Skip to content

v1.2.0

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 21 Oct 17:50
· 5 commits to main since this release
Immutable release. Only release title and notes can be modified.
v1.2.0
c765a57
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

🚀 TT CLI v1.2.0 is out!

A few highlights since v1.1.0:

🚀 New Features

  • Hermes agent: tt launch hermes connects the Hermes agent to a model you're serving.
  • tt model serve: The tt model group now has a serve command, the same as tt serve. tt model package, package-thin and push now accept -h like every other tt command.
  • Open WebUI without a login: tt launch openwebui now opens straight into the chat instead of the create-an-account screen.
  • Readable bundle info: Bundle details now print as a human-readable table by default.

🤖 Model Support

New community bundles, and bundles now available on p100:

Model What's new Hardware
Qwen3-Coder-Next New p150x2, p150x4, p300, p300x2
GLM-4.5-Air New p150x4, p300x2
gpt-oss-20b (fast) New p150, p150x2, p150x4
MiniMax-H3 New p150, p300x2
MiMo V2.6 Distill (Qwen 9B) New p150
Voxtral Mini 4B Realtime New p300x2
Nemotron 3.5 Lightning 30B-A3B Fixed successor p300x2
CLEF New p150x2
CLM v0.1 8B New p150, p150x4
Laya New p150, p150x4
Laya Typed Decisions New p150, p150x4
Ornith 1.0 35B Now on p100 p100, p150, p150x2, p150x4
Ornith 1.5 35B-A3B Now on p100 p100, p150, p300x2
MiniMax Music 3 Now on p100 p100, p150
S2 Pro Now on p100 p100, p150, p300, p300x2, p300x4
Gemma 4 12B Now on p100 p100, p150
Llama 3.2 3B Now on p100 p100, p150
SmolLM2 135M Now on p100 p100, p150

🐛 Bug Fixes

  • Fixed served model weights not being readable by the inference-server container on systems with a restrictive umask or newer huggingface_hub cache layouts.
  • Fixed Hugging Face tokens and other secrets possibly appearing in on-disk logs, crash reports and resume commands. Log files are now private (0600/0700) and streamed output is redacted.
  • Fixed bundles pulled before a board was added to the catalog showing an outdated hardware list.
  • Fixed --workflow evals|benchmarks not being recorded in tt serve usage telemetry.
  • Error messages and hints no longer tell you to run the standalone tt-model binary, which usually isn't on your PATH. They now point to the matching tt commands.

🔧 Technical Improvements

  • Bumped the tt-inference-server pin to v0.25.0 and refreshed the release model spec.
  • Bumped the pinned versions of the upstream tools tt installs.
  • New docs: prerequisites for installing and running tt, and the clients tt launch can connect along with what each needs installed.
  • Release docs now say that hotfixes land on the RC branch.
  • Installer tests now stay fully offline.

Full changelog → v1.1.0...v1.2.0

Assets 7
Loading