path stringlengths 23 47 | content stringlengths 183 97.7k |
|---|---|
tables/leaderboard_k8.csv | method,behaviour,waveform,waveform_rank,behaviour_rank
xgb,0.8395719489997332,0.022831001381045237,1.0,7.0
tabpfn,0.06135674665373589,0.023608787420868634,2.0,1.0
knn,0.21408181836965717,0.03238496173122841,3.0,3.0
mlp,0.21387545410592407,0.037030248275181116,4.0,2.0
ridge,1.7352645938665583,0.07411999069882372,5.0,11.... |
tables/support_budget_results.csv | axis,method,K,n_entries,mean,median,p90,p95,repeat_sd,bootstrap_lo,bootstrap_hi
waveform,arx,2,30,28.97070310960401,0.01689651798597956,0.35770909874025786,0.383192024722958,12.112094104871376,0.05399166282290052,86.74770575474147
waveform,arx,4,30,24.34532930417188,0.015356103503077237,0.3403043484708243,0.35898874689... |
tables/waveform_entry_results.csv | method,K,entry_id,nrmse
arx,2,amp_004_folded_cascode,0.0011715768821226877
arx,2,amp_032_ti_ldo_error_selfbias,0.006250496529639271
arx,2,amp_033_ti_ldo_ref_selfbias,0.0118185536419581
arx,2,ldo_007_pmos,0.0011362971803403324
arx,2,ldo_008_fer_mirror_ota,0.009726838760142648
arx,2,ldo_009_fer_5t_pass,0.0034387424622742... |
tables/behavior_entry_results.csv | "method,K,entry_id,error_normalized\narx,2,amp_004_folded_cascode,0.0012537240731341447\narx,2,amp_0(...TRUNCATED) |
tables/family_split.csv | "family,total,train,validation,test\r\nAmplifier / OTA / op-amp,28,20,1,7\r\nRegulator,14,6,3,5\r\nS(...TRUNCATED) |
tables/resource_statistics.json | "{\n \"traces\": 152273,\n \"entries\": 80,\n \"points\": 11313572202,\n \"duration_sum_s\": 120(...TRUNCATED) |
supplement_data/partition_entries.csv | "entry_id,core_group,partition,reporting_family,fixed_design_tasks\r\namp_004_folded_cascode,core_04(...TRUNCATED) |
supplement_data/task_inventory.csv | "task_id,entry_id,design_id,partition,core_group,support_pool_count,query_count\r\namp_004_folded_ca(...TRUNCATED) |
supplement_data/entry_labels.csv | "entry_id,display_name\r\namp_004_folded_cascode,Amplifier (004)\r\namp_032_ti_ldo_error_selfbias,Am(...TRUNCATED) |
supplement_data/resource_source_composition.csv | "source,entries,design_identifiers,trajectories\r\nanalogtobi,38,3700,89568\r\nspicexplorer,16,250,6(...TRUNCATED) |
AnalogDynBench
A benchmark for few-shot analog transient response prediction and behavioral evaluation.
Release status
Public release is in progress. This initial snapshot provides numerical result tables and task/partition metadata accompanying the manuscript. It is not the complete benchmark release.
Available now
Download numerical_companion.jsonl from the Files and versions tab. Each JSONL record stores an original relative file path and its UTF-8 content, preserving the numerical files listed below:
tables/leaderboard_k8.csv: the principal waveform leaderboard.tables/support_budget_results.csv: waveform and behavioral comparisons across support budgets.tables/waveform_entry_results.csvandtables/behavior_entry_results.csv: entry-level results.tables/family_split.csvandtables/resource_statistics.json: resource composition and scale.supplement_data/partition_entries.csv,task_inventory.csv, andentry_labels.csv: partition/task inventories and readable entry labels.supplement_data/resource_source_composition.csv,source_inventory.csv, andsource_inventory_notes.txt: source information and its recorded coverage.
Numerical values retain their existing precision. This snapshot contains the accompanying results and inventories; it does not contain all data or software needed to rerun the benchmark.
To recover the original CSV/JSON/text files after downloading:
import json
from pathlib import Path
for line in Path("numerical_companion.jsonl").read_text().splitlines():
item = json.loads(line)
path = Path(item["path"])
path.parent.mkdir(parents=True, exist_ok=True)
path.write_bytes(item["content"].encode("utf-8"))
The dataset viewer displays packaged files, not individual benchmark trajectories.
Materials being prepared
Full trajectory files, baseline training/inference implementations, benchmark-pretraining checkpoints, and the complete scoring pipeline are being prepared for release. Additional materials will be added as they are assembled.
Circuit and trajectory materials remain subject to their applicable source terms. This initial snapshot does not redistribute circuit netlists, PDKs, model weights, or full trajectory archives.
Associated manuscript
AnalogDynBench: A Benchmark for Few-Shot Analog Transient Response Prediction and Behavioral Evaluation
- Downloads last month
- 19