alex/LocalAGI
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Simplify

Browse Source 
Signed-off-by: mudler <mudler@localai.io>
This commit is contained in:
mudler
2025-04-12 18:38:20 +02:00
parent 0ac5c13c4d
commit 289a6ce4c8
4 changed files with 35 additions and 112 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
Makefile
View File

@@ -9,7 +9,7 @@ cleanup-tests:
docker compose down docker compose down
tests: prepare-tests tests: prepare-tests
LOCALAGI_MODEL="openthinker-7b" LOCALAI_API_URL="http://localhost:8081" LOCALAGI_API_URL="http://localhost:8080" $(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --fail-fast -v -r ./... LOCALAGI_MODEL="gemma-3-4b-it" LOCALAI_API_URL="http://localhost:8081" LOCALAGI_API_URL="http://localhost:8080" $(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --fail-fast -v -r ./...
run-nokb: run-nokb:
$(MAKE) run KBDISABLEINDEX=true $(MAKE) run KBDISABLEINDEX=true

6
README.md
View File

@@ -83,7 +83,7 @@ LocalAGI supports multiple hardware configurations through Docker Compose profil
- Supports text, multimodal, and image generation models - Supports text, multimodal, and image generation models
- Run with: `docker compose --profile nvidia up` - Run with: `docker compose --profile nvidia up`
- Default models: - Default models:
- Text: `openthinker-7b` - Text: `gemma-3-4b-it`
- Multimodal: `minicpm-v-2_6` - Multimodal: `minicpm-v-2_6`
- Image: `flux.1-dev` - Image: `flux.1-dev`
- Environment variables: - Environment variables:
@@ -99,7 +99,7 @@ LocalAGI supports multiple hardware configurations through Docker Compose profil
- Supports text, multimodal, and image generation models - Supports text, multimodal, and image generation models
- Run with: `docker compose --profile intel up` - Run with: `docker compose --profile intel up`
- Default models: - Default models:
- Text: `openthinker-7b` - Text: `gemma-3-4b-it`
- Multimodal: `minicpm-v-2_6` - Multimodal: `minicpm-v-2_6`
- Image: `sd-1.5-ggml` - Image: `sd-1.5-ggml`
- Environment variables: - Environment variables:
@@ -130,7 +130,7 @@ docker compose --profile intel up
``` ```
If no models are specified, it will use the defaults: If no models are specified, it will use the defaults:
- Text model: `openthinker-7b` - Text model: `gemma-3-4b-it`
- Multimodal model: `minicpm-v-2_6` - Multimodal model: `minicpm-v-2_6`
- Image model: `flux.1-dev` (NVIDIA) or `sd-1.5-ggml` (Intel) - Image model: `flux.1-dev` (NVIDIA) or `sd-1.5-ggml` (Intel)

127
core/agent/actions.go
View File

@@ -429,96 +429,28 @@ func (a *Agent) pickAction(ctx context.Context, templ string, messages []openai.
}, c...) }, c...)
} }
// thoughtPromptStringBuilder := strings.Builder{} thought, err := a.decision(ctx,
// thoughtPromptStringBuilder.WriteString("You have to pick an action based on the conversation and the prompt. Describe the full reasoning process for your choice. Here is a list of actions: ") c,
// for _, m := range a.availableActions() { types.Actions{action.NewReasoning()}.ToTools(),
// thoughtPromptStringBuilder.WriteString( action.NewReasoning().Definition().Name.String(), maxRetries)
// m.Definition().Name.String() + ": " + m.Definition().Description + "\n", if err != nil {
// ) return nil, nil, "", err
// } }
originalReasoning := ""
// thoughtPromptStringBuilder.WriteString("To not use any action, respond with 'none'") response := &action.ReasoningResponse{}
if thought.actionParams != nil {
//thoughtPromptStringBuilder.WriteString("\n\nConversation: " + Messages(c).RemoveIf(func(msg openai.ChatCompletionMessage) bool { if err := thought.actionParams.Unmarshal(response); err != nil {
// return msg.Role == "system" return nil, nil, "", err
//}).String()) }
originalReasoning = response.Reasoning
//thoughtPrompt := thoughtPromptStringBuilder.String() }
if thought.message != "" {
//thoughtConv := []openai.ChatCompletionMessage{} originalReasoning = thought.message
}
xlog.Debug("[pickAction] picking action", "messages", c) xlog.Debug("[pickAction] picking action", "messages", c)
thought, err := a.askLLM(ctx, // thought, err := a.askLLM(ctx,
c, // c,
maxRetries,
)
if err != nil {
return nil, nil, "", err
}
originalReasoning := thought.Content
xlog.Debug("[pickAction] original reasoning", "originalReasoning", originalReasoning)
// From the thought, get the action call
// Get all the available actions IDs
// by grammar, let's decide if we have achieved the goal
// 1. analyze response and check if goal is achieved
// Extract the goal first
params, err := a.decision(ctx,
append(
[]openai.ChatCompletionMessage{
{
Role: "system",
Content: "Your only task is to extract the goal from the following conversation",
}}, messages...),
types.Actions{action.NewGoal()}.ToTools(),
action.NewGoal().Definition().Name.String(), maxRetries)
if err != nil {
return nil, nil, "", fmt.Errorf("failed to get the action tool parameters: %v", err)
}
goalResponse := action.GoalResponse{}
err = params.actionParams.Unmarshal(&goalResponse)
if err != nil {
return nil, nil, "", err
}
if goalResponse.Goal == "" {
xlog.Debug("[pickAction] no goal found")
return nil, nil, "", nil
}
// Check if the goal was achieved
params, err = a.decision(ctx,
[]openai.ChatCompletionMessage{
{
Role: "system",
Content: "You have to understand if the goal is achieved or not from the following reasoning. The goal: " + goalResponse.Goal,
},
{
Role: "user",
Content: originalReasoning,
}},
types.Actions{action.NewGoal()}.ToTools(),
action.NewGoal().Definition().Name.String(), maxRetries)
if err != nil {
return nil, nil, "", fmt.Errorf("failed to get the action tool parameters: %v", err)
}
err = params.actionParams.Unmarshal(&goalResponse)
if err != nil {
return nil, nil, "", err
}
if goalResponse.Achieved {
xlog.Debug("[pickAction] goal achieved", "goal", goalResponse.Goal)
return nil, nil, "", nil
}
// if the goal is not achieved, pick an action
xlog.Debug("[pickAction] goal not achieved", "goal", goalResponse.Goal)
xlog.Debug("[pickAction] thought", "conv", c, "originalReasoning", originalReasoning)
actionsID := []string{"reply"} actionsID := []string{"reply"}
for _, m := range a.availableActions() { for _, m := range a.availableActions() {
@@ -533,20 +465,11 @@ func (a *Agent) pickAction(ctx context.Context, templ string, messages []openai.
// to avoid hallucinations // to avoid hallucinations
// Extract an action // Extract an action
params, err = a.decision(ctx, params, err := a.decision(ctx,
[]openai.ChatCompletionMessage{ append(c, openai.ChatCompletionMessage{
{ Role: "system",
Role: "system", Content: "Pick the relevant action given the following reasoning: " + originalReasoning,
Content: prompt, }),
},
{
Role: "system",
Content: "Extract an action to perform from the following reasoning: ",
},
{
Role: "user",
Content: originalReasoning,
}},
types.Actions{intentionsTools}.ToTools(), types.Actions{intentionsTools}.ToTools(),
intentionsTools.Definition().Name.String(), maxRetries) intentionsTools.Definition().Name.String(), maxRetries)
if err != nil { if err != nil {

12
docker-compose.yaml
View File

@@ -8,7 +8,7 @@ services:
image: localai/localai:master-ffmpeg-core image: localai/localai:master-ffmpeg-core
command: command:
# - gemma-3-12b-it # - gemma-3-12b-it
- ${MODEL_NAME:-openthinker-7b} - ${MODEL_NAME:-gemma-3-4b-it}
- granite-embedding-107m-multilingual - granite-embedding-107m-multilingual
healthcheck: healthcheck:
test: ["CMD", "curl", "-f", "http://localhost:8080/readyz"] test: ["CMD", "curl", "-f", "http://localhost:8080/readyz"]
@@ -39,7 +39,7 @@ services:
count: 1 count: 1
capabilities: [gpu] capabilities: [gpu]
command: command:
- ${MODEL_NAME:-openthinker-7b} - ${MODEL_NAME:-gemma-3-4b-it}
- ${MULTIMODAL_MODEL:-minicpm-v-2_6} - ${MULTIMODAL_MODEL:-minicpm-v-2_6}
- ${IMAGE_MODEL:-flux.1-dev} - ${IMAGE_MODEL:-flux.1-dev}
- granite-embedding-107m-multilingual - granite-embedding-107m-multilingual
@@ -57,7 +57,7 @@ services:
- /dev/dri/card1 - /dev/dri/card1
- /dev/dri/renderD129 - /dev/dri/renderD129
command: command:
- ${MODEL_NAME:-openthinker-7b} - ${MODEL_NAME:-gemma-3-4b-it}
- ${MULTIMODAL_MODEL:-minicpm-v-2_6} - ${MULTIMODAL_MODEL:-minicpm-v-2_6}
- ${IMAGE_MODEL:-sd-1.5-ggml} - ${IMAGE_MODEL:-sd-1.5-ggml}
- granite-embedding-107m-multilingual - granite-embedding-107m-multilingual
@@ -96,7 +96,7 @@ services:
- 8080:3000 - 8080:3000
#image: quay.io/mudler/localagi:master #image: quay.io/mudler/localagi:master
environment: environment:
- LOCALAGI_MODEL=${MODEL_NAME:-openthinker-7b} - LOCALAGI_MODEL=${MODEL_NAME:-gemma-3-4b-it}
- LOCALAGI_LLM_API_URL=http://localai:8080 - LOCALAGI_LLM_API_URL=http://localai:8080
#- LOCALAGI_LLM_API_KEY=sk-1234567890 #- LOCALAGI_LLM_API_KEY=sk-1234567890
- LOCALAGI_LOCALRAG_URL=http://localrecall:8080 - LOCALAGI_LOCALRAG_URL=http://localrecall:8080
@@ -113,7 +113,7 @@ services:
extends: extends:
service: localagi service: localagi
environment: environment:
- LOCALAGI_MODEL=${MODEL_NAME:-openthinker-7b} - LOCALAGI_MODEL=${MODEL_NAME:-gemma-3-4b-it}
- LOCALAGI_MULTIMODAL_MODEL=${MULTIMODAL_MODEL:-minicpm-v-2_6} - LOCALAGI_MULTIMODAL_MODEL=${MULTIMODAL_MODEL:-minicpm-v-2_6}
- LOCALAGI_IMAGE_MODEL=${IMAGE_MODEL:-flux.1-dev} - LOCALAGI_IMAGE_MODEL=${IMAGE_MODEL:-flux.1-dev}
- LOCALAGI_LLM_API_URL=http://localai:8080 - LOCALAGI_LLM_API_URL=http://localai:8080
@@ -127,7 +127,7 @@ services:
extends: extends:
service: localagi service: localagi
environment: environment:
- LOCALAGI_MODEL=${MODEL_NAME:-openthinker-7b} - LOCALAGI_MODEL=${MODEL_NAME:-gemma-3-4b-it}
- LOCALAGI_MULTIMODAL_MODEL=${MULTIMODAL_MODEL:-minicpm-v-2_6} - LOCALAGI_MULTIMODAL_MODEL=${MULTIMODAL_MODEL:-minicpm-v-2_6}
- LOCALAGI_IMAGE_MODEL=${IMAGE_MODEL:-sd-1.5-ggml} - LOCALAGI_IMAGE_MODEL=${IMAGE_MODEL:-sd-1.5-ggml}
- LOCALAGI_LLM_API_URL=http://localai:8080 - LOCALAGI_LLM_API_URL=http://localai:8080