Skip to content

pmetal eval

Evaluate a model’s perplexity on a dataset to measure generation quality.

Terminal window
pmetal eval \
--model <MODEL> \
--dataset <DATASET> \
[OPTIONS]
Terminal window
# Evaluate perplexity
pmetal eval \
--model Qwen/Qwen3-0.6B \
--dataset eval.jsonl
# Evaluate with LoRA adapter
pmetal eval \
--model Qwen/Qwen3-0.6B \
--dataset eval.jsonl \
--lora ./output/lora_weights.safetensors