Skip to content

MODEL_CONFIG.md

Latest commit

 

History

History
199 lines (157 loc) · 5.75 KB

MODEL_CONFIG.md

File metadata and controls

199 lines (157 loc) · 5.75 KB

Model Configuration Guide

The autonomous trading agent is model agnostic and supports OpenAI, Groq, and Cerebras models via LiteLLM.

Supported Models

Cerebras (RECOMMENDED FOR TRADING - Ultra-Fast)

  • Model: gpt-oss-120b (120B parameter model)
  • Speed: ⚡ ULTRA-FAST - 1800+ tokens/second
  • Pros: Blazing fast inference (10x faster than OpenAI), excellent for real-time trading, 120B parameters for strong reasoning
  • Cons: Newer provider, limited track record
  • Best for: Real-time trading where speed is critical
  • API: https://api.cerebras.ai/v1

Groq (Fast & Cost-Effective)

  • Model: qwen/qwen3-32b (Qwen 3, 32B parameters)
  • Speed: ⚡ Fast - 200-500 tokens/second
  • Pros: Very fast inference, cost-effective, strong multilingual support
  • Cons: Smaller model size compared to others
  • Best for: Development, testing, high-frequency decisions

OpenAI (Default - High Quality)

  • Model: gpt-5-mini
  • Speed: 🐌 Slower - 50-100 tokens/second
  • Pros: High quality, reliable, best reasoning
  • Cons: Slower, higher cost per token
  • Best for: Production trading with accurate decision-making where speed is less critical

Configuration

.env File Settings

Add these variables to your .env file:

# Model Provider: "cerebras" (recommended), "groq", or "openai" (default)
MODEL_PROVIDER=cerebras

# Cerebras API Key (required if using Cerebras)
CEREBRAS_API_KEY=csk-your-cerebras-api-key-here

# Groq API Key (required if using Groq)
GROQ_API_KEY=gsk-your-groq-api-key-here

# OpenAI API Key (required if using OpenAI)
OPENAI_API_KEY=sk-your-openai-api-key-here

# OpenAlgo Settings (always required)
OPENALGO_API_KEY=your-openalgo-key
OPENALGO_HOST=http://127.0.0.1:5000

Switching Models

Use Cerebras (Recommended for Speed)

  1. Set MODEL_PROVIDER=cerebras in .env
  2. Ensure CEREBRAS_API_KEY is configured
  3. Run the agent: 
    uv run python agent.py
  4. You'll see: [MODEL] Using Cerebras (cerebras/gpt-oss-120b) - ULTRA-FAST

Use Groq

  1. Set MODEL_PROVIDER=groq in .env
  2. Ensure GROQ_API_KEY is configured
  3. Run the agent: 
    uv run python agent.py
  4. You'll see: [MODEL] Using Groq (groq/qwen/qwen3-32b)

Use OpenAI (Default)

  1. Set MODEL_PROVIDER=openai in .env (or leave it unset)
  2. Ensure OPENAI_API_KEY is configured
  3. Run the agent: 
    uv run python agent.py
  4. You'll see: [MODEL] Using OpenAI (openai/gpt-5-mini)

Advanced: Custom Model Selection

You can override the default model for each provider:

# Use a different Cerebras model
MODEL_PROVIDER=cerebras
CEREBRAS_MODEL=cerebras/gpt-oss-120b  # Default, or try cerebras/llama3.1-70b
CEREBRAS_API_KEY=csk-...

# Use a different Groq model
MODEL_PROVIDER=groq
GROQ_MODEL=groq/qwen/qwen3-32b  # Default, or try groq/llama-3.3-70b-versatile
GROQ_API_KEY=gsk-...

# Use a different OpenAI model
MODEL_PROVIDER=openai
OPENAI_MODEL=openai/gpt-4o
OPENAI_API_KEY=sk-...

# Use any custom LiteLLM-compatible model
MODEL_PROVIDER=custom
CUSTOM_MODEL=anthropic/claude-3-5-sonnet-20241022
CUSTOM_API_KEY=sk-ant-...
CUSTOM_API_BASE=https://api.anthropic.com/v1  # Optional

Model Selection Logic

The agent automatically selects the model based on MODEL_PROVIDER:

# From agent.py:36-56
MODEL_PROVIDER = os.getenv("MODEL_PROVIDER", "openai").lower()

if MODEL_PROVIDER == "groq":
    trading_model = LitellmModel(
        model="groq/llama-3.3-70b-versatile",
        api_key=os.getenv("GROQ_API_KEY")
    )
else:
    trading_model = LitellmModel(
        model="openai/gpt-5-mini",
        api_key=os.getenv("OPENAI_API_KEY")
    )

Adding Custom Models

To add support for other LiteLLM-compatible models:

  1. Edit agent.py lines 36-56
  2. Add a new condition for your provider: 
    elif MODEL_PROVIDER == "anthropic":
    trading_model = LitellmModel(
        model="anthropic/claude-3-5-sonnet-20241022",
        api_key=os.getenv("ANTHROPIC_API_KEY")
    )
  3. Update this documentation

Performance Comparison

Metric Cerebras gpt-oss-120b Groq qwen3-32b OpenAI gpt-5-mini
Model Size 120B parameters 32B parameters Unknown (mini)
Speed ⚡⚡⚡ ~50ms/request ⚡⚡ ~100ms/request 🐌 ~500ms/request
Tokens/sec 1800+ 200-500 50-100
Cost $0.60/1M in + $0.60/1M out $0.05/1M tokens $0.15/1M in + $0.60/1M out
Quality Excellent (120B) Very Good (32B) Excellent
Reasoning Best-in-class Strong Best-in-class
Availability 99% uptime 99% uptime 99.9% uptime
Best For Real-time trading Fast decisions Accuracy critical

Troubleshooting

Import Error

ModuleNotFoundError: No module named 'agents.extensions.models.litellm_model'

Fix: Install with LiteLLM support:

uv add openai-agents[litellm]
uv sync

API Key Error

litellm.AuthenticationError: API key not found

Fix: Ensure the correct API key is set in .env:

  • For OpenAI: OPENAI_API_KEY=sk-...
  • For Groq: GROQ_API_KEY=gsk-...

Model Not Found

litellm.exceptions.BadRequestError: model 'gpt-5-mini' not found

Fix: Check that the model name is correct and available. If the model name changes, update agent.py line 53.

Cost Estimation

Daily Trading Costs (Estimated)

Assumptions:

  • 375 trading cycles per day (9:15 AM - 3:15 PM, every minute)
  • 5 symbols per cycle
  • ~2000 tokens per cycle (analysis + decisions)
  • Total: 750,000 tokens/day

OpenAI (gpt-5-mini):

  • Cost: $0.11/day ($3.30/month)

Groq (llama-3.3-70b):

  • Cost: $0.04/day ($1.20/month)

Note: Actual costs may vary based on market conditions and decision complexity.