Merge branch 'master' into test-authv2

mudler · Sep 17, 2024 · e04ad74 · e04ad74
2 parents c8d9d3c + eee1fb2
commit e04ad74
Show file tree

Hide file tree

Showing 10 changed files with 37 additions and 24 deletions.
diff --git a/backend/python/openvoice/requirements-intel.txt b/backend/python/openvoice/requirements-intel.txt
@@ -15,7 +15,7 @@ unidecode==1.3.7
 whisper-timestamped==1.15.4
 openai
 python-dotenv
-pypinyin==0.50.0
+pypinyin==0.53.0
 cn2an==0.5.22
 jieba==0.42.1
 gradio==4.38.1

diff --git a/backend/python/sentencetransformers/backend.py b/backend/python/sentencetransformers/backend.py
@@ -55,7 +55,7 @@ def LoadModel(self, request, context):
         """
         model_name = request.Model
         try:
-            self.model = SentenceTransformer(model_name)
+            self.model = SentenceTransformer(model_name, trust_remote_code=request.TrustRemoteCode)
         except Exception as err:
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
 

diff --git a/backend/python/sentencetransformers/requirements.txt b/backend/python/sentencetransformers/requirements.txt
@@ -1,3 +1,5 @@
 grpcio==1.66.1
 protobuf
-certifi
+certifi
+datasets
+einops
diff --git a/backend/python/transformers/requirements.txt b/backend/python/transformers/requirements.txt
@@ -1,4 +1,4 @@
 grpcio==1.66.1
 protobuf
 certifi
-setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
diff --git a/examples/functions/requirements.txt b/examples/functions/requirements.txt
@@ -1,2 +1,2 @@
 langchain==0.3.0
-openai==1.44.0
+openai==1.45.1
diff --git a/examples/langchain-chroma/requirements.txt b/examples/langchain-chroma/requirements.txt
@@ -1,4 +1,4 @@
-langchain==0.2.16
+langchain==0.3.0
 openai==1.45.1
 chromadb==0.5.5
 llama-index==0.11.7
diff --git a/examples/langchain/langchainpy-localai-example/requirements.txt b/examples/langchain/langchainpy-localai-example/requirements.txt
@@ -10,15 +10,15 @@ debugpy==1.8.2
 frozenlist==1.4.1
 greenlet==3.1.0
 idna==3.8
-langchain==0.2.16
+langchain==0.3.0
 langchain-community==0.2.16
 marshmallow==3.22.0
 marshmallow-enum==1.5.1
 multidict==6.0.5
 mypy-extensions==1.0.0
 numexpr==2.10.1
 numpy==2.1.1
-openai==1.44.0
+openai==1.45.1
 openapi-schema-pydantic==1.2.4
 packaging>=23.2
 pydantic==2.8.2

diff --git a/gallery/index.yaml b/gallery/index.yaml
@@ -1281,7 +1281,7 @@
 - !!merge <<: *mistral03
   name: "mn-12b-lyra-v4-iq-imatrix"
   icon: https://cdn-uploads.huggingface.co/production/uploads/65d4cf2693a0a3744a27536c/dVoru83WOpwVjMlgZ_xhA.png
-  #chatml
+  # chatml
   url: "github:mudler/LocalAI/gallery/chatml.yaml@master"
   urls:
     - https://huggingface.co/Lewdiculous/MN-12B-Lyra-v4-GGUF-IQ-Imatrix

diff --git a/pkg/model/loader.go b/pkg/model/loader.go
@@ -69,6 +69,8 @@ var knownModelsNameSuffixToSkip []string = []string{
 	".tar.gz",
 }
 
+const retryTimeout = time.Duration(2 * time.Minute)
+
 func (ml *ModelLoader) ListFilesInModelPath() ([]string, error) {
 	files, err := os.ReadDir(ml.ModelPath)
 	if err != nil {
@@ -146,15 +148,23 @@ func (ml *ModelLoader) ShutdownModel(modelName string) error {
 	ml.mu.Lock()
 	defer ml.mu.Unlock()
 
-	return ml.stopModel(modelName)
-}
-
-func (ml *ModelLoader) stopModel(modelName string) error {
-	defer ml.deleteProcess(modelName)
-	if _, ok := ml.models[modelName]; !ok {
+	_, ok := ml.models[modelName]
+	if !ok {
 		return fmt.Errorf("model %s not found", modelName)
 	}
-	return nil
+
+	retries := 1
+	for ml.models[modelName].GRPC(false, ml.wd).IsBusy() {
+		log.Debug().Msgf("%s busy. Waiting.", modelName)
+		dur := time.Duration(retries*2) * time.Second
+		if dur > retryTimeout {
+			dur = retryTimeout
+		}
+		time.Sleep(dur)
+		retries++
+	}
+
+	return ml.deleteProcess(modelName)
 }
 
 func (ml *ModelLoader) CheckIsLoaded(s string) *Model {

diff --git a/pkg/model/process.go b/pkg/model/process.go
@@ -18,15 +18,16 @@ import (
 
 func (ml *ModelLoader) StopAllExcept(s string) error {
 	return ml.StopGRPC(func(id string, p *process.Process) bool {
-		if id != s {
-			for ml.models[id].GRPC(false, ml.wd).IsBusy() {
-				log.Debug().Msgf("%s busy. Waiting.", id)
-				time.Sleep(2 * time.Second)
-			}
-			log.Debug().Msgf("[single-backend] Stopping %s", id)
-			return true
+		if id == s {
+			return false
 		}
-		return false
+
+		for ml.models[id].GRPC(false, ml.wd).IsBusy() {
+			log.Debug().Msgf("%s busy. Waiting.", id)
+			time.Sleep(2 * time.Second)
+		}
+		log.Debug().Msgf("[single-backend] Stopping %s", id)
+		return true
 	})
 }