mirror of
https://github.com/ollama/ollama.git
synced 2026-05-13 06:21:28 +00:00
app: source featured models from experimental recommendations endpoint (#15909)
Replace the hardcoded FEATURED_MODELS list with the /api/experimental/model-recommendations endpoint so the picker stays in sync with server-driven recommendations. Inline the merge into useModels (recommendations first, then the rest of /api/tags) and drop the standalone mergeModels util.
This commit is contained in:
Bruce MacDonald
GitHub
parent
8f39fff70b
commit
938ca6e274
7 changed files with 75 additions and 258 deletions
|
|
@ -406,6 +406,31 @@ export async function* pullModel(
|
|||
}
|
||||
}
|
||||
|
||||
export interface ModelRecommendation {
|
||||
model: string;
|
||||
description: string;
|
||||
context_length?: number;
|
||||
max_output_tokens?: number;
|
||||
vram_bytes?: number;
|
||||
}
|
||||
|
||||
export interface ModelRecommendationsResponse {
|
||||
recommendations: ModelRecommendation[];
|
||||
}
|
||||
|
||||
export async function getModelRecommendations(): Promise<ModelRecommendation[]> {
|
||||
const response = await fetch(
|
||||
`${API_BASE}/api/experimental/model-recommendations`,
|
||||
);
|
||||
if (!response.ok) {
|
||||
throw new Error(
|
||||
`Failed to fetch model recommendations: ${response.statusText}`,
|
||||
);
|
||||
}
|
||||
const data: ModelRecommendationsResponse = await response.json();
|
||||
return data.recommendations || [];
|
||||
}
|
||||
|
||||
export async function getInferenceCompute(): Promise<InferenceComputeResponse> {
|
||||
const response = await fetch(`${API_BASE}/api/v1/inference-compute`);
|
||||
if (!response.ok) {
|
||||
|
|
|
|||
13
app/ui/app/src/hooks/useFeaturedModels.ts
Normal file
13
app/ui/app/src/hooks/useFeaturedModels.ts
Normal file
|
|
@ -0,0 +1,13 @@
|
|||
import { useQuery } from "@tanstack/react-query";
|
||||
import { getModelRecommendations } from "@/api";
|
||||
import type { ModelRecommendation } from "@/api";
|
||||
|
||||
export function useFeaturedModels() {
|
||||
return useQuery<ModelRecommendation[], Error>({
|
||||
queryKey: ["modelRecommendations"],
|
||||
queryFn: getModelRecommendations,
|
||||
staleTime: 5 * 60 * 1000,
|
||||
gcTime: 30 * 60 * 1000,
|
||||
refetchOnWindowFocus: false,
|
||||
});
|
||||
}
|
||||
|
|
@ -1,51 +1,49 @@
|
|||
import { useQuery } from "@tanstack/react-query";
|
||||
import { Model } from "@/gotypes";
|
||||
import { getModels } from "@/api";
|
||||
import { mergeModels } from "@/utils/mergeModels";
|
||||
import { useMemo } from "react";
|
||||
import { useCloudStatus } from "./useCloudStatus";
|
||||
import { useFeaturedModels } from "./useFeaturedModels";
|
||||
|
||||
export function useModels(searchQuery = "") {
|
||||
const { cloudDisabled } = useCloudStatus();
|
||||
const { data: recommendations, isLoading: recommendationsLoading } =
|
||||
useFeaturedModels();
|
||||
const localQuery = useQuery<Model[], Error>({
|
||||
queryKey: ["models", searchQuery],
|
||||
queryFn: () => getModels(searchQuery),
|
||||
gcTime: 10 * 60 * 1000, // Keep in cache for 10 minutes
|
||||
gcTime: 10 * 60 * 1000,
|
||||
retry: 10,
|
||||
// exponential backoff, starting at 100ms and capping at 5s
|
||||
retryDelay: (attemptIndex) => Math.min(100 * 2 ** attemptIndex, 5000),
|
||||
refetchOnWindowFocus: true,
|
||||
refetchInterval: 30 * 1000, // Refetch every 30 seconds to keep models updated
|
||||
refetchInterval: 30 * 1000,
|
||||
refetchIntervalInBackground: true,
|
||||
});
|
||||
|
||||
const allModels = useMemo(() => {
|
||||
const models = mergeModels(localQuery.data || [], cloudDisabled);
|
||||
const local = localQuery.data || [];
|
||||
const featured = (recommendations || []).map((r) => r.model);
|
||||
const featuredSet = new Set(featured);
|
||||
|
||||
if (searchQuery && searchQuery.trim()) {
|
||||
const query = searchQuery.toLowerCase().trim();
|
||||
const filteredModels = models.filter((model) =>
|
||||
model.model.toLowerCase().includes(query),
|
||||
);
|
||||
// Recommended models first (using the local copy when downloaded),
|
||||
// then everything else from /api/tags in tags order.
|
||||
const recommended = featured.map(
|
||||
(name) =>
|
||||
local.find((m) => m.model === name) || new Model({ model: name }),
|
||||
);
|
||||
const rest = local.filter((m) => !featuredSet.has(m.model));
|
||||
const merged = [...recommended, ...rest];
|
||||
|
||||
const seen = new Set<string>();
|
||||
return filteredModels.filter((model) => {
|
||||
const currentModel = model.model.toLowerCase();
|
||||
if (seen.has(currentModel)) {
|
||||
return false;
|
||||
}
|
||||
seen.add(currentModel);
|
||||
return true;
|
||||
});
|
||||
}
|
||||
|
||||
return models;
|
||||
}, [localQuery.data, searchQuery, cloudDisabled]);
|
||||
const visible = cloudDisabled
|
||||
? merged.filter((m) => !m.isCloud())
|
||||
: merged;
|
||||
return filterBySearch(visible, searchQuery);
|
||||
}, [localQuery.data, searchQuery, cloudDisabled, recommendations]);
|
||||
|
||||
return {
|
||||
...localQuery,
|
||||
data: allModels,
|
||||
isLoading: localQuery.isLoading,
|
||||
isLoading: localQuery.isLoading || recommendationsLoading,
|
||||
};
|
||||
}
|
||||
|
||||
|
|
@ -53,3 +51,16 @@ export function useRefetchModels() {
|
|||
const { refetch } = useModels();
|
||||
return refetch;
|
||||
}
|
||||
|
||||
function filterBySearch(models: Model[], query: string): Model[] {
|
||||
const q = query.trim().toLowerCase();
|
||||
if (!q) return models;
|
||||
|
||||
const seen = new Set<string>();
|
||||
return models.filter((m) => {
|
||||
const name = m.model.toLowerCase();
|
||||
if (!name.includes(q) || seen.has(name)) return false;
|
||||
seen.add(name);
|
||||
return true;
|
||||
});
|
||||
}
|
||||
|
|
|
|||
|
|
@ -4,7 +4,6 @@ import { useModels } from "./useModels";
|
|||
import { useChat } from "./useChats";
|
||||
import { useSettings } from "./useSettings.ts";
|
||||
import { Model } from "@/gotypes";
|
||||
import { FEATURED_MODELS } from "@/utils/mergeModels";
|
||||
import { getTotalVRAM } from "@/utils/vram.ts";
|
||||
import { getInferenceCompute } from "@/api";
|
||||
import { useCloudStatus } from "./useCloudStatus";
|
||||
|
|
@ -92,9 +91,7 @@ export function useSelectedModel(currentChatId?: string, searchQuery?: string) {
|
|||
(settings.selectedModel &&
|
||||
new Model({
|
||||
model: settings.selectedModel,
|
||||
cloud: FEATURED_MODELS.some(
|
||||
(f) => f.endsWith("cloud") && f === settings.selectedModel,
|
||||
),
|
||||
cloud: settings.selectedModel.endsWith("cloud"),
|
||||
ollama_host: false,
|
||||
})) ||
|
||||
null
|
||||
|
|
|
|||
|
|
@ -1,128 +0,0 @@
|
|||
import { describe, it, expect } from "vitest";
|
||||
import { Model } from "@/gotypes";
|
||||
import { mergeModels, FEATURED_MODELS } from "@/utils/mergeModels";
|
||||
import "@/api";
|
||||
|
||||
describe("Model merging logic", () => {
|
||||
it("should handle cloud models with -cloud suffix", () => {
|
||||
const localModels: Model[] = [
|
||||
new Model({ model: "gpt-oss:120b-cloud" }),
|
||||
new Model({ model: "llama3:latest" }),
|
||||
new Model({ model: "mistral:latest" }),
|
||||
];
|
||||
|
||||
const merged = mergeModels(localModels);
|
||||
|
||||
// First verify cloud models are first and in FEATURED_MODELS order
|
||||
const cloudModels = FEATURED_MODELS.filter((m: string) =>
|
||||
m.endsWith("cloud"),
|
||||
);
|
||||
for (let i = 0; i < cloudModels.length; i++) {
|
||||
expect(merged[i].model).toBe(cloudModels[i]);
|
||||
expect(merged[i].isCloud()).toBe(true);
|
||||
}
|
||||
|
||||
// Then verify non-cloud featured models are next and in FEATURED_MODELS order
|
||||
const nonCloudFeatured = FEATURED_MODELS.filter(
|
||||
(m: string) => !m.endsWith("cloud"),
|
||||
);
|
||||
for (let i = 0; i < nonCloudFeatured.length; i++) {
|
||||
const model = merged[i + cloudModels.length];
|
||||
expect(model.model).toBe(nonCloudFeatured[i]);
|
||||
expect(model.isCloud()).toBe(false);
|
||||
}
|
||||
|
||||
// Verify local models are preserved and come after featured models
|
||||
const featuredCount = FEATURED_MODELS.length;
|
||||
expect(merged[featuredCount].model).toBe("llama3:latest");
|
||||
expect(merged[featuredCount + 1].model).toBe("mistral:latest");
|
||||
|
||||
// Length should be exactly featured models plus our local models
|
||||
expect(merged.length).toBe(FEATURED_MODELS.length + 2);
|
||||
});
|
||||
|
||||
it("should hide cloud models when cloud is disabled", () => {
|
||||
const localModels: Model[] = [
|
||||
new Model({ model: "gpt-oss:120b-cloud" }),
|
||||
new Model({ model: "llama3:latest" }),
|
||||
new Model({ model: "mistral:latest" }),
|
||||
];
|
||||
|
||||
const merged = mergeModels(localModels, true); // cloud disabled = true
|
||||
|
||||
// No cloud models should be present
|
||||
const cloudModels = merged.filter((m) => m.isCloud());
|
||||
expect(cloudModels.length).toBe(0);
|
||||
|
||||
// Should have non-cloud featured models
|
||||
const nonCloudFeatured = FEATURED_MODELS.filter(
|
||||
(m) => !m.endsWith("cloud"),
|
||||
);
|
||||
for (let i = 0; i < nonCloudFeatured.length; i++) {
|
||||
const model = merged[i];
|
||||
expect(model.model).toBe(nonCloudFeatured[i]);
|
||||
expect(model.isCloud()).toBe(false);
|
||||
}
|
||||
|
||||
// Local models should be preserved
|
||||
const featuredCount = nonCloudFeatured.length;
|
||||
expect(merged[featuredCount].model).toBe("llama3:latest");
|
||||
expect(merged[featuredCount + 1].model).toBe("mistral:latest");
|
||||
});
|
||||
|
||||
it("should handle empty input", () => {
|
||||
const merged = mergeModels([]);
|
||||
|
||||
// First verify cloud models are first and in FEATURED_MODELS order
|
||||
const cloudModels = FEATURED_MODELS.filter((m) => m.endsWith("cloud"));
|
||||
for (let i = 0; i < cloudModels.length; i++) {
|
||||
expect(merged[i].model).toBe(cloudModels[i]);
|
||||
expect(merged[i].isCloud()).toBe(true);
|
||||
}
|
||||
|
||||
// Then verify non-cloud featured models are next and in FEATURED_MODELS order
|
||||
const nonCloudFeatured = FEATURED_MODELS.filter(
|
||||
(m) => !m.endsWith("cloud"),
|
||||
);
|
||||
for (let i = 0; i < nonCloudFeatured.length; i++) {
|
||||
const model = merged[i + cloudModels.length];
|
||||
expect(model.model).toBe(nonCloudFeatured[i]);
|
||||
expect(model.isCloud()).toBe(false);
|
||||
}
|
||||
|
||||
// Length should be exactly FEATURED_MODELS length
|
||||
expect(merged.length).toBe(FEATURED_MODELS.length);
|
||||
});
|
||||
|
||||
it("should sort models correctly", () => {
|
||||
const localModels: Model[] = [
|
||||
new Model({ model: "zephyr:latest" }),
|
||||
new Model({ model: "alpha:latest" }),
|
||||
new Model({ model: "gpt-oss:120b-cloud" }),
|
||||
];
|
||||
|
||||
const merged = mergeModels(localModels);
|
||||
|
||||
// First verify cloud models are first and in FEATURED_MODELS order
|
||||
const cloudModels = FEATURED_MODELS.filter((m) => m.endsWith("cloud"));
|
||||
for (let i = 0; i < cloudModels.length; i++) {
|
||||
expect(merged[i].model).toBe(cloudModels[i]);
|
||||
expect(merged[i].isCloud()).toBe(true);
|
||||
}
|
||||
|
||||
// Then verify non-cloud featured models are next and in FEATURED_MODELS order
|
||||
const nonCloudFeatured = FEATURED_MODELS.filter(
|
||||
(m) => !m.endsWith("cloud"),
|
||||
);
|
||||
for (let i = 0; i < nonCloudFeatured.length; i++) {
|
||||
const model = merged[i + cloudModels.length];
|
||||
expect(model.model).toBe(nonCloudFeatured[i]);
|
||||
expect(model.isCloud()).toBe(false);
|
||||
}
|
||||
|
||||
// Non-featured local models should be at the end in alphabetical order
|
||||
const featuredCount = FEATURED_MODELS.length;
|
||||
expect(merged[featuredCount].model).toBe("alpha:latest");
|
||||
expect(merged[featuredCount + 1].model).toBe("zephyr:latest");
|
||||
});
|
||||
});
|
||||
|
|
@ -1,102 +0,0 @@
|
|||
import { Model } from "@/gotypes";
|
||||
|
||||
// Featured models list (in priority order)
|
||||
export const FEATURED_MODELS = [
|
||||
"kimi-k2.5:cloud",
|
||||
"glm-5:cloud",
|
||||
"minimax-m2.7:cloud",
|
||||
"gemma4:31b-cloud",
|
||||
"qwen3.5:397b-cloud",
|
||||
"gpt-oss:120b-cloud",
|
||||
"gpt-oss:20b-cloud",
|
||||
"deepseek-v3.1:671b-cloud",
|
||||
"gpt-oss:120b",
|
||||
"gpt-oss:20b",
|
||||
"gemma4:31b",
|
||||
"gemma4:26b",
|
||||
"gemma4:e4b",
|
||||
"gemma4:e2b",
|
||||
"deepseek-r1:8b",
|
||||
"qwen3-coder:30b",
|
||||
"qwen3-vl:30b",
|
||||
"qwen3-vl:8b",
|
||||
"qwen3-vl:4b",
|
||||
"qwen3.5:27b",
|
||||
"qwen3.5:9b",
|
||||
"qwen3.5:4b",
|
||||
];
|
||||
|
||||
function alphabeticalSort(a: Model, b: Model): number {
|
||||
return a.model.toLowerCase().localeCompare(b.model.toLowerCase());
|
||||
}
|
||||
|
||||
//Merges models, sorting cloud models first, then other models
|
||||
export function mergeModels(
|
||||
localModels: Model[],
|
||||
hideCloudModels: boolean = false,
|
||||
): Model[] {
|
||||
const allModels = (localModels || []).map((model) => model);
|
||||
|
||||
// 1. Get cloud models from local models and featured list
|
||||
const cloudModels = [...allModels.filter((m) => m.isCloud())];
|
||||
|
||||
// Add any cloud models from FEATURED_MODELS that aren't in local models
|
||||
FEATURED_MODELS.filter((f) => f.endsWith("cloud")).forEach((cloudModel) => {
|
||||
if (!cloudModels.some((m) => m.model === cloudModel)) {
|
||||
cloudModels.push(new Model({ model: cloudModel }));
|
||||
}
|
||||
});
|
||||
|
||||
// 2. Get other featured models (non-cloud)
|
||||
const featuredModels = FEATURED_MODELS.filter(
|
||||
(f) => !f.endsWith("cloud"),
|
||||
).map((model) => {
|
||||
// Check if this model exists in local models
|
||||
const localMatch = allModels.find(
|
||||
(m) => m.model.toLowerCase() === model.toLowerCase(),
|
||||
);
|
||||
|
||||
if (localMatch) return localMatch;
|
||||
|
||||
return new Model({
|
||||
model,
|
||||
});
|
||||
});
|
||||
|
||||
// 3. Get remaining local models that aren't featured and aren't cloud models
|
||||
const remainingModels = allModels.filter(
|
||||
(model) =>
|
||||
!model.isCloud() &&
|
||||
!FEATURED_MODELS.some(
|
||||
(f) => f.toLowerCase() === model.model.toLowerCase(),
|
||||
),
|
||||
);
|
||||
|
||||
cloudModels.sort((a, b) => {
|
||||
const aIndex = FEATURED_MODELS.indexOf(a.model);
|
||||
const bIndex = FEATURED_MODELS.indexOf(b.model);
|
||||
|
||||
// If both are featured, sort by their position in FEATURED_MODELS
|
||||
if (aIndex !== -1 && bIndex !== -1) {
|
||||
return aIndex - bIndex;
|
||||
}
|
||||
|
||||
// If only one is featured, featured model comes first
|
||||
if (aIndex !== -1 && bIndex === -1) return -1;
|
||||
if (aIndex === -1 && bIndex !== -1) return 1;
|
||||
|
||||
// If neither is featured, sort alphabetically
|
||||
return a.model.toLowerCase().localeCompare(b.model.toLowerCase());
|
||||
});
|
||||
|
||||
featuredModels.sort(
|
||||
(a, b) =>
|
||||
FEATURED_MODELS.indexOf(a.model) - FEATURED_MODELS.indexOf(b.model),
|
||||
);
|
||||
|
||||
remainingModels.sort(alphabeticalSort);
|
||||
|
||||
return hideCloudModels
|
||||
? [...featuredModels, ...remainingModels]
|
||||
: [...cloudModels, ...featuredModels, ...remainingModels];
|
||||
}
|
||||
|
|
@ -302,6 +302,7 @@ func (s *Server) Handler() http.Handler {
|
|||
mux.Handle("HEAD /api/version", ollamaProxy)
|
||||
mux.Handle("POST /api/me", ollamaProxy)
|
||||
mux.Handle("POST /api/signout", ollamaProxy)
|
||||
mux.Handle("GET /api/experimental/model-recommendations", ollamaProxy)
|
||||
|
||||
// React app - catch all non-API routes and serve the React app
|
||||
mux.Handle("GET /", s.appHandler())
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue