Cerebras
View SourceUltra-fast inference with OpenAI-compatible API and Cerebras-specific optimizations.
Configuration
CEREBRAS_API_KEY=csk_...
Provider Options
No custom provider options - uses OpenAI-compatible defaults with Cerebras-specific handling.
Passed via standard ReqLLM options:
temperature,max_tokens,top_ptools(with automaticstrict: truefor non-Qwen models)tool_choice("auto"or"none"only)
Implementation Notes
System Messages
System messages have stronger influence compared to OpenAI's implementation.
Tool Calling
- Requires
strict: truein tool schemas (automatically added) - Qwen models do NOT support
strict: true(automatically excluded) - Only supports
tool_choice: "auto"or"none"(not function-specific)
Streaming Limitations
Streaming not supported with:
- Reasoning models in JSON mode
- Tool calling scenarios
Unsupported OpenAI Features
These fields will result in a 400 error:
frequency_penaltylogit_biaspresence_penaltyparallel_tool_callsservice_tier
All restrictions handled automatically by ReqLLM.