TT-Metal v0.65.0 Release Notes
This release contains significant improvements and new features.
Changes
See CHANGELOG.txt for detailed commit history.
Installation
Refer to INSTALLING.md for installation instructions.
Model Updates
New
- New Op Infrastructure Enablement for LLM & Diffusion Models
Core transformer execution paths (QKV, rotary embeddings, SDPA decode) migrated to the new op infra, forming the backbone for scalable LLM and diffusion support.
PR #33209 – Migrate op to new infra: sdpa_decode
Model Performance & Accuracy Updates
- Stable Diffusion / SDXL Accuracy Fix
Corrected SDXL VAE accuracy issues that impacted image quality and downstream validation.
PR #33156 – SDXL vae batch encode accuracy fix
Improvements and New Features
-
Sub-Core Grid Scaling Across Ops
Enabled sub-core grid support for core unary ops, unblocking better utilization and scaling on large devices.
PR #33157 – Add sub_core_grids to unary infra and ops -
Numerical Accuracy Fixes in Core Math Ops
Fixed accuracy issues in exponential-related ops that directly affect model convergence and output quality.
PR #33139 – Fix expm1 accuracy -
Large-Kernel Support
Added support for huge kernels, enabling execution of larger and more complex workloads without fragmentation.
PR #32956 – Huge kernel support -
Improved Error Propagation in Build System
Ensured exceptions in build threads correctly propagate to the main thread, preventing silent failures.
PR #33205 – Ensure exceptions in build threads are propagated -
Fabric Router Heartbeat
Added heartbeat support to the fabric router, significantly improving detection of stalled or unhealthy links.
PR #31255 – Fabric router heartbeat feature -
Telemetry Firmware Visibility
Exposed remaining firmware versions via telemetry, improving fleet visibility and debugging.
PR #33158 – Telemetry: Expose remaining firmware versions -
CI & Workflow Hardening
Embedded pytest commands directly into Galaxy workflows, reducing CI flakiness and improving debuggability.
PR #32991 – Embed Pytest commands in Galaxy workflows
Full Changelog: v0.64.5...v0.65.0