mlship v0.2.0 - Benchmarking

What's New

Built-in Benchmarking (mlship benchmark)

Measure model serving performance with a single command. Closes #1.

mlship benchmark model.pkl
mlship benchmark model.pkl --requests 1000 --warmup 10
mlship benchmark distilbert-base-uncased-finetuned-sst-2-english --source huggingface
mlship benchmark model.pkl --output json > results.json

Metrics reported: cold start latency, avg/min/p50/p95/p99/max latency, and throughput (requests/sec).

All frameworks supported: sklearn, PyTorch, TensorFlow, HuggingFace — with auto-generated test payloads.

CLI options:

Option	Default	Description
--requests N	100	Number of benchmark requests
--warmup N	5	Number of warmup requests
--port PORT	8000	Server port
--payload JSON	auto	Custom test payload
--source	local	Model source (local or huggingface)
--output	text	Output format (text or json)

Other Changes

• Updated documentation across README, ARCHITECTURE, and WHY_MLSHIP
• Added comprehensive test suite (unit + integration tests across all frameworks)
• Fixed request timeouts for slow models (e.g., GPT-2 on CPU)

Full Changelog: v0.1.5...v0.2.0