diff --git a/src/resources/model-metadata.json b/src/resources/model-metadata.json index 1fd4db9..aed9f56 100644 --- a/src/resources/model-metadata.json +++ b/src/resources/model-metadata.json @@ -701,5 +701,410 @@ "together_ai/Qwen/Qwen3-235B-A22B-fp8-tput": { "input_cost_per_token": 0.0000002, "output_cost_per_token": 0.0000006 - } + }, + "meta-llama/llama-3.1-8b-instruct": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.0000002, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.0000004, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 6000, + "tpm": 600000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama3.1" +}, +"meta-llama/llama-3.1-70b-instruct": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.0000015, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.0000020, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 2000, + "tpm": 200000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama3.1" +}, +"meta-llama/llama-3.1-405b-instruct": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.000008, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.000016, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 300, + "tpm": 50000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama3.1" +}, +"meta-llama/llama-3-8b-instruct": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.0000001, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.0000003, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 5000, + "tpm": 500000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama3" +}, +"meta-llama/llama-3-70b-instruct": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.0000008, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.0000015, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 1500, + "tpm": 150000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama3" +}, +"meta-llama/llama-2-7b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.0000001, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.0000003, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 6000, + "tpm": 600000, + "supports_system_messages": true, + "supports_function_calling": false, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": false, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama" +}, +"meta-llama/llama-2-13b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.00000015, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.00000035, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 5000, + "tpm": 500000, + "supports_system_messages": true, + "supports_function_calling": false, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": false, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama" +}, +"meta-llama/llama-2-70b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.0000008, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.0000015, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 1500, + "tpm": 150000, + "supports_system_messages": true, + "supports_function_calling": false, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": false, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama" +}, +"meta-llama/llama-3-8b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.0000002, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.0000004, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 6000, + "tpm": 600000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama3" +}, +"meta-llama/llama-3-70b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.000001, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.000002, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 2000, + "tpm": 200000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama3" +}, +"meta-llama/llama-3.1-8b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.0000002, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.0000004, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 6000, + "tpm": 600000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama3.1" +}, +"meta-llama/llama-3.1-70b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.0000015, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.0000020, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 2000, + "tpm": 200000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama3.1" +}, +"meta-llama/llama-3.1-405b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.000008, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.000016, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 300, + "tpm": 50000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama3.1" +}, +"meta-llama/llama-3.2-1b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.00000005, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.0000001, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 12000, + "tpm": 1200000, + "supports_system_messages": true, + "supports_function_calling": false, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": false, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama-models/tree/main/models/llama3_2" +}, +"meta-llama/llama-3.2-3b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "max_images_per_prompt": 0, + "max_videos_per_prompt": 0, + "max_video_length": 0, + "max_audio_length_hours": 0, + "max_audio_per_prompt": 0, + "max_pdf_size_mb": 0, + "input_cost_per_token": 0.00000012, + "input_cost_per_audio_token": 0, + "output_cost_per_token": 0.00000025, + "litellm_provider": "meta", + "mode": "chat", + "rpm": 8000, + "tpm": 800000, + "supports_system_messages": true, + "supports_function_calling": false, + "supports_vision": false, + "supports_response_schema": false, + "supports_audio_output": false, + "supports_tool_choice": false, + "supported_modalities": ["text"], + "supported_output_modalities": ["text"], + "source": "https://github.com/meta-llama/llama-models/tree/main/models/llama3_2" +} }