Files
majordomo/provider/llamaswap/music_embed_test.go
T
steve cacce61ddc feat: musicgen + embeddings/rerank surfaces (ADR-0021, ADR-0022)
- NEW musicgen leaf package: blocking Generate over ACE-Step's async job
  queue (release_task -> poll query_result -> fetch file, all via
  /upstream); tolerant envelope parsing, double-encoded result handled
- NEW embeddings leaf package: EmbedModel + RerankModel as separate mints
  (two server instances on the host, llama.cpp #20085); InstructedQuery
  helper for Qwen3-style query/document asymmetry
- provider/llamaswap: /v1/embeddings + /v1/rerank clients with strict
  validation (index-ordered vectors, count mismatch and out-of-range
  index are hard errors; rerank sorted descending, minimal parser)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-13 00:50:36 -04:00

225 lines
7.7 KiB
Go

package llamaswap
import (
"context"
"encoding/json"
"errors"
"net/http"
"net/http/httptest"
"sync/atomic"
"testing"
"time"
"gitea.stevedudenhoeffer.com/steve/majordomo/embeddings"
"gitea.stevedudenhoeffer.com/steve/majordomo/llm"
"gitea.stevedudenhoeffer.com/steve/majordomo/musicgen"
)
// aceStepStub emulates the ACE-Step job API: one queued poll, then success.
func aceStepStub(t *testing.T, mp3 []byte) (*httptest.Server, *atomic.Int32) {
t.Helper()
var polls atomic.Int32
var gotRelease map[string]any
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
switch r.URL.Path {
case "/upstream/musicgen-acestep/release_task":
_ = json.NewDecoder(r.Body).Decode(&gotRelease)
if gotRelease["task_type"] != "text2music" {
t.Errorf("task_type = %v", gotRelease["task_type"])
}
_, _ = w.Write([]byte(`{"data": {"task_id": "t-1", "status": "queued"}}`))
case "/upstream/musicgen-acestep/query_result":
n := polls.Add(1)
if n == 1 {
_, _ = w.Write([]byte(`{"data": [{"task_id": "t-1", "status": 0, "result": ""}]}`))
return
}
_, _ = w.Write([]byte(`{"data": [{"task_id": "t-1", "status": 1,
"result": "{\"file\": \"/v1/audio?path=out.mp3\", \"metas\": {\"bpm\": 120}}"}]}`))
case "/upstream/musicgen-acestep/v1/audio":
if got := r.URL.Query().Get("path"); got != "out.mp3" {
t.Errorf("audio path = %q", got)
}
w.Header().Set("Content-Type", "audio/mpeg")
_, _ = w.Write(mp3)
default:
t.Errorf("unexpected path %q", r.URL.Path)
w.WriteHeader(404)
}
}))
return srv, &polls
}
func TestMusicGenerate(t *testing.T) {
mp3 := []byte("ID3fakeaudio")
srv, polls := aceStepStub(t, mp3)
defer srv.Close()
p := New(WithBaseURL(srv.URL), WithHTTPClient(srv.Client()))
mm, err := p.MusicModel("musicgen-acestep")
if err != nil {
t.Fatalf("MusicModel: %v", err)
}
// Shrink the poll interval indirectly by bounding the whole call: the
// stub succeeds on poll #2, so a generous deadline still finishes fast.
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer cancel()
res, err := mm.Generate(ctx,
musicgen.Request{Prompt: "chiptune anthem about mortbux"},
musicgen.WithLyrics("mort mort mort"), musicgen.WithDuration(30))
if err != nil {
t.Fatalf("Generate: %v", err)
}
if polls.Load() < 2 {
t.Errorf("polls = %d, want >= 2 (queued then done)", polls.Load())
}
if res.Audio.MIME != "audio/mpeg" || string(res.Audio.Data) != string(mp3) {
t.Fatalf("audio = %q/%d bytes", res.Audio.MIME, len(res.Audio.Data))
}
}
func TestMusicGenerateUpstreamFailure(t *testing.T) {
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
switch r.URL.Path {
case "/upstream/musicgen-acestep/release_task":
_, _ = w.Write([]byte(`{"data": {"task_id": "t-2"}}`))
case "/upstream/musicgen-acestep/query_result":
_, _ = w.Write([]byte(`{"data": [{"task_id": "t-2", "status": 2, "result": "{\"error\": \"OOM\"}"}]}`))
}
}))
defer srv.Close()
p := New(WithBaseURL(srv.URL), WithHTTPClient(srv.Client()))
mm, _ := p.MusicModel("musicgen-acestep")
_, err := mm.Generate(context.Background(), musicgen.Request{Prompt: "p"})
var apiErr *llm.APIError
if !errors.As(err, &apiErr) {
t.Fatalf("err = %v, want APIError for failed job", err)
}
}
func TestMusicGenerateRejectsBadArgs(t *testing.T) {
p := New(WithBaseURL("http://unused"))
mm, _ := p.MusicModel("musicgen-acestep")
if _, err := mm.Generate(context.Background(), musicgen.Request{}); !errors.Is(err, llm.ErrUnsupported) {
t.Errorf("no prompt: err = %v, want ErrUnsupported", err)
}
}
func TestEmbed(t *testing.T) {
var gotBody map[string]any
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.URL.Path != "/v1/embeddings" {
t.Errorf("path = %q", r.URL.Path)
}
_ = json.NewDecoder(r.Body).Decode(&gotBody)
// Deliberately out of order: the client must sort by index.
_, _ = w.Write([]byte(`{"object":"list","data":[
{"object":"embedding","index":1,"embedding":[0.3,0.4]},
{"object":"embedding","index":0,"embedding":[0.1,0.2]}
]}`))
}))
defer srv.Close()
p := New(WithBaseURL(srv.URL), WithHTTPClient(srv.Client()))
em, err := p.EmbedModel("embed-qwen3-0.6b")
if err != nil {
t.Fatalf("EmbedModel: %v", err)
}
res, err := em.Embed(context.Background(), embeddings.EmbedRequest{Inputs: []string{"a", "b"}})
if err != nil {
t.Fatalf("Embed: %v", err)
}
if gotBody["model"] != "embed-qwen3-0.6b" {
t.Errorf("model = %v", gotBody["model"])
}
if len(res.Vectors) != 2 || res.Vectors[0][0] != 0.1 || res.Vectors[1][0] != 0.3 {
t.Fatalf("vectors = %+v (index ordering broken?)", res.Vectors)
}
}
func TestEmbedCountMismatchIsError(t *testing.T) {
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
_, _ = w.Write([]byte(`{"data":[{"index":0,"embedding":[0.1]}]}`))
}))
defer srv.Close()
p := New(WithBaseURL(srv.URL), WithHTTPClient(srv.Client()))
em, _ := p.EmbedModel("embed-qwen3-0.6b")
_, err := em.Embed(context.Background(), embeddings.EmbedRequest{Inputs: []string{"a", "b"}})
var apiErr *llm.APIError
if !errors.As(err, &apiErr) {
t.Fatalf("err = %v, want APIError for count mismatch", err)
}
}
func TestEmbedRejectsEmptyInputs(t *testing.T) {
p := New(WithBaseURL("http://unused"))
em, _ := p.EmbedModel("embed-qwen3-0.6b")
if _, err := em.Embed(context.Background(), embeddings.EmbedRequest{}); !errors.Is(err, llm.ErrUnsupported) {
t.Errorf("no inputs: err = %v, want ErrUnsupported", err)
}
if _, err := em.Embed(context.Background(), embeddings.EmbedRequest{Inputs: []string{"a", " "}}); !errors.Is(err, llm.ErrUnsupported) {
t.Errorf("blank input: err = %v, want ErrUnsupported", err)
}
}
func TestRerank(t *testing.T) {
var gotBody map[string]any
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.URL.Path != "/v1/rerank" {
t.Errorf("path = %q", r.URL.Path)
}
_ = json.NewDecoder(r.Body).Decode(&gotBody)
// Out of score order: the client must sort descending.
_, _ = w.Write([]byte(`{"results":[
{"index":0,"relevance_score":0.11},
{"index":2,"relevance_score":0.93},
{"index":1,"relevance_score":0.42}
]}`))
}))
defer srv.Close()
p := New(WithBaseURL(srv.URL), WithHTTPClient(srv.Client()))
rm, err := p.RerankModel("rerank-bge-v2-m3")
if err != nil {
t.Fatalf("RerankModel: %v", err)
}
res, err := rm.Rerank(context.Background(),
embeddings.RerankRequest{Query: "what is a panda?", Documents: []string{"a", "b", "c"}},
embeddings.WithTopN(3))
if err != nil {
t.Fatalf("Rerank: %v", err)
}
if gotBody["top_n"] != float64(3) || gotBody["query"] != "what is a panda?" {
t.Errorf("top_n/query = %v/%v", gotBody["top_n"], gotBody["query"])
}
if len(res.Results) != 3 || res.Results[0].Index != 2 || res.Results[2].Index != 0 {
t.Fatalf("results = %+v (descending sort broken?)", res.Results)
}
}
func TestRerankRejectsOutOfRangeIndex(t *testing.T) {
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
_, _ = w.Write([]byte(`{"results":[{"index":7,"relevance_score":0.9}]}`))
}))
defer srv.Close()
p := New(WithBaseURL(srv.URL), WithHTTPClient(srv.Client()))
rm, _ := p.RerankModel("rerank-bge-v2-m3")
_, err := rm.Rerank(context.Background(),
embeddings.RerankRequest{Query: "q", Documents: []string{"a"}})
var apiErr *llm.APIError
if !errors.As(err, &apiErr) {
t.Fatalf("err = %v, want APIError for out-of-range index", err)
}
}
func TestInstructedQuery(t *testing.T) {
got := embeddings.InstructedQuery("", "how tall is everest")
want := "Instruct: Given a web search query, retrieve relevant passages that answer the query\nQuery: how tall is everest"
if got != want {
t.Errorf("InstructedQuery = %q", got)
}
}