Agent Q3 — HF Dataset Bucket (Unified)

Canonical HF dataset for all Agent Q3 variants Contains data for both Agent Q3 [Evo] Platform: Agent-Q3

Structure

agent-q3 evo/
│
├── Q3 Evo/                      ← Agent Q3 [Evo] data
│   ├── exports/                 # Domain JSONL exports (Solidity, prediction markets, LangGraph)
│   ├── feedback/                # RLHF / DPO feedback signals from Q3 Evo inference runs
│   ├── chroma/                  # ChromaDB vector store snapshots (384-dim nomic embeddings)
│   ├── benchmarks/              # Q3 Evo evaluation results and accuracy benchmarks
│   └── pipelines/               # Q3 Evo pipeline artifacts and run metadata
│   ├── corpora/                 # Domain-filtered training text
│   ├── lora_signals/            # LoRA feedback pairs (prompt → chosen / rejected)
│   ├── arxiv/                   # arXiv paper embeddings (384-dim)
│   ├── benchmarks/              # Q3 Evo benchmark results per training iteration
│   ├── lora_runs/               # LoRA experiment checkpoints and loss curves
│   ├── arxiv_ingestion/         # Processed arXiv chunks
│   └── run_logs/                # Training run metadata
│   ├── metadata/                # Dataset cards, version logs, provenance
    └── eval_sets/               # Cross-variant QA evaluation sets

Usage

from datasets import load_dataset

# Full unified dataset
ds = load_dataset("madDegen/agent-q3", split="train")

# Q3 Evo feedback signals only
ds_hq = load_dataset("madDegen/agent-q3", data_dir="hq/feedback", split="train")

# Q3 Evo LoRA signals only
ds_evo = load_dataset("madDegen/agent-q3", data_dir="evo/lora_signals", split="train")

Related Assets

License

Apache 2.0