Meta (Llama)
View SourceGeneric provider for Meta Llama models using Meta's native prompt format.
Important Usage Note
Most deployments use OpenAI-compatible APIs and should NOT use this provider directly:
- Azure AI Foundry → Use OpenAI-compatible API
- Google Cloud Vertex AI → Use OpenAI-compatible API
- vLLM (self-hosted) → Use OpenAI-compatible API
- Ollama (self-hosted) → Use OpenAI-compatible API
- llama.cpp (self-hosted) → Use OpenAI-compatible API
This provider is for services using Meta's native format with prompt, max_gen_len, generation fields.
Current Use Cases
- AWS Bedrock: Uses native Meta format via
ReqLLM.Providers.AmazonBedrock.Meta
For AWS Bedrock, see Amazon Bedrock Provider Guide.
Configuration
No direct configuration - wrapped by cloud providers using native format.
Native Format Details
Request Format
prompt- Formatted text with Llama special tokensmax_gen_len- Maximum tokens to generatetemperature- Sampling temperaturetop_p- Nucleus sampling parameter
Response Format
generation- Generated textprompt_token_count- Input token countgeneration_token_count- Output token countstop_reason- Why generation stopped
Llama Prompt Format
Llama 3+ uses structured prompt format with special tokens:
- System:
<|start_header_id|>system<|end_header_id|> - User:
<|start_header_id|>user<|end_header_id|> - Assistant:
<|start_header_id|>assistant<|end_header_id|>
Provider Options
No custom provider options - uses standard ReqLLM options translated to native format.