AI Training Datasets

This dataset repository contains structured session telemetry, real-time trade signals, risk gate evaluations, agent decisions, and accuracy improvement reports generated by the Nifty AI Brain Agent running autonomously on Kaggle dual T4 GPU environment. 1. Continuous Learning: Track which setups (CPR Breakout, VWAP Rejection, Trap Hunter) perform best. 2. Defect & Weakness Isolation: Automatically identify why signals fail (e.g. low VIX chop, opening 15-min noise, wrong delta). 3. Agent Tuning: Use improvementnotes.md to refine strategy thresholds, risk weights, and LLM reasoning prompts. Automatically synced from Kaggle dual T4 runners via HFSessionPusher.

Environment adaptation, training evidence, latency profiles, raw demonstrations and evaluation traces, organized by environment and experiment stage. - AirRaid: zero-latency and profile-latency experiments. Only raw demonstrations are distributed. Generate converted training datasets on the training server. Model weights remain in the dedicated model repository and are referenced from the experiment records. Pin a commit revision for reproducible downloads. All seven environments use condition/model-family/experiment/stage directories, with each environment’s shared zero-latency data stored once. Existing dataset configuration names and split names are retained and point to the reorganized…

sveneziale/activationsandbarcodes3108 Compute artifacts pushed by tda-for-llms's Hugging-Face-backed pipeline (hf.enabled: true in experiment.yaml). Layout activations/{modelslug}/{corpus}/{revision}/{actname}/ Raw per-cloud activation matrices extracted from the model, one independent copy per checkpoint (revision). Independent of topology.metric — the same activations are reused across every metric or topology config

Dataset · Text generation

thai-semi-pretrain-qwen

AI

ชุดข้อมูลภาษาไทยสำหรับ semi-pretraining: เอกสารเว็บภาษาไทยที่ถูกทำความสะอาดและเรียบเรียงใหม่ด้วย LLM โดยมีบล็อกการวิเคราะห์สั้น ๆ … ฝังอยู่หน้าข้อความ และมีคู่คำถาม-คำตอบ (SFT) พร้อม think full.jsonl (หนึ่งแถวต่อเอกสาร) pretrainwiththink.jsonl / pretrainclean.jsonl: id, type (rewrite), text sft.jsonl: id ( -q ), docid, question, think, answer, messages (รูปแบบแชท: user = คำถาม, assistant = … + คำตอบ) 1. แหล่งข้อมูลดิบ: สตรีมตัวอย่างจาก FineWeb-2 ภาษาไทย (HuggingFaceFW/fineweb-2, thaThai) และ Wikipedia ไทย (wikimedia/wikipedia, 20231101.th) รวม 80,000 เอกสาร (CulturaX ไม่ได้ใช้ เพราะยังไม่ได้รับสิทธิ์เข้าถึง) 2. กรองด้วยกฎ (ไม่ใช้ LLM): normalize Unicode (NFC), ลบอักขระล่องหน/ควบคุม, ลบ URL…

Dataset

global_sat_metar_v3

Meteolibre.dev

Corrected republish of meteolibre-dev/globalsatmetarv2. Same shards, schema, patch grid and satellite/radar/elevation bytes; only the rasterized METAR/SYNOP channel was fixed. The v2 rasterizer used a closed 1-hour window [t-1h, t], so every observation stamped at exactly HH:00:00 (all SYNOP/FM-12 reports, and any station with reporting gaps > 1h) was written into two consecutive hourly frames. ~20-40% of consecutive obs pairs in v2 were exact copies of the same report, which trains a copy-prior ("next frame = last frame") into sequence models. In v3, a frame t+1 whose tmpc, dwpc and mslp are all bit-exact equal to frame t (at least one finite) is treated as the same report leaked across…

taskcategories: - text-generation - agent

Sweeperbench run database runs.csv is the append-only raw table. Each row describes one prediction, core evaluation, or preservation evaluation. schema.json documents the fields. Primary key: runid (UUID). Each top-level UUID directory is exactly that run, with matching metadata.json, artifact manifest, prompts, traces, verdicts and any collected recordings. Relationships: experimentid groups a launch; evaluations reference the exact predictionrunid. Model, harness

Dataset

fec-dumps

Nick Crews

parquet versions of the Schedule A and Schedule B tables from the weekly postgres.dump backups from the Federal Election Commission's database. Published on a weekly cron job by https://github.com/NickCrews/fec-dumps, see that for more info.

Public browser assets for the C4D scene viewer, including the wider-family gallery. The full training exports are in smerkd/c4d-scenes. This repository contains browser previews and display-mapped supervision examples, not full training supervision. Browser videos are compressed previews; training RGB is stored losslessly in the scene archives. Gaussian displays use a bounded subset with experimental, uncertified covariance/opacity conventions. Preview assets are checksum-verified and gallery URLs refer to immutable revisions.

Synthetic Nepali instruction-tuning data generated with NVIDIA NeMo Data Designer from authoritative Nepali documents (agriculture manuals, legal texts). Answers are grounded strictly in the source; unanswerable questions get an explicit refusal. Records use chat messages format plus metadata and per-record qualityscores (grounding / correctness / naturalness, 1-5, LLM-as-judge). One data/train-.jsonl per source document; shards are overwritten idempotently on re-runs. (grounding>=4, correctness>=3, naturalness>=3). Companion repos hold quality-gate rejects (rejected-) and records whose judge call failed (unjudged-).