🐦⬛ RavenX-Trade 8B v1.1 — Autonomous Trading Intelligence
MLX 4-bit · Apple Silicon · 128K context · 4-step MAP protocol · 318K training examples
Qwen3-8B fine-tuned for autonomous trading analysis. Trained on 318,675 examples from Polymarket, crypto, stocks, options, hedge fund, and quantitative trading data. Follow the MAP protocol: Market → Analyze → Predict → Trade.
📦
GGUF (Ollama / llama.cpp): RavenX-Trade-8B-GGUF🛡️
Security model: RavenX-Sec-8B-RATH-128k (610K examples, 6-step RATH protocol)
Built by @DeadByDawn101 · RavenX LLC
What This Model Does
A fine-tuned Qwen3-8B that follows the MAP protocol for every trading analysis:
| MAP Step | What It Does |
|---|
| M — Market | Identifies asset, price, timeframe, market context |
| A — Analyze | Technical indicators, volume, momentum, on-chain signals |
| P — Predict | Signal direction, confidence level, target price |
| T — Trade | Entry, stop loss, target, risk/reward, position sizing |
Example Output
MAP STEP 1: MARKET
- Asset: Bitcoin (BTC)
- Price: 67500.00
- Timeframe: 15-minute
- Market Context: Early phase of bullish setup
MAP STEP 2: ANALYZE
- RSI at 28: Deep oversold, potential bounce
- MACD just bullish crossover: Confirmation of bullish momentum
- 3x average volume: High conviction, likely accumulation
MAP STEP 3: PREDICT
- Signal: Strong potential for bullish reversal
- Confidence: 85%
- Target: 68000.00
MAP STEP 4: TRADE
- Action: Buy
- Entry: 67500.00
- Stop Loss: 66500.00
- Target: 68000.00
- Risk/Reward: 1:1.5
- Position Size: 2.5% of portfolio
Quick Start (MLX)
1from mlx_lm import load, generate
2
3model, tokenizer = load("deadbydawn101/RavenX-Trade-8B-MAP-128k-mlx-4bit")
4
5messages = [
6 {"role": "system", "content": "You are RavenX-Trade. Follow the MAP protocol (Market, Analyze, Predict, Trade) for every analysis."},
7 {"role": "user", "content": "BTC at 67500. RSI 28. MACD bullish crossover. Volume 3x average. Analyze."}
8]
9
10prompt = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
11response = generate(model, tokenizer, prompt=prompt, max_tokens=2048)
12print(response)
Chat REPL
python3 -m mlx_lm chat --model deadbydawn101/RavenX-Trade-8B-MAP-128k-mlx-4bit --max-tokens 2048OpenAI-Compatible Server
python3 -m mlx_lm.server --model deadbydawn101/RavenX-Trade-8B-MAP-128k-mlx-4bit
Training Details
| Parameter | Value |
|---|
| Architecture | Qwen3-8B (Josiefied abliterated) |
| Training Data | 318,675 examples |
| Context Window | 128K (YaRN rope scaling, factor 4.0) |
| Method | MLX LoRA (rank 32, 8 layers) |
| Learning Rate | 1e-5 |
| Batch Size | 2 |
| Iterations | 2,000 |
| Sequence Length | 8,192 |
| Hardware | Apple M1 Max 64GB |
Dataset Composition (318K examples)
| Category | Examples | Weight |
|---|
| Polymarket | 86,800 | 2x |
| Stock/Options Trading | 36,000 | 2x |
| Hedge Fund Strategies | 14,200 | 2x |
| Crypto Trading | 7,100 | 2x |
| Agent/Tool Calling | 16,100 | 1x |
| GitHub Repos (15) | 3,422 | 3x |
| Synthetic MAP | custom | 2x |
Datasets Used (21)
Polymarket: polymarket-quant-bench (36.8K) · polymarket-transactions (50K streamed)
Crypto: Solana-blockchain-360-Coding · crypto-agent-safe-function-calling · crypto-trading-r1-rl · crypto-ai-trading-signals
Stocks/Quant: stock-trader-sft-v2 (36K) · QuantitativeEquityInvesting · stock-market-chart-patterns
Hedge Fund: Hedgefunddataset (14.2K)
Agent/Reasoning: Claude-TraceInversion · ToolACE
15 GitHub Repos: polyverse · polymarket-trade-engine · tradingagents · ai-hedge-fund · MiroFish · Polymarket-BTC-15-Minute-Trading-Bot · pump-fun-bundler-bot · pumpfun-mcp-server · auto-researchtrading · degen-financial-plugins · polyterm · gopher · trading-bot · open-reasoning · minimind
The RavenX Model Family
Source Code & Training Pipeline
License
Apache-2.0
"We don't give up. We do what others don't and build what isn't possible." — RavenX LLC
