[pull] master from mudler:master

backend/cpp/llama-cpp/Makefile

@@ -1,5 +1,5 @@
 
-LLAMA_VERSION?=fd1234cb468935ea087d6929b2487926c3afff4b
+LLAMA_VERSION?=e725a1a982ca870404a9c4935df52466327bbd02
 LLAMA_REPO?=https://github.com/ggerganov/llama.cpp
 
 CMAKE_ARGS?=

backend/cpp/llama-cpp/grpc-server.cpp

@@ -313,9 +313,11 @@ static void params_parse(const backend::ModelOptions* request,
         params.pooling_type = LLAMA_POOLING_TYPE_RANK;
     }
 
+
     if (request->ropescaling() == "none")   { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_NONE; }
     else if (request->ropescaling() == "yarn")   { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_YARN; }
-    else { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_LINEAR; }
+    else if (request->ropescaling() == "linear")   {  params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_LINEAR; }
+
     if ( request->yarnextfactor() != 0.0f ) {
         params.yarn_ext_factor = request->yarnextfactor();
     }

core/http/endpoints/openai/chat.go

@@ -305,7 +305,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 		// If we are using the tokenizer template, we don't need to process the messages
 		// unless we are processing functions
 		if !config.TemplateConfig.UseTokenizerTemplate || shouldUseFn {
-			predInput = evaluator.TemplateMessages(input.Messages, config, funcs, shouldUseFn)
+			predInput = evaluator.TemplateMessages(*input, input.Messages, config, funcs, shouldUseFn)
 
 			log.Debug().Msgf("Prompt (after templating): %s", predInput)
 			if config.Grammar != "" {

core/http/endpoints/openai/completion.go

@@ -109,8 +109,10 @@ func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, e
 			predInput := config.PromptStrings[0]
 
 			templatedInput, err := evaluator.EvaluateTemplateForPrompt(templates.CompletionPromptTemplate, *config, templates.PromptTemplateData{
-				Input:        predInput,
-				SystemPrompt: config.SystemPrompt,
+				Input:           predInput,
+				SystemPrompt:    config.SystemPrompt,
+				ReasoningEffort: input.ReasoningEffort,
+				Metadata:        input.Metadata,
 			})
 			if err == nil {
 				predInput = templatedInput
@@ -160,8 +162,10 @@ func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, e
 
 		for k, i := range config.PromptStrings {
 			templatedInput, err := evaluator.EvaluateTemplateForPrompt(templates.CompletionPromptTemplate, *config, templates.PromptTemplateData{
-				SystemPrompt: config.SystemPrompt,
-				Input:        i,
+				SystemPrompt:    config.SystemPrompt,
+				Input:           i,
+				ReasoningEffort: input.ReasoningEffort,
+				Metadata:        input.Metadata,
 			})
 			if err == nil {
 				i = templatedInput

core/http/endpoints/openai/edit.go

@@ -47,9 +47,11 @@ func EditEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 
 		for _, i := range config.InputStrings {
 			templatedInput, err := evaluator.EvaluateTemplateForPrompt(templates.EditPromptTemplate, *config, templates.PromptTemplateData{
-				Input:        i,
-				Instruction:  input.Instruction,
-				SystemPrompt: config.SystemPrompt,
+				Input:           i,
+				Instruction:     input.Instruction,
+				SystemPrompt:    config.SystemPrompt,
+				ReasoningEffort: input.ReasoningEffort,
+				Metadata:        input.Metadata,
 			})
 			if err == nil {
 				i = templatedInput

core/schema/openai.go

@@ -183,6 +183,10 @@ type OpenAIRequest struct {
 	Backend string `json:"backend" yaml:"backend"`
 
 	ModelBaseName string `json:"model_base_name" yaml:"model_base_name"`
+
+	ReasoningEffort string `json:"reasoning_effort" yaml:"reasoning_effort"`
+
+	Metadata map[string]string `json:"metadata" yaml:"metadata"`
 }
 
 type ModelsDataResponse struct {

core/templates/evaluator.go

@@ -21,6 +21,8 @@ type PromptTemplateData struct {
 	Instruction          string
 	Functions            []functions.Function
 	MessageIndex         int
+	ReasoningEffort      string
+	Metadata             map[string]string
 }
 
 type ChatMessageTemplateData struct {
@@ -133,7 +135,7 @@ func (e *Evaluator) evaluateJinjaTemplateForPrompt(templateType TemplateType, te
 	return e.cache.evaluateJinjaTemplate(templateType, templateName, conversation)
 }
 
-func (e *Evaluator) TemplateMessages(messages []schema.Message, config *config.BackendConfig, funcs []functions.Function, shouldUseFn bool) string {
+func (e *Evaluator) TemplateMessages(input schema.OpenAIRequest, messages []schema.Message, config *config.BackendConfig, funcs []functions.Function, shouldUseFn bool) string {
 
 	if config.TemplateConfig.JinjaTemplate {
 		var messageData []ChatMessageTemplateData
@@ -283,6 +285,8 @@ func (e *Evaluator) TemplateMessages(messages []schema.Message, config *config.B
 		SuppressSystemPrompt: suppressConfigSystemPrompt,
 		Input:                predInput,
 		Functions:            funcs,
+		ReasoningEffort:      input.ReasoningEffort,
+		Metadata:             input.Metadata,
 	})
 	if err == nil {
 		predInput = templatedInput

core/templates/evaluator_test.go

@@ -219,7 +219,7 @@ var _ = Describe("Templates", func() {
 		for key := range chatMLTestMatch {
 			foo := chatMLTestMatch[key]
 			It("renders correctly `"+key+"`", func() {
-				templated := evaluator.TemplateMessages(foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
+				templated := evaluator.TemplateMessages(schema.OpenAIRequest{}, foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
 				Expect(templated).To(Equal(foo["expected"]), templated)
 			})
 		}
@@ -232,7 +232,7 @@ var _ = Describe("Templates", func() {
 		for key := range llama3TestMatch {
 			foo := llama3TestMatch[key]
 			It("renders correctly `"+key+"`", func() {
-				templated := evaluator.TemplateMessages(foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
+				templated := evaluator.TemplateMessages(schema.OpenAIRequest{}, foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
 				Expect(templated).To(Equal(foo["expected"]), templated)
 			})
 		}
@@ -245,7 +245,7 @@ var _ = Describe("Templates", func() {
 		for key := range jinjaTest {
 			foo := jinjaTest[key]
 			It("renders correctly `"+key+"`", func() {
-				templated := evaluator.TemplateMessages(foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
+				templated := evaluator.TemplateMessages(schema.OpenAIRequest{}, foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
 				Expect(templated).To(Equal(foo["expected"]), templated)
 			})
 		}

gallery/harmony.yaml

@@ -5,23 +5,22 @@ config_file: |
   mmap: true
   backend: "llama-cpp"
   template:
-    chat_message: |
-      <|start|>{{ if .FunctionCall -}}functions.{{ .FunctionCall.Name }} to=assistant{{ else if eq .RoleName "assistant"}}<|channel|>final<|message|>{{else}}{{ .RoleName }}{{end}}<|message|>
-      {{ if .Content -}}
-      {{.Content }}
-      {{ end -}}
-      {{ if .FunctionCall -}}
-      {{toJson .FunctionCall}}
-      {{ end -}}<|end|>
-    function: |
-      <|im_start|>system
-      You are a function calling AI model. You are provided with functions to execute. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
-
+    chat_message: |-
+      <|start|>{{ if .FunctionCall -}}functions.{{ .FunctionCall.Name }} to=assistant{{ else if eq .RoleName "assistant"}}assistant<|channel|>final<|message|>{{else}}{{ .RoleName }}{{end}}<|message|>
+      {{- if .Content -}}
+      {{- .Content -}}
+      {{- end -}}
+      {{- if .FunctionCall -}}
+      {{- toJson .FunctionCall -}}
+      {{- end -}}<|end|>
+    function: |-
       <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
       Knowledge cutoff: 2024-06
-      Current date: {{ now | date }}
+      Current date: {{ now | date "Mon Jan 2 15:04:05 MST 2006" }}
+
+      Reasoning: {{if eq .ReasoningEffort ""}}medium{{else}}{{.ReasoningEffort}}{{end}}
 
-      Reasoning: medium
+      # {{with .Metadata}}{{ if ne .system_prompt "" }}{{ .system_prompt }}{{ end }}{{else}}You are a friendly and helpful assistant.{{ end }}<|end|>{{- .Input -}}<|start|>assistant
 
       # Tools
 
@@ -49,21 +48,15 @@ config_file: |
 
       # Instructions
 
-      <|end|>
-      {{.Input -}}
-      <|start|>assistant
-    chat: |
+      <|end|>{{.Input -}}<|start|>assistant
+    chat: |-
       <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
       Knowledge cutoff: 2024-06
-      Current date: {{ now | date }}
+      Current date: {{ now | date "Mon Jan 2 15:04:05 MST 2006" }}
 
-      Reasoning: medium
-
-      # Instructions
+      Reasoning: {{if eq .ReasoningEffort ""}}medium{{else}}{{.ReasoningEffort}}{{end}}
 
-      <|end|>
-      {{.Input -}}
-      <|im_start|>assistant
+      # {{with .Metadata}}{{ if ne .system_prompt "" }}{{ .system_prompt }}{{ end }}{{else}}You are a friendly and helpful assistant.{{ end }}<|end|>{{- .Input -}}<|start|>assistant
     completion: |
       {{.Input}}
   context_size: 8192

gallery/index.yaml

@@ -127,6 +127,30 @@
     - filename: gpt-oss-120b-mxfp4-00003-of-00003.gguf
       sha256: b326bfd8ac696c4b9a14e9e84d5529b2bb86847aea0e65443cbf075accba8b71
       uri: huggingface://ggml-org/gpt-oss-120b-GGUF/gpt-oss-120b-mxfp4-00003-of-00003.gguf
+- !!merge <<: *gptoss
+  name: "openai_gpt-oss-20b-neo"
+  icon: https://huggingface.co/DavidAU/Openai_gpt-oss-20b-NEO-GGUF/resolve/main/matrix1.gif
+  urls:
+    - https://huggingface.co/DavidAU/Openai_gpt-oss-20b-NEO-GGUF
+  description: |
+    These are NEO Imatrix GGUFs, NEO dataset by DavidAU.
+
+    NEO dataset improves overall performance, and is for all use cases.
+
+    Example output below (creative), using settings below.
+
+    Model also passed "hard" coding test too (6 experts); no issues (IQ4_NL).
+
+    (Forcing the model to create code with no dependencies and limits of coding short cuts, with multiple loops, and in real time with no blocking in a language that does not support it normally.)
+
+    Due to quanting issues with this model (which result in oddball quant sizes / mixtures), only TESTED quants will be uploaded (at the moment).
+  overrides:
+    parameters:
+      model: OpenAI-20B-NEO-MXFP4_MOE4.gguf
+  files:
+    - filename: OpenAI-20B-NEO-MXFP4_MOE4.gguf
+      sha256: 066c84a0844b1f1f4515e5c64095fe4c67e86d5eb70db4e368e283b1134d9c1e
+      uri: huggingface://DavidAU/Openai_gpt-oss-20b-NEO-GGUF/OpenAI-20B-NEO-MXFP4_MOE4.gguf
 - &afm
   name: "arcee-ai_afm-4.5b"
   url: "github:mudler/LocalAI/gallery/chatml.yaml@master"
@@ -2090,7 +2114,7 @@
     - gemma3
     - gemma-3
   overrides:
     #mmproj: gemma-3-27b-it-mmproj-f16.gguf
     parameters:
       model: gemma-3-27b-it-Q4_K_M.gguf
   files:
@@ -2108,7 +2132,7 @@
   description: |
     google/gemma-3-12b-it is an open-source, state-of-the-art, lightweight, multimodal model built from the same research and technology used to create the Gemini models. It is capable of handling text and image input and generating text output. It has a large context window of 128K tokens and supports over 140 languages. The 12B variant has been fine-tuned using the instruction-tuning approach. Gemma 3 models are suitable for a variety of text generation and image understanding tasks, including question answering, summarization, and reasoning. Their relatively small size makes them deployable in environments with limited resources such as laptops, desktops, or your own cloud infrastructure.
   overrides:
     #mmproj: gemma-3-12b-it-mmproj-f16.gguf
     parameters:
       model: gemma-3-12b-it-Q4_K_M.gguf
   files:
@@ -2126,7 +2150,7 @@
   description: |
     Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. Gemma 3 models are multimodal, handling text and image input and generating text output, with open weights for both pre-trained variants and instruction-tuned variants. Gemma 3 has a large, 128K context window, multilingual support in over 140 languages, and is available in more sizes than previous versions. Gemma 3 models are well-suited for a variety of text generation and image understanding tasks, including question answering, summarization, and reasoning. Their relatively small size makes it possible to deploy them in environments with limited resources such as laptops, desktops or your own cloud infrastructure, democratizing access to state of the art AI models and helping foster innovation for everyone. Gemma-3-4b-it is a 4 billion parameter model.
   overrides:
     #mmproj: gemma-3-4b-it-mmproj-f16.gguf
     parameters:
       model: gemma-3-4b-it-Q4_K_M.gguf
   files:
@@ -5592,7 +5616,7 @@
       sha256: 2756551de7d8ff7093c2c5eec1cd00f1868bc128433af53f5a8d434091d4eb5a
       uri: huggingface://Triangle104/Nano_Imp_1B-Q8_0-GGUF/nano_imp_1b-q8_0.gguf
 - &qwen25
   name: "qwen2.5-14b-instruct" ## Qwen2.5
   icon: https://avatars.githubusercontent.com/u/141221163
   url: "github:mudler/LocalAI/gallery/chatml.yaml@master"
   license: apache-2.0
@@ -7553,7 +7577,7 @@
       sha256: 0fec82625f74a9a340837de7af287b1d9042e5aeb70cda2621426db99958b0af
       uri: huggingface://bartowski/Chuluun-Qwen2.5-72B-v0.08-GGUF/Chuluun-Qwen2.5-72B-v0.08-Q4_K_M.gguf
 - &smollm
   url: "github:mudler/LocalAI/gallery/chatml.yaml@master" ## SmolLM
   name: "smollm-1.7b-instruct"
   icon: https://huggingface.co/datasets/HuggingFaceTB/images/resolve/main/banner_smol.png
   tags:

swagger/docs.go

@@ -1535,6 +1535,12 @@ const docTemplate = `{
                         "$ref": "#/definitions/schema.Message"
                     }
                 },
+                "metadata": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                },
                 "mode": {
                     "description": "Image (not supported by OpenAI)",
                     "type": "integer"
@@ -1567,6 +1573,9 @@ const docTemplate = `{
                 "quality": {
                     "type": "string"
                 },
+                "reasoning_effort": {
+                    "type": "string"
+                },
                 "ref_images": {
                     "description": "Reference images for models that support them (e.g., Flux Kontext)",
                     "type": "array",

swagger/swagger.json

@@ -1528,6 +1528,12 @@
                         "$ref": "#/definitions/schema.Message"
                     }
                 },
+                "metadata": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                },
                 "mode": {
                     "description": "Image (not supported by OpenAI)",
                     "type": "integer"
@@ -1560,6 +1566,9 @@
                 "quality": {
                     "type": "string"
                 },
+                "reasoning_effort": {
+                    "type": "string"
+                },
                 "ref_images": {
                     "description": "Reference images for models that support them (e.g., Flux Kontext)",
                     "type": "array",

swagger/swagger.yaml

@@ -491,6 +491,10 @@ definitions:
         items:
           $ref: '#/definitions/schema.Message'
         type: array
+      metadata:
+        additionalProperties:
+          type: string
+        type: object
       mode:
         description: Image (not supported by OpenAI)
         type: integer
@@ -514,6 +518,8 @@ definitions:
         description: Prompt is read only by completion/image API calls
       quality:
         type: string
+      reasoning_effort:
+        type: string
       ref_images:
         description: Reference images for models that support them (e.g., Flux Kontext)
         items: