Skip to content

v2.7.0

Latest
Latest

Choose a tag to compare

View all tags
@rnabeelTT rnabeelTT released this 16 Jun 14:57
v2.7.0
7d19c0d
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

The official release of v2.7.0 for TT Studio is here! Please pull the latest updates, explore the new features, and continue to use the in-app bug reporting tool to submit feedback.

What's Changed

Full Release v2.7.0 in 7d19c0d via PR #884

Changelog v2.7.0

🚀 New Features

  • Wake Mode & VAD for Voice Agent: Added wake-word activation alongside voice-activity detection and push-to-talk, letting the Voice Agent listen and respond hands-free or on demand. (#806)
  • Image Pull & Download Progress: Introduced a live progress bar during Docker image pulls, with an accurate hand-off into container-start tracking so deployment feedback stays continuous. (#758, #866)
  • Media-Model Progress Bar: Extended deployment progress visibility to TTS and Whisper media models. (#812)
  • Startup Freshness Check: run.py now detects stale local state on bring-up and surfaces a warning/prompt before proceeding. (#724)
  • --cleanup vs --cleanup-all: Reworked teardown so --cleanup preserves deployment containers and saved secrets while --cleanup-all fully resets TT Studio to a fresh-clone state — wiping generated logs, preferences, virtualenvs, named model-weight volumes, deployment containers, and locally built images behind a single confirmation prompt (--yes for non-interactive use). (#799, #833)
  • Quicker Parallel Delete & Reset: Faster delete and reset for individual deployments within multi-slot configurations. (#841)
  • Enhanced Device Management & Reset: Improved device allocation and reset handling within the Docker control service. (#788)
  • Public Landing Page: Added a GitHub Pages landing site served from the docs/ root. (#839, #844)

🐛 Bug Fixes

  • Deployment History Status: Stopped models now report a real end time and a proper stopped status instead of showing "N/A" or "Still running". (#868)
  • API Examples for Non-LLM Models: Resolved 404s when fetching example snippets for TTS and Whisper models. (#876)
  • TPS for Base Models: Corrected tokens-per-second reporting for base (non-instruct) models. (#822)
  • Qwen3-32B Device IDs: Fixed incorrect device IDs displayed on the Models Deployed page. (#837)
  • HuggingFace Integration: Updated docker-compose.yml to ensure correct first-run HuggingFace integration. (#842)
  • Terms Acceptance Persistence: User terms acceptance is now persisted across restarts, removing the repeat prompt. (#847)
  • QB2 Preloaded Weights: Reuse preloaded weights under the pinned release instead of re-downloading them. (#848)

🔧 Technical Improvements

  • Unified Streaming Views: Stop and reset actions now share unified streaming views for consistent live feedback. (#872)
  • Single Source of Truth for Read State: Consolidated all read paths onto one authoritative state source. (#814)
  • Model Catalog Cleanup: Removed non-working and deprecated models, including Qwen3-8B. (#869, #878)
  • P300 Naming & Model Visibility: Bumped the inference artifact to v0.15.0 and corrected P300 labels and model visibility. (#815)
  • tt-smi Upgrade: Updated tt-smi from v4.0.0 to v4.1.2. (#824)
  • RAG Data Source: Refreshed the RAG knowledge base to the latest Tenstorrent documentation. (#870)
  • Progress-Bar Color Scheme: Polished the progress-bar color scheme. (#834)
  • Deprecated Legacy Startup Script: Removed the legacy startup.sh in favor of run.py. (#805)

📄 Docs & Chores

  • Slimmed the root README and linked a dedicated setup guide. (#823)
  • Added license files and a compliance-check script. (#875)
  • Cleaned up stray test files and refreshed the footer. (#845)

🧪 Testing

Validated against the v2.7.0-rc test plan on a p300x2 (QB2) host.

As always, thank you for your help in testing this release!

Assets 2
Loading