Fireworks

WatchThatSite integrates with Fireworks' API, providing high-performance inference on open-source language models.

Get Your API Key

1. Go to Fireworks Console
2. Sign up or log in with your account
3. Navigate to API Keys
4. Click Create API Key
5. Name it (e.g., "WatchThatSite")
6. Click Generate
7. Copy the API key immediately

Add to WatchThatSite

1. Go to Settings > AI Providers
2. Click Add Provider Key
3. Select Fireworks from the provider dropdown
4. Paste your API key
5. Optionally select specific models
6. Click Save

Recommended Models

Latest Fireworks models with tool calling:

• Llama 3.1 70B — Large, capable open model
• Mixtral 8x7B — Fast, capable MoE model
• Qwen 72B — Alibaba's powerful model
• Mistral 7B Instruct — Fast, efficient

Plus many additional open-source models.

All Available Models

Your account has access to all Fireworks models supporting tool calling:

1. Go to Settings > AI Providers
2. Click Add Provider Key or Edit existing
3. Models dropdown shows all available models
4. Select specific models to restrict, or leave empty for all

Browse all models on Fireworks catalog.

Key Features

• Ultra-fast inference: Optimized serving infrastructure
• Open-source focus: Emphasis on transparent, auditable models
• Cost-effective: Competitive pricing on open models
• Function calling: Tool use support
• Streaming: Real-time token generation
• Fine-tuning: Ability to fine-tune models

Pricing

Fireworks offers competitive pricing:

• Per 1M tokens: Standard pricing model
• Input vs output: Different rates for input/output tokens
• Varies by model: Different models have different costs
• Free trial: May be available for new accounts

Check Fireworks pricing for current rates.

Use Cases

Fireworks excels for:

• Cost-sensitive projects: Open models at low cost
• Privacy concerns: Run on your infrastructure option
• Fine-tuning: Customize models for your domain
• Control and auditability: Open-source transparency
• High throughput: Optimized for scale

Advantages

• No vendor lock-in: Open-source models available elsewhere
• Transparent: Audit models and behavior
• Community models: Access to cutting-edge research
• Customizable: Fine-tune for specific domains
• Self-hosted option: Run models yourself if needed

Account Setup

Set Up Billing

1. Go to Account Settings
2. Add payment method
3. Set usage budget if desired

Monitor Usage

1. Go to Usage Dashboard
2. View token consumption
3. Monitor costs
4. Check rate limits

Model Selection Tips

• Speed vs capability: Smaller models are faster, larger are more capable
• Cost: Smaller models are cheaper
• Common choice: Mixtral 8x7B offers good balance
• Latest: Check for newly released models
• Benchmarks: Fireworks provides performance comparisons

Troubleshooting

"Invalid API Key"

• Verify complete key is pasted without spaces
• Ensure key is active in Fireworks account
• Check key hasn't been revoked
• Regenerate if needed

"Model Not Found"

• Verify model ID is exactly correct
• Check Fireworks model list for availability
• Model names are case-sensitive
• Some models may be beta or region-restricted

Rate Limits

For free tier:

• Limited concurrent requests
• Request quota per time period
• Upgrade for higher limits

For paid accounts:

• Contact Fireworks support for quota increase
• Implement exponential backoff
• Use request batching

High Latency

To optimize latency:

• Use smaller models
• Enable streaming for incremental results
• Batch similar requests
• Check network conditions

Billing Issues

• Verify payment method is current
• Check account hasn't been suspended
• Confirm credit limit not exceeded
• Contact Fireworks support for issues

Performance Tips

• Streaming: Enable for real-time token generation
• Batch processing: Group requests for efficiency
• Temperature: Lower (0.2-0.5) for consistency
• Token limits: Set reasonable max_tokens values
• Connection pooling: Reuse connections

Advanced Features

Fine-Tuning

Fine-tune open-source models for your domain:

1. Prepare training data
2. Use Fireworks fine-tuning API
3. Deploy custom model
4. Use in WatchThatSite

Check Fireworks fine-tuning docs for details.

Embeddings

Some models support embeddings for semantic search:

• Text embedding models available
• Use for similarity search
• Lower cost than generation models