Skip to main content

​
Overview

check runs a full hardware analysis and scores compatible models across four dimensions: Quality, Speed, Fit, and Context. It is the primary command for answering “what can I run on this machine right now?” 
llm-checker check

​
Example Output

SYSTEM SUMMARY
──────────────────────────────────────────────────
CPU: Apple M4 Pro (12 cores)
Memory: 24GB RAM
GPU: Apple M4 Pro
Architecture: ARM64
Hardware Tier: HIGH

RECOMMENDED MODEL
──────────────────────────────────────────────────
Model: Qwen 2.5 Coder 14B
Size: ~9GB (Q4_K_M)
Compatibility Score: 82/100
Fine-tuning: LoRA+QLoRA
Reason: Highest compatibility score for your hardware
Estimated Speed: 28 tokens/sec
Runtime: Ollama
Status: Available for installation

Pull: ollama pull qwen2.5-coder:14b

QUICK START
──────────────────────────────────────────────────
1. Install the recommended model:
   ollama pull qwen2.5-coder:14b
2. Start using it:
   ollama run qwen2.5-coder:14b

​
Flags

​
-u, --use-case
string
Filter and weight recommendations by use case. Accepted values: general, coding, chat, reasoning, multimodal, embeddings, creative.Default: general
​
--runtime
string
Specify the inference runtime. Accepted values: ollama, vllm, mlx, llama.cpp.Default: ollama
​
-l, --limit
number
Number of compatible models to display in output.Default: 1
​
--policy
string
Path to a policy.yaml file. Evaluates all candidate models against enterprise governance rules and reports violations.
​
--no-verbose
flag
Disable the step-by-step progress display. Useful for scripting and CI environments.
​
--max-size
string
Maximum model size to consider. Accepts B (parameters) or GB format, e.g. 30B or 15GB.
​
--min-size
string
Minimum model size to consider. Accepts B or GB format, e.g. 7B or 4GB.
​
--include-cloud
flag
Include cloud-hosted models in the analysis in addition to local Ollama models.
​
--simulate
string
Simulate a hardware profile instead of detecting real hardware. Use --simulate list to see all available profiles.
​
--gpu
string
Custom GPU model for hardware simulation, e.g. "RTX 5060" or "RX 7800 XT".
​
--ram
number
Custom RAM in GB for hardware simulation, e.g. 32.
​
--cpu
string
Custom CPU model for hardware simulation, e.g. "AMD Ryzen 7 5700X".
​
--vram
number
Override GPU VRAM in GB for simulation. Requires --gpu.
​
-d, --detailed
flag
Show detailed hardware information in the output.
​
-f, --filter
string
Filter results by model type.
​
--ollama-only
flag
Only show models that are available in the Ollama registry.
​
--performance-test
flag
Run active performance benchmarks against installed models. Produces real tok/s measurements instead of hardware-estimated speeds.
​
--show-ollama-analysis
flag
Show a detailed breakdown of Ollama model analysis in the output.

​
Use Case Examples

# General analysis
llm-checker check

# Find best coding model, show top 3
llm-checker check --use-case coding --limit 3

# Specify runtime and filter size
llm-checker check --runtime vllm --max-size 24B

# Use enterprise policy governance
llm-checker check --policy ./policy.yaml

# Suppress progress for CI
llm-checker check --use-case coding --no-verbose

# Detailed hardware information
llm-checker check --detailed

# Only show Ollama-available models
llm-checker check --ollama-only

# Run real performance benchmarks
llm-checker check --performance-test

# Simulate an RTX 4090 system
llm-checker check --simulate rtx4090

# Custom hardware simulation
llm-checker check --gpu "RTX 5060" --ram 32 --cpu "AMD Ryzen 7 5700X"

​
Scoring System

check uses the deterministic 4D scoring engine. Weights are adjusted by use case:
CategoryQualitySpeedFitContext
general45%35%15%5%
coding55%20%15%10%
reasoning60%10%20%10%
multimodal50%15%20%15%

​
Policy Enforcement

When --policy <file> is supplied, check evaluates all compatible and marginal candidates discovered during analysis — not just the models shown in the --limit output.
  • In audit mode, policy violations are reported but the command exits with 0.
  • In enforce mode, blocking violations cause a non-zero exit (default 1).
  • Override the exit code with enforcement.exit_code in policy.yaml.
llm-checker check --policy ./policy.yaml --use-case coding --runtime vllm
Run llm-checker check --simulate list to see all built-in hardware profiles for testing without needing the real hardware.

Build docs developers (and LLMs) love

Get started for freeTalk to us