OpenAI: /v1/models/{model} compatibility

[royjhan]

Adds compatibility for /v1/models/{model}

E.g
curl http://localhost:11434/v1/models/llama3

{
    "id": "llama3",
    "object": "model",
    "created": 1718141294,
    "owned_by": "library"
}

[JerrettDavis]

Looks good! Way better approach than what I was attempting in my branch.

[jmorganca]

We most likely don't need toDeleteCompletion and can just provide the struct inline here

[jmorganca]

This works, however what would be even better is some tests in openai.go that test the middleware in isolation. Here's an example I had an LLM generate me for a middleware that changes formats (not the same although I think you get the idea)

package main

import (
    "net/http"
    "net/http/httptest"
    "testing"

    "github.com/gin-gonic/gin"
)

func TestFormatMiddleware(t *testing.T) {
    // Create a test router
    r := gin.New()
    r.Use(FormatMiddleware())
    r.GET("/test", func(c *gin.Context) {
        format := c.GetString("format")
        if format == "json" {
            c.JSON(http.StatusOK, gin.H{"message": "Success"})
        } else if format == "xml" {
            c.XML(http.StatusOK, gin.H{"message": "Success"})
        } else {
            c.String(http.StatusOK, "Success")
        }
    })

    // Define test cases
    tests := []struct {
        name           string
        queryParam     string
        expectedStatus int
        expectedBody   string
    }{
        {"JSON Format", "json", http.StatusOK, `{"message":"Success"}`},
        {"XML Format", "xml", http.StatusOK, `<map><message>Success</message></map>`},
        {"Default Format", "", http.StatusOK, "Success"},
    }

    for _, tt := range tests {
        t.Run(tt.name, func(t *testing.T) {
            req, _ := http.NewRequest("GET", "/test?format="+tt.queryParam, nil)
            w := httptest.NewRecorder()
            r.ServeHTTP(w, req)

            if w.Code != tt.expectedStatus {
                t.Errorf("Expected status %d, got %d", tt.expectedStatus, w.Code)
            }

            if w.Body.String() != tt.expectedBody {
                t.Errorf("Expected body %s, got %s", tt.expectedBody, w.Body.String())
            }
        })
    }
}

[jmorganca]

should we change this to toModel?

Suggested change

	func toRetrieveCompletion(r api.ShowResponse, model string) Model {
	func toModel(r api.ShowResponse, model string) Model {

[jmorganca]

This should be the Ollama username similar to /v1/models

[jmorganca]

Can we remove Gin from this test file (in the PR this branches off of)

[royjhan]

Gin is needed to test the middleware, do you mind explaining what you mean?

[jmorganca]

Let's not use assert either – just t.Fatal t.Fatalf etc in the standard testing library is best

[jmorganca]

retrieve model

[jmorganca]

Is this used? I believe it's implemented in middleware now?