 ██ ██ ▓▓ ██   ModelRoom
 ██ ██ ██ ░░   Which local model packages fit your machine.
 ▓▓ ██ ░░ ░░   modelroom 1.0.0 · github.com/jm-vis/modelroom

 folder    not chosen yet          the first question asks where results live
 daemon    Ollama                  not reachable
 machine   workstation             not measured in this results folder yet

 Five steps: configuration, packages, context, measurement, results.
 Files are written as the run goes on. Esc leaves a list, Ctrl-C leaves at any point.

 ── Step 1 of 5  Configuration ──────────────────────────────────────────
 ? Where should results live?  this folder
   another computer: pick its profile below, or import the file made there
 ? Which machines should the result cover?  this machine, a machine entered by hand
 ? What kind of machine is this?  laptop
   graphics  NVIDIA RTX PRO 3000 Blackwell Generation Laptop GPU · 12.0 GB · nvidia-smi · usable (CUDA)
   graphics  Intel(R) Graphics · shared memory · windows · not counted: integrated graphics shares the system memory
   measured: laptop, one graphics card, 12 GB, 128 GB memory, confirmed by llmfit
 ? What is the machine called?  not kept: the machine gets a neutral name
 ? What kind of machine is it?  workstation
 ? How much memory does it have, in GiB?  32 GiB
 ? What runs the model?  unified memory (system and graphics share one pool)
   entered workstation-00000000: workstation, 32 GiB memory, shared memory
 ✓ Configuration   <results>, 2 machines

 ── Step 2 of 5  Packages ───────────────────────────────────────────────
 ? What are you looking for?  qwen
 ? Show only repositories of a publisher or a listed packager?  Yes
   catalog: shipped <day>
   searched Hugging Face at the publisher Qwen and the five listed packagers
   4 repositories, 2 of them models you can pick from
 ? Which of these models should the result cover?  Qwen3.5-9B, DeepSeek-R1-0528-Qwen3-8B
 ✓ Packages        2 models, 24 packages from 3 repositories

 ── Step 3 of 5  Context ────────────────────────────────────────────────
 ? How many people use it on a typical day?  1 (1 request)
 ? How much text should a model handle at once?  L  32k  24,000 words  a report or a long contract
 ✓ Context         L 32k

 ── Step 4 of 5  Measurement ────────────────────────────────────────────
 ? Measure the speed on this machine now?  No
   no load test: cannot reach the Ollama daemon at http://127.0.0.1:11434: http://127.0.0.1:11434: the Ollama daemon did not answer (refused)
 – Measurement     none in this run

 ── Step 5 of 5  Results ────────────────────────────────────────────────

 folder    <results>
 machine   laptop-00000000         laptop, 12 GB graphics, 128 GB memory, measured today
 machine   workstation-00000000 (entered)  workstation, shared memory 32 GB, entered
 models    Qwen3.5-9B, DeepSeek-R1-0528-Qwen3-8B  24 packages from unsloth and Ollama
 context   32k (entered), KV cache f16 (assumed), 1 request (from 1 user)
 speed     not measured            say Yes in step 4 to measure an installed package
 result    docs\models.md          24 packages ranked, 0 no fit

 laptop-00000000 · context L 32k
 room      graphics 11.0 GB for models (12.0 GB card minus reserve)
           memory 111.7 GB for models (127.7 GB minus reserve)
           disk 41.2 GB free on C: (Ollama models)
 #    Model                   Package                Fit          Speed         Memory    Download
 ───────────────────────────────────────────────────────────────────────────────────────────────────
 1    DeepSeek-R1-0528-Qwen3  unsloth · Q4_K_M       tight        –             10.2 GB   4.7 GB
 2    Qwen3.5-9B              unsloth · UD-Q6_K_XL   tight        –             10.5 GB   8.2 GB
 shown     2 rows of 24 packages, one row per model
           ranked for this machine; another machine may show other rows
 memory    #1–2 fit into graphics memory, budget 11.0 GB (card minus reserve, from the profile)
 speed     nothing measured in the rows shown · say Yes in step 4 to measure an installed package
 key       Fit is memory only: the package's need against a pool · ample: up to 60 % of the pool
           fits: up to 85 % · tight: up to 98 % · no fit: more
           the pool is what the mode names, minus its reserve (per card) · gpu: one card
           gpu_split: all cards together · cpu_gpu, cpu (RAM): the system memory
           system memory keeps a quarter as reserve, 2 to 16 GB, unless modelroom.toml sets it
           one memory shared by system and card counts as gpu, a quarter kept, no upper limit
           off the card, and for a fit from size the highest class is fits
           best: the least compression of the same model that fits
           modelroom rates no model family against another
           one row per model, its least compression that fits
           a run with one model shows its first 10 packages, then the measured ones
           Speed is measured, see its column.

 install   #1  ollama pull hf.co/unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_K_M

 workstation-00000000 (entered) · another computer · context L 32k
 room      shared memory 23.0 GB for models (32.0 GB minus both reserves)
           disk unknown on another computer
 #    Model                   Package                Fit          Speed         Memory    Download
 ───────────────────────────────────────────────────────────────────────────────────────────────────
 1    DeepSeek-R1-0528-Qwen3  unsloth · Q4_K_M       ample        –             10.2 GB   4.7 GB
 2    Qwen3.5-9B              unsloth · BF16         tight        –             19.9 GB   16.7 GB
 shown     2 rows of 24 packages, one row per model
           ranked for this machine; another machine may show other rows
 memory    #1–2 fit into shared memory, budget 23.0 GB
           (memory minus both reserves, from the profile)
 speed     nothing measured in the rows shown · a machine entered by hand is never measured
 key       Fit is memory only: the package's need against a pool · ample: up to 60 % of the pool
           fits: up to 85 % · tight: up to 98 % · no fit: more
           the pool is what the mode names, minus its reserve (per card) · gpu: one card
           gpu_split: all cards together · cpu_gpu, cpu (RAM): the system memory
           system memory keeps a quarter as reserve, 2 to 16 GB, unless modelroom.toml sets it
           one memory shared by system and card counts as gpu, a quarter kept, no upper limit
           off the card, and for a fit from size the highest class is fits
           best: the least compression of the same model that fits
           modelroom rates no model family against another
           one row per model, its least compression that fits
           a run with one model shows its first 10 packages, then the measured ones
           Speed is measured, see its column.

 install   #1 on workstation-00000000 (entered)  ollama pull hf.co/unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_K_M
           run it on that computer, or choose a package for it below
 Quality of these models for your task: https://artificialanalysis.ai

 ✓ Results         docs\models.md
