Datasets:
File size: 4,811 Bytes
1836fe9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 | #!/usr/bin/env python3
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: LicenseRef-NVIDIA-OneWay-Noncommercial
"""Stage 1: run a VLM over every (benchmark sample, prompt type) through an API.
Writes <output-dir>/inference/<model_slug>/<prompt_type>/idx_NNNN_<side>.json.
"""
from __future__ import annotations
import argparse
import datetime
import os
import sys
from . import _api
from ._shared import (
PROMPT_TYPE_NAMES,
build_question,
load_eval_items,
result_path,
save_result,
)
def cmd_infer(args) -> None:
"""Run inference on every benchmark sample."""
items = load_eval_items(args.scenarios)
scenarios_dir = os.path.dirname(os.path.abspath(args.scenarios))
if not items:
print("No valid scenarios found.", file=sys.stderr)
sys.exit(1)
prompt_types = args.prompt_type
total = len(items) * len(prompt_types)
done = 0
skipped = 0
print(
f"Inference: {len(items)} eval items x {len(prompt_types)} prompt type(s) "
f"= {total} calls | model={args.model}"
)
for pt in prompt_types:
for item in items:
idx = item["scenario_index"]
side = item["side"]
rpath = result_path(args.output_dir, "inference", args.model, pt, idx, side)
# Resume: skip if already done
if not args.no_resume and os.path.isfile(rpath):
skipped += 1
continue
question = build_question(pt, item["task"])
# Clips ship pre-trimmed to their annotated window and are used as-is.
video = item["video_path"]
if not os.path.isabs(video):
video = os.path.join(scenarios_dir, video)
if not os.path.isfile(video):
print(f" Video not found, skipping: {video}", file=sys.stderr)
continue
done += 1
print(
f" [{done + skipped}/{total}] idx={idx} side={side} pt={pt}",
end="",
flush=True,
)
try:
answer = _api.query_model(
video, question, args.model, args.api_url, args.api_key,
args.input_mode, args.max_frames, robot_type=item["robot"],
)
except SystemExit:
print(" [FAILED]")
continue
save_result(rpath, {
"scenario_index": idx,
"side": side,
"prompt_type": pt,
"model": args.model,
"question": question,
"video_path": item["video_path"],
"model_response": answer,
"timestamp": datetime.datetime.now(datetime.timezone.utc).isoformat(),
})
print(f" [OK] {len(answer)} chars")
print(f"Inference done: {done} new, {skipped} resumed, {total} total.")
def main() -> None:
parser = argparse.ArgumentParser(
description="Run VLM inference on the benchmark videos through an API.",
)
parser.add_argument("--scenarios", required=True, help="Path to benchmark.json")
parser.add_argument(
"--model", required=True, help="Model name as your API endpoint knows it",
)
parser.add_argument(
"--prompt-type", nargs="+", default=list(PROMPT_TYPE_NAMES),
choices=PROMPT_TYPE_NAMES,
help="Prompt type(s) to evaluate (default: all four)",
)
parser.add_argument("--output-dir", default="eval_results", help="Output directory")
parser.add_argument(
"--api-url", default=_api.CHAT_COMPLETIONS_URL,
help="Full chat-completions URL (default: $VLM_API_BASE/v1/chat/completions)",
)
parser.add_argument(
"--input-mode", choices=("auto", "gemini_video", "openai_frames"), default="auto",
help="Send the whole clip ('gemini_video') or sampled frames ('openai_frames'). "
"'auto' sends frames to GPT models and the clip to others.",
)
parser.add_argument(
"--max-frames", type=int, default=16,
help="Frames sent in 'openai_frames' mode, spread evenly over the clip",
)
parser.add_argument("--no-resume", action="store_true", help="Re-run all, ignore existing")
args = parser.parse_args()
args.api_key = os.environ.get("VLM_API_KEY", "")
if args.max_frames < 1:
parser.error("--max-frames must be at least 1")
if not args.api_key.strip():
print("Missing API key: set VLM_API_KEY.", file=sys.stderr)
sys.exit(1)
if not args.api_url:
print("Missing endpoint: set VLM_API_BASE or pass --api-url.", file=sys.stderr)
sys.exit(1)
cmd_infer(args)
if __name__ == "__main__":
main()
|