mirror of
https://github.com/headroomlabs-ai/headroom.git
synced 2026-08-27 14:17:10 -04:00
20 lines
688 B
Python
20 lines
688 B
Python
|
"""Tests for local token/cache mode benchmark harness."""
|
||
|
|
|
||
|
|
from benchmarks.proxy_mode_benchmark import run_local_benchmark
|
||
|
|
|
||
|
|
|
||
|
|
def test_local_mode_benchmark_shows_compression_and_cache_tradeoff() -> None:
|
||
|
|
results = run_local_benchmark(turns=6)
|
||
|
|
|
||
|
|
baseline = results["baseline"]
|
||
|
|
token = results["token"]
|
||
|
|
cache = results["cache"]
|
||
|
|
|
||
|
|
assert token.total_tokens_saved > 0
|
||
|
|
assert cache.total_tokens_saved > 0
|
||
|
|
assert token.total_sent_tokens < baseline.total_sent_tokens
|
||
|
|
assert cache.total_sent_tokens < baseline.total_sent_tokens
|
||
|
|
|
||
|
|
# Cache mode should preserve prefix better than token mode.
|
||
|
|
assert cache.total_cache_read_tokens >= token.total_cache_read_tokens
|