BladeDISC

Visit Tool

BladeDISC is a Coding & Development tool that optimizes machine learning workloads. It provides an end-to-end dynamic shape compiler for TensorFlow and PyTorch on GPGPU and CPU backends.

Claim this tool

2Views

At a glance

Pricing

Open Source

Free tier

Yes

API

Skill level

Technical

About

What is BladeDISC?

BladeDISC is an end-to-end Dynamic Shape Compiler project designed to optimize machine learning workloads, serving as a key component of Alibaba's PAI-Blade. It offers general, transparent, and user-friendly performance optimization for both TensorFlow and PyTorch workloads across GPGPU and CPU backends. The architecture inherently supports dynamic shape workloads, with careful consideration for performance in both static and dynamic shape scenarios. BladeDISC also provides multiple flexible deployment solutions, including a Plugin Mode for integration within TensorFlow/PyTorch runtimes and a Standalone Mode for AOT standalone execution. The project is built upon MLIR and closely collaborates with the mlir-hlo and Torch-MLIR projects, aiming to unify and automate compiler solutions for both inference and training.

Best used for

Ideal for developers who need to optimize TensorFlow and PyTorch machine learning workloads, accelerate both inference and training, and efficiently handle dynamic shape scenarios. Especially valuable for those working with GPGPU and CPU backends and seeking a unified, automated compiler solution.

Common actions

optimize machine learning

accelerate AI workloads

compile dynamic models

enhance model performance

collaborationdeepfakeopen-sourcelow-code/no-codeautomated workflow"AI Agents"github copilotface swappingworkflows

Capabilities

Key features

Dynamic shape compilation
TensorFlow/PyTorch optimization
GPGPU/CPU backend support
Plugin mode deployment
Standalone mode execution
MLIR-based architecture

Target Audience

developer

Integrations

Not yet documented

Pricing & Plans

Open Source

Free

FAQs

What machine learning frameworks does BladeDISC support?

BladeDISC provides performance optimization for both TensorFlow (versions 1.12, 1.15, 2.4, 2.5) and PyTorch (version >= 1.6.0) workloads. It aims to offer broad compatibility across different versions, with some adaptation potentially needed for others.

How does BladeDISC handle dynamic shape workloads?

BladeDISC's architecture natively supports dynamic shape workloads, providing significant performance advantages. For example, it shows a 1.75x speedup for BERT large inference on T4 GPU compared to static compilers like XLA, which can suffer from severe performance degradation.

What are the deployment options for BladeDISC?

BladeDISC offers two main deployment solutions: Plugin Mode, where it works as a plugin within TensorFlow or PyTorch runtimes, and Standalone Mode, which compiles workloads into a self-executable binary that does not rely on framework runtimes.

Trending

Subcategories trending in Coding & Development

Open Source & Models DevOps & Infrastructure No-Code / Low-Code Testing & QA Backend & APIs Prompt Engineering

Trending

Explore

Browse AI tools by category

Content & Design Productivity & Business Coding & Development AI Agents & Automation Research & Education Wellness & Lifestyle Career Development Marketing & Growth Data & Analytics Customer Support & CX Finance E-commerce