CERBERUS/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-10-09 09:47:06 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

chore: remove old imagegen LLMs models (#14597)

Browse source 
These models are implemented in the x/mlxrunner instead.
This commit is contained in:
Patrick Devine 2026-03-03 13:23:40 -08:00 • committed by GitHub
parent 799e51d419
commit 110eff01a9
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
20 changed files with 49 additions and 3919 deletions
Download patch file Download diff file Expand all files Collapse all files

4
server/routes.go
View file

@ -161,7 +161,7 @@ func (s *Server) scheduleRunner(ctx context.Context, name string, caps []model.C
return nil, nil, nil, fmt.Errorf("%s %w", name, err)
}
useImagegen, _ := requestOpts["use_imagegen_runner"].(bool)
// Deprecated runner override option; ignore if present.
delete(requestOpts, "use_imagegen_runner")
opts, err := s.modelOptions(model, requestOpts)
@ -169,7 +169,7 @@ func (s *Server) scheduleRunner(ctx context.Context, name string, caps []model.C
return nil, nil, nil, err
}
runnerCh, errCh := s.sched.GetRunner(ctx, model, opts, keepAlive, useImagegen)
runnerCh, errCh := s.sched.GetRunner(ctx, model, opts, keepAlive)
var runner *runnerRef
select {
case runner = <-runnerCh:

21
server/sched.go
View file

@ -33,7 +33,6 @@ type LlmRequest struct {
successCh chan *runnerRef
errCh chan error
schedAttempts uint
useImagegen bool
}
type Scheduler struct {
@ -106,7 +105,7 @@ func schedulerModelKey(m *Model) string {
}
// context must be canceled to decrement ref count and release the runner
func (s *Scheduler) GetRunner(c context.Context, m *Model, opts api.Options, sessionDuration *api.Duration, useImagegen bool) (chan *runnerRef, chan error) {
func (s *Scheduler) GetRunner(c context.Context, m *Model, opts api.Options, sessionDuration *api.Duration) (chan *runnerRef, chan error) {
if opts.NumCtx < 4 {
opts.NumCtx = 4
}
@ -123,7 +122,6 @@ func (s *Scheduler) GetRunner(c context.Context, m *Model, opts api.Options, ses
sessionDuration: sessionDuration,
successCh: make(chan *runnerRef, 1),
errCh: make(chan error, 1),
useImagegen: useImagegen,
}
key := schedulerModelKey(req.model)
@ -593,20 +591,15 @@ iGPUScan:
return false
}
// loadMLX loads an experimental safetensors model using the unified MLX runner.
// This supports both LLM (completion) and image generation models.
// loadMLX loads an experimental safetensors model using MLX runners.
// Image models use x/imagegen; LLM models use x/mlxrunner.
func (s *Scheduler) loadMLX(req *LlmRequest) bool {
modelName := req.model.ShortName
var server llm.LlamaServer
var err error
isImagegen := false
if slices.Contains(req.model.Config.Capabilities, "image") {
server, err = imagegen.NewServer(modelName, imagegen.ModeImageGen)
isImagegen = true
} else if req.useImagegen {
server, err = imagegen.NewServer(modelName, imagegen.ModeLLM)
isImagegen = true
server, err = imagegen.NewServer(modelName)
} else {
server, err = mlxrunner.NewClient(modelName)
}
@ -628,7 +621,7 @@ func (s *Scheduler) loadMLX(req *LlmRequest) bool {
llama: server,
Options: &req.opts,
loading: false,
isImagegen: isImagegen,
isImagegen: slices.Contains(req.model.Config.Capabilities, "image"),
sessionDuration: sessionDuration,
totalSize: totalSize,
vramSize: vramSize,
@ -737,8 +730,8 @@ func (runner *runnerRef) needsReload(ctx context.Context, req *LlmRequest) bool
runner.refMu.Lock()
defer runner.refMu.Unlock()
// Check if runner type (imagegen vs mlxrunner) matches what's requested
wantImagegen := req.useImagegen || slices.Contains(req.model.Config.Capabilities, "image")
// Check if runner type (imagegen vs mlxrunner) matches what's requested.
wantImagegen := slices.Contains(req.model.Config.Capabilities, "image")
if runner.isImagegen != wantImagegen {
return true
}

12
server/sched_test.go
View file

@ -408,10 +408,10 @@ func TestSchedGetRunner(t *testing.T) {
s.getSystemInfoFn = getSystemInfoFn
s.newServerFn = a.newServer
slog.Info("a")
successCh1a, errCh1a := s.GetRunner(a.ctx, a.req.model, a.req.opts, a.req.sessionDuration, false)
successCh1a, errCh1a := s.GetRunner(a.ctx, a.req.model, a.req.opts, a.req.sessionDuration)
require.Len(t, s.pendingReqCh, 1)
slog.Info("b")
successCh1b, errCh1b := s.GetRunner(b.ctx, b.req.model, b.req.opts, b.req.sessionDuration, false)
successCh1b, errCh1b := s.GetRunner(b.ctx, b.req.model, b.req.opts, b.req.sessionDuration)
require.Len(t, s.pendingReqCh, 1)
require.Empty(t, successCh1b)
require.Len(t, errCh1b, 1)
@ -435,7 +435,7 @@ func TestSchedGetRunner(t *testing.T) {
c.req.model.ModelPath = "bad path"
slog.Info("c")
successCh1c, errCh1c := s.GetRunner(c.ctx, c.req.model, c.req.opts, c.req.sessionDuration, false)
successCh1c, errCh1c := s.GetRunner(c.ctx, c.req.model, c.req.opts, c.req.sessionDuration)
// Starts in pending channel, then should be quickly processed to return an error
time.Sleep(50 * time.Millisecond) // Long enough for the "a" model to expire and unload
require.Empty(t, successCh1c)
@ -470,7 +470,7 @@ func TestSchedGetRunnerUsesDigestKeyWhenModelPathEmpty(t *testing.T) {
s.loadedMu.Unlock()
reqModel := &Model{Name: "safetensors-b", Digest: "sha-b"}
successCh, errCh := s.GetRunner(ctx, reqModel, opts, nil, false)
successCh, errCh := s.GetRunner(ctx, reqModel, opts, nil)
require.Empty(t, successCh)
require.Empty(t, errCh)
@ -499,7 +499,7 @@ func TestSchedGetRunnerReusesSameDigestWhenModelPathEmpty(t *testing.T) {
s.loadedMu.Unlock()
reqCtx, cancelReq := context.WithCancel(ctx)
successCh, errCh := s.GetRunner(reqCtx, &Model{Name: "safetensors-a-copy", Digest: "sha-a"}, opts, nil, false)
successCh, errCh := s.GetRunner(reqCtx, &Model{Name: "safetensors-a-copy", Digest: "sha-a"}, opts, nil)
cancelReq()
select {
@ -574,7 +574,7 @@ func TestSchedPrematureExpired(t *testing.T) {
s.getGpuFn = getGpuFn
s.getSystemInfoFn = getSystemInfoFn
s.newServerFn = scenario1a.newServer
successCh1a, errCh1a := s.GetRunner(scenario1a.ctx, scenario1a.req.model, scenario1a.req.opts, scenario1a.req.sessionDuration, false)
successCh1a, errCh1a := s.GetRunner(scenario1a.ctx, scenario1a.req.model, scenario1a.req.opts, scenario1a.req.sessionDuration)
require.Len(t, s.pendingReqCh, 1)
s.Run(ctx)
select {