mirror of
https://github.com/hanzoai/ai.git
synced 2026-11-20 14:13:23 +00:00
The Zen family (zen5*, zen3*, zen4*, and the zen-* aliases) is discovered from zen's GET /v1/models and routed through the zen provider (zen_client.go), which owns the SKU→upstream mapping, pricing, identity, and reasoning vocabulary. Delete the static zen routes, the zen entries in conf/models.yaml, the per-SKU identity prompts (and the now-unused GetIdentityPrompt machinery), and the zen pricing tables. ai keeps only non-zen routes (raw upstreams stay runtime-configurable via admin.hanzo.ai). Tests assert ai now holds no static zen route.
141 lines
4.4 KiB
Go
141 lines
4.4 KiB
Go
// Copyright 2023-2025 Hanzo AI Inc. All Rights Reserved.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
package controllers
|
|
|
|
import (
|
|
"os"
|
|
"path/filepath"
|
|
"testing"
|
|
)
|
|
|
|
// The same model is served at different sizes by different providers: GLM-5.2
|
|
// is a 1M-context model, but DigitalOcean serves it at 262144. The window is
|
|
// therefore a property of the (provider, upstream) pair, and a prompt too big
|
|
// for one provider should be routed to one that can hold it — not refused.
|
|
const ctxRouteYAML = `
|
|
version: 1
|
|
models:
|
|
# The model default: what glm-5.2 IS.
|
|
glm-5.2:
|
|
provider: do-ai
|
|
upstream: glm-5.2
|
|
context_window: 262144 # ...but this is what DO serves it at.
|
|
|
|
# A provider that serves the SAME model with its full window.
|
|
glm-5.2-wide:
|
|
provider: nebius
|
|
upstream: glm-5.2
|
|
context_window: 1048576
|
|
|
|
zen5:
|
|
provider: do-ai
|
|
upstream: glm-5.2
|
|
fallbacks:
|
|
- { provider: nebius, upstream: glm-5.2, context_window: 1048576 }
|
|
|
|
# No window declared anywhere in the chain.
|
|
mystery:
|
|
provider: do-ai
|
|
upstream: some-new-model
|
|
`
|
|
|
|
func loadCtxRouteConfig(t *testing.T) *ModelConfig {
|
|
t.Helper()
|
|
path := filepath.Join(t.TempDir(), "models.yaml")
|
|
if err := os.WriteFile(path, []byte(ctxRouteYAML), 0o644); err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
mc := &ModelConfig{
|
|
routes: make(map[string]modelRoute),
|
|
pricing: make(map[string]modelPrice),
|
|
stopCh: make(chan struct{}),
|
|
}
|
|
if err := mc.loadFromFile(path); err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
return mc
|
|
}
|
|
|
|
func TestContextWindow_AliasInheritsUpstream(t *testing.T) {
|
|
mc := loadCtxRouteConfig(t)
|
|
|
|
// zen5 declares no window; it inherits what its upstream is served at.
|
|
if got := mc.ContextWindow("zen5"); got != 262144 {
|
|
t.Errorf("zen5 should inherit glm-5.2's served window 262144, got %d", got)
|
|
}
|
|
// Undeclared anywhere → 0, and the caller applies the floor.
|
|
if got := mc.ContextWindow("mystery"); got != 0 {
|
|
t.Errorf("undeclared model: want 0 (caller uses the floor), got %d", got)
|
|
}
|
|
}
|
|
|
|
func TestRouteForContext_FitsPrimary(t *testing.T) {
|
|
mc := loadCtxRouteConfig(t)
|
|
|
|
// A prompt the primary can hold stays on the primary.
|
|
prov, up, win, ok := mc.RouteForContext("zen5", 100_000)
|
|
if !ok {
|
|
t.Fatal("zen5 should resolve")
|
|
}
|
|
if prov != "do-ai" || up != "glm-5.2" || win != 262144 {
|
|
t.Errorf("a prompt that fits must stay on the primary: got provider=%s upstream=%s window=%d", prov, up, win)
|
|
}
|
|
}
|
|
|
|
func TestRouteForContext_RerouteToProviderThatFits(t *testing.T) {
|
|
mc := loadCtxRouteConfig(t)
|
|
|
|
// 400K exceeds what DO serves glm-5.2 at (262144). The request must NOT be
|
|
// refused — it must be routed to the provider that serves the same model
|
|
// with a window big enough to hold it.
|
|
prov, up, win, ok := mc.RouteForContext("zen5", 400_000)
|
|
if !ok {
|
|
t.Fatal("zen5 should resolve")
|
|
}
|
|
if prov != "nebius" {
|
|
t.Errorf("a 400K prompt must reroute to the provider that can serve it: got provider=%s (window %d)", prov, win)
|
|
}
|
|
if up != "glm-5.2" {
|
|
t.Errorf("rerouting must keep the SAME model, got upstream=%s", up)
|
|
}
|
|
if win != 1048576 {
|
|
t.Errorf("window should be the fallback provider's: want 1048576, got %d", win)
|
|
}
|
|
}
|
|
|
|
func TestRouteForContext_NothingFitsKeepsPrimary(t *testing.T) {
|
|
mc := loadCtxRouteConfig(t)
|
|
|
|
// Bigger than every provider in the chain: return the primary and let the
|
|
// upstream reject it. An honest error from the model beats a guess from us.
|
|
prov, _, _, ok := mc.RouteForContext("zen5", 5_000_000)
|
|
if !ok {
|
|
t.Fatal("zen5 should resolve")
|
|
}
|
|
if prov != "do-ai" {
|
|
t.Errorf("when nothing fits, keep the primary and let the upstream speak: got %s", prov)
|
|
}
|
|
}
|
|
|
|
func TestRouteForContext_UnknownSizeTakesPrimary(t *testing.T) {
|
|
mc := loadCtxRouteConfig(t)
|
|
|
|
// tokens <= 0 means "size unknown" — no basis to reroute.
|
|
prov, _, _, ok := mc.RouteForContext("zen5", 0)
|
|
if !ok || prov != "do-ai" {
|
|
t.Errorf("unknown prompt size must take the primary, got provider=%s ok=%v", prov, ok)
|
|
}
|
|
}
|