FlagRelease/Qwen3.8-27B-BF16-ascend-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-ascend-FlagOS is a 27 billion parameter vision-language model, Qwen3.8-27B, adapted by the Zhongzhi FlagOS community for multi-chip deployment. This version is specifically configured for Ascend AI chips, running in BF16 precision, and is part of a broader initiative to enable 'develop once, run anywhere' across diverse AI accelerators. It provides out-of-the-box inference solutions with pre-configured hardware and software parameters, making it suitable for developers seeking efficient deployment on Ascend platforms.

Loading preview...

FlagRelease/Qwen3.8-27B-BF16-ascend-FlagOS Overview

This model is an adaptation of Alibaba's Qwen3.8-27B vision-language model, specifically optimized for deployment on Ascend AI chips within the FlagOS ecosystem. Developed by the Zhongzhi FlagOS community, it represents a significant effort in achieving hardware abstraction and unified deployment across various AI accelerators.

Key Features and Differentiators

  • Multi-Chip Adaptation: While this specific model targets Ascend, the FlagOS initiative ensures broad compatibility across 11 AI chips, including NVIDIA, Moore Threads, and Kunlunxin, with precision alignment and deployment verification.
  • BF16 Precision: This Ascend-specific release operates using BF16 precision, ensuring a balance between performance and numerical stability.
  • Out-of-the-Box Deployment: FlagOS provides integrated deployment solutions, including pre-configured container images and inference scripts, enabling rapid setup and execution on Ascend hardware.
  • Consistency Validation: The model's performance and results on the FlagOS software stack are rigorously benchmarked against native stacks to ensure consistency and reliability.
  • FlagOS Ecosystem: Leverages core FlagOS technologies like FlagScale (distributed training/inference), FlagGems (universal operator library), FlagCX (communication library), and FlagTree (unified compiler) to streamline model migration and deployment.

Use Cases

This model is ideal for developers and organizations looking to:

  • Deploy the Qwen3.8-27B model efficiently on Ascend AI hardware.
  • Benefit from a unified software stack that simplifies AI workload porting and maintenance.
  • Utilize a pre-validated and optimized solution for vision-language tasks on specific hardware.

Finetunes
1 models