pickfit

BrowseLocal AI tools

MFQ

MFQ quantizes large models and runs them locally with native CUDA and Metal runtimes.

Tested on LinuxKnown vulnerabilities

Use it

Before you install: known vulnerabilities

pickfit found 10 known high or critical vulnerabilities in the packages a fresh install brings (OSV.dev, 3 Nov 2026). Weigh this before you install it.

  • crates:zip 8.6.0 zip: Critical Parser Differential & Security Control Bypass via Dual-Header Inconsistencies (Zip Cloaking) Leading to Arbitrary File Overwrite & RCE (GHSA-m665-v7pm-7grj, critical, fixed in 9.0.5)
  • npm:dompurify 3.4.16 DOMPurify SAFE_FOR_TEMPLATES bypass: split template expressions survive in attached shadow root content on DOM output modes (GHSA-cg2g-xh64-v4q7, critical, fixed in 3.4.17)
See 8 more

Linux x86_64

  1. git clone MFQ

    git clone https://github.com/Tylogi/MFQ.git MFQ
    
  2. uv sync base deps

    cd MFQ && uv sync
  3. uv sync --extra train to get torch for quantize

    cd MFQ && uv sync --extra train

Check that it works

cd MFQ && uv run mfq --version 2>&1

The output should contain “mfq 0.1.0”.

First thing to try

cd MFQ && uv run python -c "
import json, os, pathlib
import torch
from safetensors.torch import save_file
# create tiny hf model
out='/tmp/tiny_hf'
os.makedirs(out, exist_ok=True)
config={
  'model_type':'qwen3',
  'hidden_size':128,
  'num_hidden_layers':2,
  'num_attention_heads':4,
  'num_key_value_heads':4,
  'intermediate_size':256,
  'vocab_size':32000,
  'max_position_embeddings':512,
  'hidden_act':'silu',
  'rms_norm_eps':1e-05
}
open(os.path.join(out,'config.json'),'w').write(json.dumps(config))
# create dummy weights matching qwen3?
# Look at expected keys: model.embed_tokens, model.layers.0 etc
sd={}
def rand(shape): return torch.randn(shape, dtype=torch.float16)
sd['model.embed_tokens.weight']=rand((32000,128))
for i in range(2):
    p=f'model.layers.{i}'
    sd[f'{p}.self_attn.q_proj.weight']=rand((128,128))
    sd[f'{p}.self_attn.k_proj.weight']=rand((128,128))
    sd[f'{p}.self_attn.v_proj.weight']=rand((128,128))
    sd[f'{p}.self_attn.o_proj.weight']=rand((128,128))
    sd[f'{p}.mlp.gate_proj.weight']=rand((256,128))
    sd[f'{p}.mlp.up_proj.weight']=rand((256,128))
    sd[f'{p}.mlp.down_proj.weight']=rand((128,256))
    sd[f'{p}.input_layernorm.weight']=rand((128,))
    sd[f'{p}.post_attention_layernorm.weight']=rand((128,))
sd['model.norm.weight']=rand((128,))
sd['lm_head.weight']=rand((32000,128))
save_file(sd, os.path.join(out,'model.safetensors'))
print('done', os.listdir(out))
" 2>&1
cd MFQ && uv run mfq quantize /tmp/tiny_hf /tmp/tiny.mfq --preset S4-M --backend cpu --dry-run 2>&1

From the MFQ folder, run uv run mfq quantize <model_dir> <output.mfq> --preset S4-M --backend cpu --dry-run to preview a quantization.

What it does

  • Convert Hugging Face checkpoints to MFQ format
    How to use it
    1. Preview the job with uv run mfq quantize <model_dir> <output.mfq> --preset S4-M --backend cpu --dry-run.
    2. Quantize with uv run mfq quantize <model_dir> <output.mfq> --preset S4-M --backend cpu --overwrite.
    3. Look for "status": "ok" and the output path in the result.
    Tested
  • Run large models on your own hardware
    Not tested
  • Download and run models in MFQ Studio
    Not tested
  • Serve models through an OpenAI-compatible API
    Not tested

How far pickfit got

Linux x86_64

  1. Installs
  2. Starts
  3. 1 of 4 features work
  4. Works offline

Good for

  • run large models on local hardware
  • download and run models in MFQ Studio
  • serve models via OpenAI-compatible API

An alternative to Strata, llama.cpp and oMLX.

How pickfit tested it

Each run starts from a clean machine and follows the steps above. How pickfit verifies

Linux x86_64

Its main function works

  • Replayed from scratch
  • Works offline
  • Checked on 1 Nov 2026
  • Version 0.1.0

Differences the test machine hit

  • README: uv sync + uv run mfq build --backend auto works on any host
    In the test: build requires CUDA toolkit nvcc/CMake or Apple Silicon Metal; on plain Linux x86 without accelerator it fails with 'no supported inference accelerator was detected'
  • README: uv run mfq serve starts server
    In the test: serve also auto-detects backend and fails with same BuildError when no accelerator present, so F3 cannot be tested in this sandbox
  • README: quantize from HF safetensors
    In the test: quantize --backend cpu works with torch+extra train deps, verified with tiny synthetic Qwen3 checkpoint; README example uses --backend auto which would require accelerator

How to read a tool card

Stitching
pickfit installed this tool itself. Everything under the stitches comes from that run. A tool pickfit has not run never gets stitches.
Verified on your OS
The install steps ran on the system you chose. “Ran on Linux only” means it has not been run on yours yet.
macOS, Linux and WindowsWhere it ran
Green OS marks: pickfit ran the install steps there. Grey ones: the README says it runs there, and pickfit has not tried.
How far the run got
Installs and starts, its main function works, its service answers, it produces model output. A window means a desktop app opened; a plug, an MCP client used it. The mark sits beside the run’s result, under “Why it fits and how we know” and in “How pickfit tested it”.
Visual check
pickfit looked at screenshots of the app. Green: it looks right. Amber with a number: things that look wrong; open it to see them. Grey and dashed: not checked yet.
Known vulnerabilities
Known high or critical vulnerabilities (OSV.dev) in what it installs. Green: none known. Red with a number: how many; open it for the list, and read it before you install. Grey and dashed: not checked, which does not mean safe.
About the run
The day it ran, and the version that ran.
Source of a reason
Every reason a pick fits rests on one profile field. Open the reason to see the field and the value it relies on.
Hidden gem
High quality, little known.
Quality and popularity
Two scores out of 100. Quality: how well it installed and worked when pickfit tried it, and how good its install docs and upkeep are. Popularity: from stars, downloads and Hacker News.
Features checked
Of the main features the README advertises, how many the run actually exercised (green), and how many failed (red). Grey ones were not checked, and the card says why.
Not verified
A well-known tool Fitz mentions from general knowledge. No run record, no steps, no score.