Adding Ollama as a conversation component for local dev/running of LLMs (#3688)

Signed-off-by: Emmanuel Auffray <emmanuel.auffray@gmail.com>
Co-authored-by: Mike Nguyen <hey@mike.ee>
Co-authored-by: Yaron Schneider <schneider.yaron@live.com>

conversation/ollama/metadata.yaml

@@ -0,0 +1,23 @@
+# yaml-language-server: $schema=../../../component-metadata-schema.json
+schemaVersion: v1
+type: conversation
+name: ollama
+version: v1
+status: alpha
+title: "Ollama"
+urls:
+  - title: Reference
+    url: https://docs.dapr.io/reference/components-reference/supported-conversation/ollama/
+metadata:
+  - name: model
+    required: false
+    description: |
+      The Ollama LLM to use. Defaults to llama3.2:latest
+    type: string
+    example: 'llama3.2:latest'
+  - name: cacheTTL
+    required: false
+    description: |
+      A time-to-live value for a prompt cache to expire. Uses Golang durations
+    type: string
+    example: '10m'

conversation/ollama/ollama.go

@@ -0,0 +1,127 @@
+/*
+Copyright 2024 The Dapr Authors
+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 ollama
+
+import (
+	"context"
+	"reflect"
+
+	"github.com/dapr/components-contrib/conversation"
+	"github.com/dapr/components-contrib/metadata"
+	"github.com/dapr/kit/logger"
+	kmeta "github.com/dapr/kit/metadata"
+
+	"github.com/tmc/langchaingo/llms"
+	"github.com/tmc/langchaingo/llms/ollama"
+)
+
+type Ollama struct {
+	llm llms.Model
+
+	logger logger.Logger
+}
+
+func NewOllama(logger logger.Logger) conversation.Conversation {
+	o := &Ollama{
+		logger: logger,
+	}
+
+	return o
+}
+
+const defaultModel = "llama3.2:latest"
+
+func (o *Ollama) Init(ctx context.Context, meta conversation.Metadata) error {
+	md := conversation.LangchainMetadata{}
+	err := kmeta.DecodeMetadata(meta.Properties, &md)
+	if err != nil {
+		return err
+	}
+
+	model := defaultModel
+	if md.Model != "" {
+		model = md.Model
+	}
+
+	llm, err := ollama.New(
+		ollama.WithModel(model),
+	)
+	if err != nil {
+		return err
+	}
+
+	o.llm = llm
+
+	if md.CacheTTL != "" {
+		cachedModel, cacheErr := conversation.CacheModel(ctx, md.CacheTTL, o.llm)
+		if cacheErr != nil {
+			return cacheErr
+		}
+
+		o.llm = cachedModel
+	}
+	return nil
+}
+
+func (o *Ollama) GetComponentMetadata() (metadataInfo metadata.MetadataMap) {
+	metadataStruct := conversation.LangchainMetadata{}
+	metadata.GetMetadataInfoFromStructType(reflect.TypeOf(metadataStruct), &metadataInfo, metadata.ConversationType)
+	return
+}
+
+func (o *Ollama) Converse(ctx context.Context, r *conversation.ConversationRequest) (res *conversation.ConversationResponse, err error) {
+	messages := make([]llms.MessageContent, 0, len(r.Inputs))
+
+	for _, input := range r.Inputs {
+		role := conversation.ConvertLangchainRole(input.Role)
+
+		messages = append(messages, llms.MessageContent{
+			Role: role,
+			Parts: []llms.ContentPart{
+				llms.TextPart(input.Message),
+			},
+		})
+	}
+
+	opts := []llms.CallOption{}
+
+	if r.Temperature > 0 {
+		opts = append(opts, conversation.LangchainTemperature(r.Temperature))
+	}
+
+	resp, err := o.llm.GenerateContent(ctx, messages, opts...)
+	if err != nil {
+		return nil, err
+	}
+
+	outputs := make([]conversation.ConversationResult, 0, len(resp.Choices))
+
+	for i := range resp.Choices {
+		outputs = append(outputs, conversation.ConversationResult{
+			Result:     resp.Choices[i].Content,
+			Parameters: r.Parameters,
+		})
+	}
+
+	res = &conversation.ConversationResponse{
+		Outputs: outputs,
+	}
+
+	return res, nil
+}
+
+func (o *Ollama) Close() error {
+	return nil
+}