Tuned Tensor Documentation

End-to-end platform for shaping the behaviour of open-weight language models through instruction tuning. Define what your model should do, and the platform handles compilation, training, evaluation, and versioning.

Key Concepts

Behaviour Specs

A behaviour spec is a structured description of what you want your model to do. It includes:

System prompt — the persona and role of the model
Guidelines — rules the model should follow
Constraints — things the model must not do
Examples — input/output pairs demonstrating desired behaviour
Base model — which open-weight model to fine-tune

Runs

A run is a single end-to-end cycle: compile your spec into training data → augment with AI → fine-tune the model → auto-evaluate against your spec. Each run captures a snapshot of the spec at run time, so you can track how changes affect model behaviour over time.

Auto-Evaluation

After training completes, Tuned Tensor automatically evaluates the fine-tuned model by running each example from your spec and scoring the output. Scoring uses an LLM-as-judge for behavioural correctness, with fallback to similarity scoring. The platform also detects regressions — examples that scored worse than the previous run.

Playground

The Playground lets you test any base or fine-tuned model interactively via Together AI serverless inference. You can run a single model or enable compare mode to see two models side by side with the same prompt. Each response shows latency and token usage metrics. See the Playground API reference for details.

Workflow

Define — Create a behaviour spec describing what your model should do
Run — Start a run. The platform compiles your examples, augments them with AI, fine-tunes the model, and evaluates it.
Inspect — Review per-example pass/fail results and check for regressions
Test — Use the Playground to compare your fine-tuned model against the base model with real prompts
Iterate — Refine your spec and run again

Supported Models

Model	Context Length
`Qwen/Qwen2.5-7B-Instruct-Turbo`	32,768
`Qwen/Qwen2.5-72B-Instruct`	32,768
`Qwen/Qwen3.5-35B-A3B`	65,536
`meta-llama/Llama-3.2-3B-Instruct`	131,072
`meta-llama/Meta-Llama-3.1-8B-Instruct-Reference`	131,072
`meta-llama/Llama-3.3-70B-Instruct-Turbo`	131,072
`mistralai/Mistral-7B-Instruct-v0.3`	32,768

API & CLI

Every feature is accessible through the REST API or the tt CLI. The dashboard is a convenience layer on top.

REST API — Base URL: https://api.tunedtensor.com/v1. All endpoints accept Authorization: Bearer tt_... API key authentication or session cookies from the dashboard.
CLI — Install with npm install -g @tuned-tensor/cli. See the CLI Tool docs for installation and usage.