Skip to content
This repository was archived by the owner on Jul 4, 2025. It is now read-only.

Commit 90694c4

Browse files
sangjanaisangjanai
committed
chore: API docs
1 parent bcc4c80 commit 90694c4

File tree

1 file changed

+229
-3
lines changed

1 file changed

+229
-3
lines changed

‎docs/static/openapi/cortex.json‎

Lines changed: 229 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -512,6 +512,73 @@
512512
}
513513
}
514514
},
515+
"/v1/models/add": {
516+
"post": {
517+
"operationId": "ModelsController_addModel",
518+
"summary": "Add a model",
519+
"description": "Add a new model configuration to the system.",
520+
"requestBody": {
521+
"required": true,
522+
"content": {
523+
"application/json": {
524+
"schema": {
525+
"$ref": "#/components/schemas/AddModelRequest"
526+
}
527+
}
528+
}
529+
},
530+
"responses": {
531+
"200": {
532+
"description": "Successful response",
533+
"content": {
534+
"application/json": {
535+
"schema": {
536+
"type": "object",
537+
"properties": {
538+
"message": {
539+
"type": "string"
540+
},
541+
"model": {
542+
"type": "object",
543+
"properties": {
544+
"model": {
545+
"type": "string"
546+
},
547+
"engine": {
548+
"type": "string"
549+
},
550+
"version": {
551+
"type": "string"
552+
}
553+
}
554+
}
555+
}
556+
},
557+
"example": {
558+
"message": "Model added successfully!",
559+
"model": {
560+
"model": "claude-3-5-sonnet-20241022",
561+
"engine": "anthropic",
562+
"version": "2023-06-01"
563+
}
564+
}
565+
}
566+
}
567+
},
568+
"400": {
569+
"description": "Bad request",
570+
"content": {
571+
"application/json": {
572+
"schema": {
573+
"$ref": "#/components/schemas/SimpleErrorResponse"
574+
}
575+
}
576+
}
577+
}
578+
},
579+
"tags": ["Pulling Models"]
580+
}
581+
},
515582
"/v1/models": {
516583
"get": {
517584
"operationId": "ModelsController_findAll",
@@ -1417,7 +1484,7 @@
14171484
"required": true,
14181485
"schema": {
14191486
"type": "string",
1420-
"enum": ["llama-cpp", "onnxruntime", "tensorrt-llm"],
1487+
"enum": ["llama-cpp", "onnxruntime", "tensorrt-llm", "openai", "anthropic"],
14211488
"default": "llama-cpp"
14221489
},
14231490
"description": "The type of engine"
@@ -1439,6 +1506,31 @@
14391506
"type": "string",
14401507
"description": "The variant of the engine to install (optional)",
14411508
"example": "mac-arm64"
1509+
},
1510+
"type": {
1511+
"type": "string",
1512+
"description": "The type of connection",
1513+
"example": "remote"
1514+
},
1515+
"url": {
1516+
"type": "string",
1517+
"description": "The URL for the API endpoint",
1518+
"example": "https://api.openai.com"
1519+
},
1520+
"api_key": {
1521+
"type": "string",
1522+
"description": "The API key for authentication",
1523+
"example": ""
1524+
},
1525+
"metadata": {
1526+
"type": "object",
1527+
"properties": {
1528+
"get_models_url": {
1529+
"type": "string",
1530+
"description": "The URL to get models",
1531+
"example": "https://api.openai.com/v1/models"
1532+
}
1533+
}
14421534
}
14431535
}
14441536
}
@@ -1475,7 +1567,7 @@
14751567
"required": true,
14761568
"schema": {
14771569
"type": "string",
1478-
"enum": ["llama-cpp", "onnxruntime", "tensorrt-llm"],
1570+
"enum": ["llama-cpp", "onnxruntime", "tensorrt-llm", "openai", "anthropic"],
14791571
"default": "llama-cpp"
14801572
},
14811573
"description": "The type of engine"
@@ -1690,7 +1782,7 @@
16901782
"required": true,
16911783
"schema": {
16921784
"type": "string",
1693-
"enum": ["llama-cpp", "onnxruntime", "tensorrt-llm"],
1785+
"enum": ["llama-cpp", "onnxruntime", "tensorrt-llm", "openai", "anthropic"],
16941786
"default": "llama-cpp"
16951787
},
16961788
"description": "The name of the engine to update"
@@ -3623,6 +3715,109 @@
36233715
}
36243716
}
36253717
},
3718+
"AddModelRequest": {
3719+
"type": "object",
3720+
"required": ["model", "engine", "version", "inference_params", "TransformReq", "TransformResp", "metadata"],
3721+
"properties": {
3722+
"model": {
3723+
"type": "string",
3724+
"description": "The identifier of the model."
3725+
},
3726+
"api_key_template": {
3727+
"type": "string",
3728+
"description": "Template for the API key header."
3729+
},
3730+
"engine": {
3731+
"type": "string",
3732+
"description": "The engine used for the model."
3733+
},
3734+
"version": {
3735+
"type": "string",
3736+
"description": "The version of the model."
3737+
},
3738+
"inference_params": {
3739+
"type": "object",
3740+
"properties": {
3741+
"temperature": {
3742+
"type": "number"
3743+
},
3744+
"top_p": {
3745+
"type": "number"
3746+
},
3747+
"frequency_penalty": {
3748+
"type": "number"
3749+
},
3750+
"presence_penalty": {
3751+
"type": "number"
3752+
},
3753+
"max_tokens": {
3754+
"type": "integer"
3755+
},
3756+
"stream": {
3757+
"type": "boolean"
3758+
}
3759+
}
3760+
},
3761+
"TransformReq": {
3762+
"type": "object",
3763+
"properties": {
3764+
"get_models": {
3765+
"type": "object"
3766+
},
3767+
"chat_completions": {
3768+
"type": "object",
3769+
"properties": {
3770+
"url": {
3771+
"type": "string"
3772+
},
3773+
"template": {
3774+
"type": "string"
3775+
}
3776+
}
3777+
},
3778+
"embeddings": {
3779+
"type": "object"
3780+
}
3781+
}
3782+
},
3783+
"TransformResp": {
3784+
"type": "object",
3785+
"properties": {
3786+
"chat_completions": {
3787+
"type": "object",
3788+
"properties": {
3789+
"template": {
3790+
"type": "string"
3791+
}
3792+
}
3793+
},
3794+
"embeddings": {
3795+
"type": "object"
3796+
}
3797+
}
3798+
},
3799+
"metadata": {
3800+
"type": "object",
3801+
"properties": {
3802+
"author": {
3803+
"type": "string"
3804+
},
3805+
"description": {
3806+
"type": "string"
3807+
},
3808+
"end_point": {
3809+
"type": "string"
3810+
},
3811+
"logo": {
3812+
"type": "string"
3813+
},
3814+
"api_key_url": {
3815+
"type": "string"
3816+
}
3817+
}
3818+
}
3819+
}
3820+
},
36263821
"CreateModelDto": {
36273822
"type": "object",
36283823
"properties": {
@@ -4292,6 +4487,37 @@
42924487
"type": "integer",
42934488
"description": "Number of GPU layers.",
42944489
"example": 33
4490+
},
4491+
"api_key_template": {
4492+
"type": "string",
4493+
"description": "Template for the API key header."
4494+
},
4495+
"version": {
4496+
"type": "string",
4497+
"description": "The version of the model."
4498+
},
4499+
"inference_params": {
4500+
"type": "object",
4501+
"properties": {
4502+
"temperature": {
4503+
"type": "number"
4504+
},
4505+
"top_p": {
4506+
"type": "number"
4507+
},
4508+
"frequency_penalty": {
4509+
"type": "number"
4510+
},
4511+
"presence_penalty": {
4512+
"type": "number"
4513+
},
4514+
"max_tokens": {
4515+
"type": "integer"
4516+
},
4517+
"stream": {
4518+
"type": "boolean"
4519+
}
4520+
}
42954521
}
42964522
}
42974523
},

0 commit comments

Comments
 (0)