DEDUPED Hot AI repos, turned into use cases Get 5 a day

How to compare local models on your card before you commit with Poly Strata

Switch between installed models on a running server to see which speed and answers suit you.

Who it is forA user who has several models downloaded and wants to choose one for daily work based on speed and answer quality on their own hardware.

The problemPublished speed numbers come from other PCs, so it is hard to know which model file runs well on your card.

The outcomeYou have a short list of models and file sizes that work on your PC, with your own rough speed notes.

What you need first

  • A Windows or Linux PC with an NVIDIA card, or an AMD card with 12 GB or more for Qwen3.8-Flash-Next
  • Enough disk space for the model file you pick, from about 11 GB to 114 GB depending on the model
  • Enough RAM for the model you pick, since large models such as GLM 5.3 Flash and DeepSeek V4 Flash keep most of their file in RAM
  • A coding agent that can point at a custom API base, if you want to use one with the local server

Steps

  1. Install Poly Strata with the one-click installer and download two or three models in different file sizes.
  2. Start the server with the first model and send it the same prompt you use day to day.
  3. Change to the next installed model while the server keeps running, using the model switch described in the docs.
  4. Note the answer speed and quality for each model and keep the one that fits your card.

Prompt for your agent

With the PolyStrata server running on localhost, send the same prompt <test prompt> to each installed model in turn, switching models between runs. Expected output: a table of model name, approximate tokens per second, and a short quality note for each. Verify that every model answered the same prompt and that the speeds were measured on this PC.

Caveat. Model switching is described in docs/BACKENDS.md, and the eight non-Qwen3.8 models are experimental on the Poly Strata engine.

Share: X LinkedIn Hacker News Email Copy link

Disclaimer

Deduped lists open-source repositories that belong to other people. We do not own, maintain, audit, or vouch for any of them, and a repo appearing here is not an endorsement or a safety check. The use cases, steps, and prompts are written by AI and may be incomplete, outdated, or just wrong. Where a use case is marked "Seen in the wild", we link to public posts where someone described something similar, but we have not verified those posts and the use case itself is still our AI's own write-up. Use all of this at your own risk. Only run code, commands, or agents on systems and accounts you are authorized to use, and read the code and every command before you run it, especially anything that touches credentials, files, or the network. Nothing here is professional, security, legal, or financial advice. All trademarks and project names belong to their respective owners, and their use here does not imply affiliation. Content is provided as is, without warranties of any kind. Found a mistake, a safety problem, or something you want removed? Email [email protected] and we will look at it promptly. Full disclaimer page