k1rl-quasar / README.md
RealComp's picture
Upload 32 files
047d482 verified
|
Raw
History Blame Contribute Delete
5.72 kB
metadata
title: K1RL QUASAR  Volatility 75
emoji: 
colorFrom: yellow
colorTo: gray
sdk: docker
pinned: false
license: mit
short_description: 8 quantum AI agents trading Volatility 75 on Deriv 24/7

⚡ K1RL QUASAR — VOLATILITY 75

⭐ THE GOLD RANGER — Precision · Momentum · Continuous Volatility

Status Index Symbol Voyage Pretrain


🧠 What is K1RL QUASAR?

K1RL QUASAR is an 8-agent quantum-enhanced reinforcement learning trading system that autonomously trades synthetic indices on the Deriv platform — 24 hours a day, 7 days a week, with zero human intervention.

Each agent runs an Actor-Critic network enhanced with Qiskit quantum circuits, coordinating via QMIX value decomposition and a QuantumVotingSystem where all 6 trading agents vote on every action before execution.


⭐ Gold Ranger Profile — Volatility 75

The Gold Ranger is the precision momentum trader. Pretrained on V75_1S weights — the same symbol, the same market structure, the same 75% annualised volatility — but now operating on the standard 1-second tick feed. Agents arrive already calibrated to V75's random-walk character and simply refine their timing to the native cadence.

Property Value
🎯 Index Volatility 75 Index
📡 Deriv Symbol R_75
〰️ Market Type Continuous random-walk (75% annualised vol)
📈 Tick Cadence 1 second
🧬 Pretrained From V75_1S (voyage v5)
📦 Voyage v10 → saves to V100/v10/
🔑 Redis Password k1rl_v75_8d2f5a1c9e4b7063
🔄 Transfer Scale 1.00× — direct weight transfer (same symbol family)

🔑 Why V75_1S is the Right Pretrain

Unlike spaces that load from V75 foundation (v1), V75 loads pretrained weights from V75_1S (v5) — its closest sibling.

Factor V75_1S → V75 V75 Foundation (v1) → V75
Symbol ✅ Identical (R_75 / 1HZ75V) ✅ Same symbol
Market structure ✅ Same random-walk character ✅ Same
Training maturity ✅ Fully trained (v5 voyage) ⚠️ Older baseline
Weight transfer scale ✅ 1.00× (no freq. correction) ⚠️ May require re-adaptation
Expected convergence Fast — same market, more evolved weights Moderate

The agents arrive already understanding V75's volatility rhythm. Fine-tuning on the standard tick feed is the only remaining step.


🏗️ Architecture

┌──────────────────────────────────────────────────────────┐
│               K1RL QUASAR — VOLATILITY 75                │
├──────────────┬──────────────┬────────────────────────────┤
│  Features.py │  Rewards.py  │    quasar_main4.py         │
│  V75         │  V75         │    8 Actor-Critic          │
│  Tick Data   │  Reward Sig  │    QMIX + Voting           │
├──────────────┴──────────────┴────────────────────────────┤
│                 Redis (V75: namespace)                    │
├──────────────────────────────────────────────────────────┤
│            HuggingFace Spaces — CPU Basic                │
└──────────────────────────────────────────────────────────┘

Checkpoint Loading:

  Startup → scan V100/v10/ → found? → YES → Resume training
                                   → NO  → Load V100/v5/ (V75_1S)
                                                → Reset steps to 0
                                                → Begin V75 fine-tune

🗺️ Fleet Voyage Map

Voyage Space Symbol Pretrain Theme
v1 K1RL-Quasar-Volatility-75 R_75 🏛️ Foundation ⬛ Black
v2 K1RL-Quasar-Crash-500 CRASH500 v1 🔵 Blue
v3 K1RL-Quasar-Volatility-50-1s 1HZ50V v1 💚 Green
v4 K1RL-Quasar-Volatility-100-1s 1HZ100V v1 🔴 Red
v5 K1RL-Quasar-Volatility-75-1s 1HZ75V v1 💛 Yellow
v6 K1RL-Quasar-Crash-1000 CRASH1000 v2 🌸 Pink
v7 K1RL-Quasar-Volatility-25 R_25 v1
v8 K1RL-Quasar-Volatility-30-1s 1HZ30V v1
v9 K1RL-Quasar-STPRNG200 STPRNG200 v1
v10 K1RL-Quasar-Volatility-75 R_75 v5 (V75_1S) ⭐ Gold

📊 Checkpoints

  • Dataset: KarlQuant/k1rl-checkpoints
  • Path: V100/v10/
  • Strategy: Resume from v10 → fallback warm-start from v5 (V75_1S) → cold start if v5 empty

Built with PyTorch · Qiskit · Redis · HuggingFace Spaces

K1RL QUASAR FleetAutonomous Quantum Trading