squash on a reasonable progress checkpoint

openai-openapi/config.yaml

@@ -0,0 +1,10 @@
+package: apiv2
+generate:
+  models: true
+  chi-server: true
+output: apiv2/localai.gen.go
+# output-options:
+#   user-templates:
+#     union.tmpl: "// SKIP"
+#     union-and-additional-properties.tmpl: "// SKIP"
+#     additional-properties.tmpl: "// SKIP"

openai-openapi/localai_model_patches.yaml

@@ -0,0 +1,64 @@
+#! LocalAI Model Patches
+#! These are fed to YTT, which patches the OpenAI OpenAPI spec to include the custom attributes used by LocalAI to pass parameters and information to local models.
+#! TODO: pretty this up. Fancy YAML features will reduce code duplication, but this is a POC.
+
+#@ load("@ytt:overlay", "overlay")
+
+#@overlay/match by=overlay.all
+---
+components:
+  schemas:
+    CreateChatCompletionRequest:
+      properties:
+        #@overlay/match missing_ok=True
+        x-localai-extensions:
+          type: object
+          nullable: true
+          properties:
+            seed:
+              type: integer
+              nullable: true
+            top_k:
+              type: integer
+              nullable: true
+            batch:
+              type: integer
+              nullable: true
+            f16:
+              type: boolean
+              nullable: true
+            ignore_eos:
+              type: boolean
+              nullable: true
+            #! skip RepeatPenalty, try using FrequencyPenalty in real model instead??
+            n_keep:
+              type: integer
+              nullable: true
+              x-go-name: Keep
+            mirostat:
+              type: integer
+              nullable: true
+            mirostat_eta:
+              type: number
+              nullable: true
+              x-go-type: float64
+            mirostat_tau:
+              type: number
+              nullable: true
+              x-go-type: float64
+    CreateImageRequest:
+      properties:
+        #@overlay/match missing_ok=True
+        x-localai-extensions:
+          type: object
+          nullable: true
+          properties:
+            seed:
+              type: integer
+              nullable: true
+            mode:
+              type: integer
+              nullable: true
+            step:
+              type: integer
+              nullable: true

openai-openapi/remove_depreciated_openapi.yaml

@@ -0,0 +1,12 @@
+#! This doesn't actually work or get used yet. It's a scratch space for an idea of mine - we might want to just strip out and ignore depreciated portions of the spec. Not reliable enough to use yet.
+
+#@ load("@ytt:overlay", "overlay")
+
+#@overlay/match by=overlay.all, expects="1+"
+---
+paths:
+  #!overlay/match by=overlay.subset({"deprecated": True})
+  #@overlay/remove
+  *:
+    *:
+      depreciated: True

openai-openapi/test_segment.yml

@@ -0,0 +1,30 @@
+#! This file is just for my reference during development and will be removed.
+components:
+  schemas:
+    CreateChatCompletionRequest:
+      type: object
+      properties:
+        model:
+          description: ID of the model to use. Currently, only `gpt-3.5-turbo` and `gpt-3.5-turbo-0301` are supported.
+          type: string
+        messages:
+          description: The messages to generate chat completions for, in the [chat format](/docs/guides/chat/introduction).
+          type: array
+          minItems: 1
+          items:
+            $ref: '#/components/schemas/ChatCompletionRequestMessage'
+        temperature:
+          type: number
+          minimum: 0
+          maximum: 2
+          default: 1
+          example: 1
+          nullable: true
+          description: *completions_temperature_description
+    CreateImageRequest:
+      type: object
+      properties:
+        prompt:
+          description: A text description of the desired image(s). The maximum length is 1000 characters.
+          type: string
+          example: "A cute baby sea otter"