needle / scripts /audit_native_budget.py
Leon4gr45's picture
Upload folder using huggingface_hub (part 2)
25a2c1c verified
Raw History Blame Contribute Delete
1.03 kB
"""Replay the captured pre-change native prompt; never execute returned calls."""
import argparse,json,time
from pathlib import Path
import needle
def main():
p=argparse.ArgumentParser();p.add_argument('--output',default='/tmp/needle-budget-replay.json');a=p.parse_args()
recorded=json.loads(Path('benchmarks/workflow/results/before-budget-probe.json').read_text())
prompt=recorded['effective_prompt'];rows=[]
for budget in (512,1024,2048):
agent=needle.Needle(tools=prompt['tools'],system=prompt['system'],generation=3)
start=time.perf_counter();result=agent.complete(prompt['text'],max_new_tokens=budget)
rows.append({'public_sdk_max_new_tokens':budget,'latency_ms':(time.perf_counter()-start)*1000,'native':result})
Path(a.output).write_text(json.dumps({'methodology':'Exact recorded pre-change native prompt; official Needle 3.0.1 default cache. Host output-budget probe, no tool execution.','effective_prompt':prompt,'rows':rows},indent=2)+'\n')
if __name__=='__main__':main()