Skip to content

[Roadmap] Primus-Turbo Roadmap H1 2026 #211

@xiaobochen-amd

Description

@xiaobochen-amd

This roadmap is the H1 2026 development plan of Primus-Turbo.

Note: The roadmap is flexible and will be updated over time based on project needs and community input.

Release Overview

Version Framework Status Date
v0.3.0 PyTorch + JAX + ROCm7.x 🚧 In Progress TBD
v1.0.0 Planning Planning TBD

Detailed Plans

v0.3.0 (In Progress)

Focus

  • Introduce FP4 foundational support.
  • Performance optimization for FP16/BF16 and FP8.
  • Support JAX frontend.

Features

  • Attention
    • Unified Sequence Parallel
  • GEMM:
    • MXFP4
    • Performance optimization
  • GroupedGEMM
    • Add Hipblaslt Multi-Stream backend
    • Performance optimization
  • MoE TokenDispatcher
    • Add UCCL backend
  • Dist
    • Support sDMA
  • AutoTune
  • JAX

v1.0.0 (Planning)

Focus

  • New Hardware.
  • Key operators provide strong out-of-the-box performance.
  • Key operators support deterministic for RL.
  • Support diffusion model key operators like sparse/linear attention.

Past Roadmap

Primus-Turbo Roadmap H2 2025

Metadata

Metadata

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions