yieldthought-agents 0.1.0

Automation for TTNN model bringup.

Automod agents

Automation for TTNN model bringup using GitHub issues/projects and codexapi tasks.

Requirements

• gh, git, python3, codex CLI
• tt-smi (or equivalent TT reset tool)
• codexapi Python package

Install Python deps:

pip install -r requirements.txt

Install as a package (adds yt-bringup-worker):

pip install -e .

Environment variables

Required:

• YT_SYSTEM = n150 | n300 | lb
• YT_OWNER (default: yieldthought)
• YT_REPO_MODELS (default: ttnn_models)
• YT_WORKER_NAME (default: hostname)

Project config:

• YT_PROJECT_NUMBER (preferred) or YT_PROJECT_TITLE

HF + Codex:

• HF_TOKEN (required for gated models)
• CODEX_BIN (optional)
• CODEX_* env vars required by your codex install

Policy defaults:

• YT_TOP1_MIN (default 0.90)
• YT_TOP5_MIN (default 0.97)
• YT_MAX_ATTEMPTS (default 10)
• YT_SLEEP_SECS (default 20)
• YT_TMP_ROOT (optional temp root)

Run the worker

python scripts/run_worker.py

Run once and exit:

python scripts/run_worker.py --once

Dry run (prints actions only):

python scripts/run_worker.py --dry-run

Dry run with a specific issue number in the output:

python scripts/run_worker.py --dry-run --issue 123

Bootstrap a machine

scripts/bootstrap_machine.sh

Tests

pytest -q