Datasets:
File size: 7,834 Bytes
5db0974 92dd734 5db0974 b9b8a4a 92dd734 5db0974 b9b8a4a 92dd734 5db0974 b9b8a4a 5db0974 b9b8a4a 5db0974 b9b8a4a 5db0974 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 | #!/usr/bin/env python3
"""Aggregate raw layerstream-plugin, stock llama.cpp, Hugging Face accelerate, and AirLLM
JSON benchmark outputs into a normalized JSON + JSONL + Parquet Hugging Face dataset bundle.
Usage:
python3 benchmarks/export_hf_dataset.py \
--results-dir benchmarks/results \
--out-dir benchmarks/hf-dataset
"""
from __future__ import annotations
import argparse
import json
from pathlib import Path
from typing import Any
def extract_json_array(raw: str) -> list[dict[str, Any]]:
"""Extract the trailing JSON array from stdout that may include bracketed library banners."""
pos = len(raw)
while True:
idx = raw.rfind("[", 0, pos)
if idx == -1:
return []
candidate = raw[idx:].strip()
try:
data = json.loads(candidate)
if isinstance(data, list) and (not data or isinstance(data[0], dict)):
return data
except json.JSONDecodeError:
pass
pos = idx
DEFAULT_HARDWARE = "NVIDIA Jetson Orin Nano (8 GiB unified LPDDR5, CUDA sm_87, dm-crypt NVMe)"
DEFAULT_HARDWARE_ID = "orin-nano-8gb-cuda"
def load_dir_hardware(round_dir: Path) -> tuple[str, str]:
meta_path = round_dir / "hardware.json"
if meta_path.is_file():
try:
meta = json.loads(meta_path.read_text(encoding="utf-8"))
if isinstance(meta, dict):
hw = meta.get("hardware") or DEFAULT_HARDWARE
hw_id = meta.get("hardware_id") or DEFAULT_HARDWARE_ID
return str(hw), str(hw_id)
except (OSError, json.JSONDecodeError):
pass
return DEFAULT_HARDWARE, DEFAULT_HARDWARE_ID
def normalize_record(
rec: dict[str, Any],
source_file: str,
round_name: str,
default_hw: str = DEFAULT_HARDWARE,
default_hw_id: str = DEFAULT_HARDWARE_ID,
) -> dict[str, Any]:
engine = rec.get("engine")
default_ver = "0.1.0" if engine == "layerstream-plugin" else None
return {
"round": round_name,
"source_file": source_file,
"hardware": rec.get("hardware") or default_hw,
"hardware_id": rec.get("hardware_id") or default_hw_id,
"engine": engine,
"engine_version": rec.get("engine_version", default_ver),
"git_rev": rec.get("git_rev"),
"mode": rec.get("mode"),
"status": rec.get("status", "ok"),
"model": rec.get("model"),
"draft_model": rec.get("draft_model"),
"workload": rec.get("workload"),
"chat_template_applied": rec.get("chat_template_applied"),
"warmup": rec.get("warmup"),
"reps": rec.get("reps"),
"temperature": rec.get("temperature"),
"seed": rec.get("seed"),
"threads": rec.get("threads"),
"kv_cache_type": rec.get("kv_cache_type"),
"vram_budget_auto": rec.get("vram_budget_auto"),
"vram_budget_mb": rec.get("vram_budget_mb"),
"prompt_tokens": rec.get("prompt_tokens"),
"completion_tokens": rec.get("completion_tokens"),
"load_time_s": rec.get("load_time_s"),
"ttft_s": rec.get("ttft_s"),
"ttft_s_min": rec.get("ttft_s_min"),
"ttft_s_max": rec.get("ttft_s_max"),
"decode_time_s": rec.get("decode_time_s"),
"total_time_s": rec.get("total_time_s"),
"decode_tok_per_s": rec.get("decode_tok_per_s"),
"decode_tok_per_s_min": rec.get("decode_tok_per_s_min"),
"decode_tok_per_s_max": rec.get("decode_tok_per_s_max"),
"total_tok_per_s": rec.get("total_tok_per_s"),
"total_tok_per_s_min": rec.get("total_tok_per_s_min"),
"total_tok_per_s_max": rec.get("total_tok_per_s_max"),
"sweeps": rec.get("sweeps"),
"drafted_tokens": rec.get("drafted_tokens"),
"accepted_tokens": rec.get("accepted_tokens"),
"resident_layers": rec.get("resident_layers"),
"streamed_layers": rec.get("streamed_layers"),
"n_layers": rec.get("n_layers"),
"resident_mib": rec.get("resident_mib"),
"slot_mib": rec.get("slot_mib"),
"expert_slot_mib": rec.get("expert_slot_mib"),
"io_mib_read": rec.get("io_mib_read"),
"pread_mib_s": rec.get("pread_mib_s"),
"effective_io_mib_s": rec.get("effective_io_mib_s"),
"expert_cache_hits": rec.get("expert_cache_hits"),
"expert_cache_misses": rec.get("expert_cache_misses"),
"expert_cache_hit_rate": rec.get("expert_cache_hit_rate"),
"expert_cache_mib_saved": rec.get("expert_cache_mib_saved"),
"moe_layers": rec.get("moe_layers"),
"moe_active_streaming": rec.get("moe_active_streaming"),
"uma_zero_copy": rec.get("uma_zero_copy"),
"is_unified_memory": rec.get("is_unified_memory"),
"direct_io_active": rec.get("direct_io_active"),
"peak_rss_mib": rec.get("peak_rss_mib"),
"current_rss_mib": rec.get("current_rss_mib"),
"mem_avail_before_mib": rec.get("mem_avail_before_mib"),
"mem_avail_after_mib": rec.get("mem_avail_after_mib"),
"swap_delta_mib": rec.get("swap_delta_mib"),
"cuda_max_allocated_mib": rec.get("cuda_max_allocated_mib"),
"page_cache_delta_mib": rec.get("page_cache_delta_mib"),
"page_cache_delta_since_start_mib": rec.get("page_cache_delta_since_start_mib"),
"idle_watts": rec.get("idle_watts"),
"avg_watts": rec.get("avg_watts"),
"net_watts": rec.get("net_watts"),
"joules_per_tok": rec.get("joules_per_tok"),
"net_joules_per_tok": rec.get("net_joules_per_tok"),
"turn1_prompt_tokens": rec.get("turn1_prompt_tokens"),
"turn1_ttft_s": rec.get("turn1_ttft_s"),
"turn2_prompt_tokens": rec.get("turn2_prompt_tokens"),
"turn2_ttft_s": rec.get("turn2_ttft_s"),
"ttft_speedup_x": rec.get("ttft_speedup_x"),
"notes": rec.get("notes"),
"output_preview": rec.get("output_preview") or rec.get("turn2_output_preview"),
}
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("--results-dir", type=Path, required=True)
parser.add_argument("--out-dir", type=Path, required=True)
args = parser.parse_args()
records: list[dict[str, Any]] = []
for path in sorted(args.results_dir.rglob("*.json")):
if path.name == "hardware.json" or path.stat().st_size == 0:
continue
raw = path.read_text(encoding="utf-8", errors="replace")
try:
items = extract_json_array(raw)
except json.JSONDecodeError as exc:
print(f"Skipping malformed JSON {path}: {exc}")
continue
round_name = path.parent.name
dir_hw, dir_hw_id = load_dir_hardware(path.parent)
for item in items:
records.append(normalize_record(item, path.name, round_name, dir_hw, dir_hw_id))
args.out_dir.mkdir(parents=True, exist_ok=True)
data_dir = args.out_dir / "data"
data_dir.mkdir(parents=True, exist_ok=True)
json_path = data_dir / "benchmarks.json"
json_path.write_text(json.dumps(records, indent=2) + "\n", encoding="utf-8")
print(f"Wrote {len(records)} normalized records -> {json_path}")
jsonl_path = data_dir / "benchmarks.jsonl"
with jsonl_path.open("w", encoding="utf-8") as f:
for rec in records:
f.write(json.dumps(rec) + "\n")
print(f"Wrote JSONL -> {jsonl_path}")
try:
import pyarrow as pa
import pyarrow.parquet as pq
table = pa.Table.from_pylist(records)
parquet_path = data_dir / "train-00000-of-00001.parquet"
pq.write_table(table, parquet_path)
print(f"Wrote Parquet -> {parquet_path}")
except ImportError:
print("pyarrow not installed; skipped Parquet export (JSON/JSONL written).")
if __name__ == "__main__":
main()
|