Contents TandemLLM Docs
v0.11.2
Star58

From the repository, docs/gallery/README.md

Gallery

The dashboard at 1440×900 and on a 390×844 phone, dark theme. dashboard/e2e/readme-shots.spec.ts renders the pictures from the dashboard's mock, and dashboard/scripts/readme-shots.sh converts them to WebP. The Setup Guide speeds come from the measured rows in models/*/model.yaml. All other numbers come from the mock.

Chat

The Chat tab: a web page the model wrote in its sandbox, open in the preview pane, with the context panel

The model writes site/index.html in its sandbox, the preview pane renders it, and the context panel shows what fills the window.

Setup Guide

The models step: one tile per model with its measured speed

The models step: one tile per model, each with its measured speed in tok/s.

Nemotron 3.5 Bolt's speed table: tok/s and KL for each size

Each size's tok/s and KL, here Nemotron 3.5 Bolt.

Community weights: turboderp's Qwen3.8-Flash-Next EXL3 packs

Community weights with their maker: turboderp's Flash-Next packs.

Performance and Usage

Performance: live decode speed and each running request

Live decode speed and every running request.

Performance history: decode speed, time to first token, prefill, tokens per block, draft acceptance

History: decode speed, first-token time, prefill, tokens per block, draft acceptance.

Usage: tokens per day over a year

Tokens per day for a year.

Phone

The Chat tab on a phone: an answer with code    The Chat tab on a phone: a run with a tool call

The Chat tab on a phone: an answer with code, and a run calling a tool.