69 lines
2.0 KiB
Python
Executable File
69 lines
2.0 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
"""Run the DS4 token oracles, then the existing endpoint smoke corpus."""
|
|
|
|
import argparse
|
|
import pathlib
|
|
import subprocess
|
|
import sys
|
|
|
|
|
|
ROOT = pathlib.Path(__file__).resolve().parent.parent
|
|
HARDWARE_TESTS = (
|
|
"flash_resident_and_ssd_streaming_choose_the_same_tokens",
|
|
"flash_ssd_streaming_maps_batched_prefill_layers",
|
|
"resident_multi_session_switching_preserves_each_kv_frontier",
|
|
"legacy_mtp_runs_a_target_owned_greedy_cycle",
|
|
"dspark_runs_a_target_owned_greedy_cycle",
|
|
"ssd_streaming_supports_legacy_mtp_and_dspark",
|
|
"directional_steering_matches_the_ds4_token_oracle",
|
|
"resident_and_streamed_glm_match_ds4_decode_oracles",
|
|
"streamed_glm_uses_ds4_indexed_prefill_for_long_prompts",
|
|
"glm_mtp_preserves_target_tokens_and_drafts",
|
|
"pro_ssd_streaming_full_layer_and_selected_batch_agree",
|
|
)
|
|
ENDPOINT_SCRIPTS = (
|
|
"endpoint_parity.py",
|
|
"endpoint_reasoning.py",
|
|
"endpoint_continuation.py",
|
|
)
|
|
|
|
|
|
def run(command):
|
|
print("+", " ".join(map(str, command)), flush=True)
|
|
subprocess.run(command, cwd=ROOT, check=True)
|
|
|
|
|
|
def main():
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument(
|
|
"--endpoint",
|
|
help="also run the existing endpoint parity scripts against this URL",
|
|
)
|
|
parser.add_argument(
|
|
"--skip-hardware",
|
|
action="store_true",
|
|
help="skip tests requiring the local Flash/MTP/DSpark GGUF fixtures",
|
|
)
|
|
args = parser.parse_args()
|
|
|
|
if not args.skip_hardware:
|
|
for test in HARDWARE_TESTS:
|
|
run(
|
|
[
|
|
"cargo",
|
|
"test",
|
|
"--all-features",
|
|
test,
|
|
"--",
|
|
"--ignored",
|
|
"--nocapture",
|
|
]
|
|
)
|
|
if args.endpoint:
|
|
for script in ENDPOINT_SCRIPTS:
|
|
run([sys.executable, ROOT / "scripts" / script, args.endpoint])
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|