DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to measure which local model runs well on your Mac with Whallm

Use the Throughput tool to test prefill and decode speed on your own hardware before committing to a large download.

Short answer: Install Whallm and download a model from the Model page, as described in the Quick start; then open Model and check Advanced Settings for the cache mode and slot count you want to test; open the Throughput tool in the app and run it on your Mac using the 16K-token input and 512-token answer from the README benchmark, if the tool allows it. All steps.

Who it is forA Mac owner deciding between Whallm models or cache settings who wants real numbers for their own chip and SSD.

The problemBenchmark numbers from other machines do not show how a given model, cache mode and SSD will perform on this Mac.

The outcomeThe user has prefill and decode speeds and peak memory for their own Mac, to compare against the README benchmark table.

What you need first

  • Apple Silicon Mac running macOS 15 or later
  • 64 GiB memory recommended, with more disk space than the chosen model's files (Qwen models about 118 to 119 GiB, DeepSeek V4.1 above 458 GiB)
  • Fast internal, Thunderbolt or USB4 SSD
  • Network access to download the Whallm release and model weights

Steps

  1. Install Whallm and download a model from the Model page, as described in the Quick start.
  2. Open Model and check Advanced Settings for the cache mode and slot count you want to test.
  3. Open the Throughput tool in the app and run it on your Mac using the 16K-token input and 512-token answer from the README benchmark, if the tool allows it.
  4. Compare your prefill tok/s, decode tok/s and peak memory with the BENCHMARK.md results and note the cache mode used.

Prompt for your agent

In the yanun0323/Whallm repo, read the Benchmark summary and the BENCHMARK.md file and list the metrics and settings a user must record to compare a run with the published results. Output a short checklist with the chip, cache mode, slots, prefill, decode and peak memory, and note that the README calls these numbers reference only.

Caveat. The README says the published benchmarks are user-provided and not a controlled comparison, and peak memory is process memory rather than total Mac memory.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for yanun0323/Whallm and "Measure which local model runs well on your Mac"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer