Benchmark Decision Forge

It's 9pm. You just swapped to that new quant variant everyone's hyping on Reddit. You ask your claw to do the same task it nailed yesterday. Complete gibberish. You try a different prompt. Still wrong. You close the laptop — was it the 4-bit quantization? The benchmark mismatch? You have no idea.

Install Benchmark Decision Forge and tell your claw "Help me pick the right model for image classification." It walks through your actual use case, compares benchmarks on identical tasks, checks quant settings against your real workflow, and gives you a evidence-backed recommendation in seconds.

What you get:

• benchmark_forge.py — Core CLI that runs side-by-side model comparisons
• config.yaml — Default settings for quantization, tasks, and model pools
• quant_tester.py — Tests different quantization levels against your actual workload
• benchmark_parser.py — Converts raw benchmark results into comparison reports
• preference_store.json — Stores your model preferences and use case history
• report_generator.py — Creates clean markdown/HTML comparison reports
• regressor_check.py — Catches performance drops when you swap models
• README.md — Setup and quick-start guide