • FlagOS Ecosystem: Leverages the FlagOS open-source software stack, including FlagScale, FlagGems, FlagCX, and FlagTree, for unified model-system-chip integration.

Use Cases & Benefits

  • Multi-chip AI Deployment: Ideal for developers needing to deploy large language models across a heterogeneous mix of AI accelerators without extensive porting efforts.
  • Reduced Development Overhead: Simplifies the process of adapting and maintaining AI workloads on different hardware, thanks to the FlagOS unified stack.
  • Benchmarking and Validation: Provides a validated solution for assessing model performance on various chips, ensuring consistent results.
  • Edge AI Applications: The W4A8 low-bit version extends its utility to resource-constrained ARM edge devices.

Finetunes
1 models