mirror of
https://github.com/agentscope-ai/ReMe.git
synced 2026-10-10 03:30:56 +00:00
Merge branch 'dev' of http://gitlab.alibaba-inc.com/OpenRepo/ExperienceMaker into dev
This commit is contained in:
鸣山
commit
0b95a3bd70
16 changed files with 1428 additions and 584 deletions
4
.gitignore
vendored
4
.gitignore
vendored
|
|
@ -26,4 +26,6 @@ build/*
|
|||
*.egg-info/*
|
||||
cookbook/appworld/data/*
|
||||
cookbook/appworld/experiments/*
|
||||
cookbook/appworld/exp_result/*
|
||||
cookbook/appworld/exp_result/*
|
||||
file_vector_store/*
|
||||
cookbook/appworld/file_vector_store/*
|
||||
373
README.md
373
README.md
|
|
@ -15,15 +15,17 @@
|
|||
<strong>A comprehensive framework for AI agent experience generation and reuse</strong><br>
|
||||
<em>Empowering agents to learn from the past and excel in the future</em>
|
||||
</p>
|
||||
|
||||
---
|
||||
|
||||
## 📰 What's New
|
||||
- **[2025-08]** 🎉 ExperienceMaker v0.1.0 is now available on [PyPI](https://pypi.org/project/experiencemaker/)!
|
||||
- **[2025-07]** 📚 Complete documentation and quick start guides released
|
||||
- **[2025-07]** 🚀 Multi-backend vector store support (Elasticsearch & ChromaDB)
|
||||
- **[2025-06]** 🚀 Multi-backend vector store support (Elasticsearch & ChromaDB)
|
||||
|
||||
---
|
||||
|
||||
## 📰 What's Next
|
||||
## 🚀 What's Next
|
||||
- **Pre-built Experience Libraries**: Domain repositories (Finance/Coding/Education/Research) + community marketplace
|
||||
- **Rich Experience Formats**: Executable code/tool configs/pipeline templates/workflows
|
||||
- **Experience Validation**: Quality analysis + cross-task effectiveness + auto-refinement
|
||||
|
|
@ -41,7 +43,7 @@ By automatically extracting, storing, and intelligently reusing experiences from
|
|||
Traditional AI agents start from scratch with every new task, wasting valuable learning opportunities.
|
||||
ExperienceMaker changes this paradigm by:
|
||||
- **🧠 Learning from History**: Automatically extract actionable insights from both successful and failed attempts
|
||||
- **🔄 Intelligent Reuse**: Apply relevant past experiences to solve new, similar challenges more effectively
|
||||
- **🔄 Intelligent Reuse**: Apply relevant experiences to solve new, similar challenges more effectively
|
||||
- **📈 Continuous Improvement**: Build a growing knowledge base that makes agents progressively smarter
|
||||
|
||||
### ✨ Core Capabilities
|
||||
|
|
@ -50,12 +52,12 @@ ExperienceMaker changes this paradigm by:
|
|||
- **Success Pattern Recognition**: Identify what works and understand the underlying principles
|
||||
- **Failure Analysis**: Learn from mistakes to avoid repeating them in future tasks
|
||||
- **Comparative Insights**: Understand the critical differences between successful and failed approaches
|
||||
- **Multi-step Trajectory Processing**: Break down complex tasks into learnable, actionable segments
|
||||
- **Multistep Trajectory Processing**: Break down complex tasks into learnable, actionable segments
|
||||
|
||||
#### 🎯 **Smart Experience Retriever**
|
||||
- **Semantic Search**: Find relevant experiences using advanced embedding models and semantic understanding
|
||||
- **Context-Aware Ranking**: Prioritize the most applicable experiences for current task contexts
|
||||
- **Dynamic Rewriting**: Intelligently adapt past experiences to fit new situations and requirements
|
||||
- **Dynamic Rewriting**: Intelligently adapt experiences to fit new situations and requirements
|
||||
- **Multi-modal Support**: Handle various input types including query, messages
|
||||
|
||||
#### 🗄️ **Scalable Experience Management**
|
||||
|
|
@ -82,7 +84,12 @@ ExperienceMaker follows a modular, production-ready architecture designed for sc
|
|||
- **🗄️ Vector Store API**: Database management and workspace operations with full CRUD support
|
||||
|
||||
#### ⚙️ **Processing Pipeline**
|
||||
Our atomic operations can be seamlessly composed into powerful processing pipelines: custom1_op->custom2_op...
|
||||
|
||||
Our atomic operations can be seamlessly composed into powerful processing pipelines:
|
||||
|
||||
```
|
||||
custom1_op->custom2_op...
|
||||
```
|
||||
|
||||
#### 🔌 **Extensible Components**
|
||||
- **LLM Integration**: OpenAI-compatible APIs with flexible model switching and provider support
|
||||
|
|
@ -110,7 +117,7 @@ pip install .
|
|||
|
||||
## ⚙️ Environment Setup
|
||||
|
||||
Create a `.env` file in your project directory:
|
||||
Create a `.env` file in your project root directory:
|
||||
|
||||
```bash
|
||||
# Required: LLM API configuration
|
||||
|
|
@ -135,13 +142,16 @@ experiencemaker \
|
|||
embedding_model.default.model_name=text-embedding-v4 \
|
||||
vector_store.default.backend=local_file
|
||||
```
|
||||
💡 **Pro Tip**: Check out our [Advanced Guide](./doc/advanced_guide.md) for detailed configuration topics including custom pipelines, operation parameters, and advanced configuration methods.
|
||||
|
||||
💡 **Pro Tip**: Check out our [Configuration Guide](./doc/configuration_guide.md) for detailed configuration topics
|
||||
including custom pipelines, operation parameters, and advanced configuration methods.
|
||||
|
||||
The service will start on `http://localhost:8001`
|
||||
|
||||
### 🔍 Production Setup with Elasticsearch Backend
|
||||
```bash
|
||||
experiencemaker \
|
||||
http_service.port=8001 \
|
||||
llm.default.model_name=qwen3-32b \
|
||||
embedding_model.default.model_name=text-embedding-v4 \
|
||||
vector_store.default.backend=elasticsearch
|
||||
|
|
@ -158,79 +168,309 @@ curl -fsSL https://elastic.co/start-local | sh
|
|||
## 📝 Your First ExperienceMaker Script
|
||||
|
||||
Here's how to get started!
|
||||
- The `load_dotenv()` function loads environment variables from your `.env` file, or you can manually export them.
|
||||
- The `base_url` points to your ExperienceMaker service.
|
||||
- The `workspace_id` serves as your experience storage namespace. Experiences in different workspaces remain completely
|
||||
isolated and cannot access each other.
|
||||
Note the `workspace_id` serves as your experience storage namespace. Experiences in different workspaces remain completely isolated and cannot access each other.
|
||||
|
||||
### 📊 Call Summarizer Examples
|
||||
|
||||
Transform conversation trajectories into valuable experiences using batch summarization. Each trajectory contains:
|
||||
|
||||
- **Message**: Complete conversation history between user and agent
|
||||
- **Score**: Performance rating (0-1 scale, where 0=failure, 1=success)
|
||||
|
||||
The summarizer analyzes these trajectories to extract actionable insights and patterns for future interactions.
|
||||
|
||||
<details open>
|
||||
<summary><b>Python</b></summary>
|
||||
|
||||
```python
|
||||
import requests
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
base_url = "http://0.0.0.0:8001/"
|
||||
workspace_id = "test_workspace"
|
||||
```
|
||||
|
||||
### 📊 Call Summarizer Examples
|
||||
Batch summarize the trajectory list, where each trajectory consists of a message and a score.
|
||||
- The message is the conversation history.
|
||||
- The score represents the rating between 0 and 1, with 0 typically indicating failure and 1 indicating success.
|
||||
|
||||
```python
|
||||
response = requests.post(url=base_url + "summarizer", json={
|
||||
"workspace_id": workspace_id,
|
||||
response = requests.post(url="http://0.0.0.0:8001/summarizer", json={
|
||||
"workspace_id": "test_workspace",
|
||||
"traj_list": [
|
||||
{"messages": messages, "score": 1.0}
|
||||
{"messages": [{"role": "user", "content": "hello world"}], "score": 1.0}
|
||||
]
|
||||
})
|
||||
|
||||
response = response.json()
|
||||
experience_list = response["experience_list"]
|
||||
experience_list = response.json()["experience_list"]
|
||||
for experience in experience_list:
|
||||
print(experience)
|
||||
```
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>curl</b></summary>
|
||||
|
||||
```bash
|
||||
curl -X POST "http://0.0.0.0:8001/summarizer" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"workspace_id": "test_workspace",
|
||||
"traj_list": [
|
||||
{
|
||||
"messages": [{"role": "user", "content": "hello world"}],
|
||||
"score": 1.0
|
||||
}
|
||||
]
|
||||
}'
|
||||
```
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>Node.js</b></summary>
|
||||
|
||||
```javascript
|
||||
const fetch = require('node-fetch');
|
||||
// or: import fetch from 'node-fetch';
|
||||
|
||||
async function callSummarizer() {
|
||||
try {
|
||||
const response = await fetch('http://0.0.0.0:8001/summarizer', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
workspace_id: "test_workspace",
|
||||
traj_list: [
|
||||
{
|
||||
messages: [{ role: "user", content: "hello world" }],
|
||||
score: 1.0
|
||||
}
|
||||
]
|
||||
})
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
const experienceList = data.experience_list;
|
||||
|
||||
experienceList.forEach(experience => {
|
||||
console.log(experience);
|
||||
});
|
||||
} catch (error) {
|
||||
console.error('Error:', error);
|
||||
}
|
||||
}
|
||||
|
||||
callSummarizer();
|
||||
```
|
||||
</details>
|
||||
|
||||
### 🔍 Call Retriever Examples
|
||||
Retrieve the top_k={top_k} experiences related to {query} in workspace=test_workspace, and finally accept the assembled context.
|
||||
Alternatively, you can also accept the raw experience_list parameter and assemble the context yourself.
|
||||
|
||||
Intelligently search and retrieve the most relevant experiences from your workspace to enhance decision-making. The retriever:
|
||||
|
||||
- **Finds** the top-k most similar experiences based on semantic similarity to your query
|
||||
- **Returns** pre-assembled context ready for immediate use, or raw experience data for custom processing
|
||||
- **Leverages** your workspace's accumulated knowledge to provide contextually relevant insights
|
||||
|
||||
<details open>
|
||||
<summary><b>Python</b></summary>
|
||||
|
||||
```python
|
||||
response = requests.post(url=base_url + "retriever", json={
|
||||
"workspace_id": workspace_id,
|
||||
"query": query,
|
||||
import requests
|
||||
|
||||
response = requests.post(url="http://0.0.0.0:8001/retriever", json={
|
||||
"workspace_id": "test_workspace",
|
||||
"query": "what is the meaning of life?",
|
||||
"top_k": 1,
|
||||
})
|
||||
|
||||
response = response.json()
|
||||
experience_merged: str = response["experience_merged"]
|
||||
experience_merged: str = response.json()["experience_merged"]
|
||||
print(f"experience_merged={experience_merged}")
|
||||
```
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>curl</b></summary>
|
||||
|
||||
```bash
|
||||
curl -X POST "http://0.0.0.0:8001/retriever" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"workspace_id": "test_workspace",
|
||||
"query": "what is the meaning of life?",
|
||||
"top_k": 1
|
||||
}'
|
||||
```
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>Node.js</b></summary>
|
||||
|
||||
```javascript
|
||||
const fetch = require('node-fetch');
|
||||
// or: import fetch from 'node-fetch';
|
||||
|
||||
async function callRetriever() {
|
||||
try {
|
||||
const response = await fetch('http://0.0.0.0:8001/retriever', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
workspace_id: "test_workspace",
|
||||
query: "what is the meaning of life?",
|
||||
top_k: 1
|
||||
})
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
const experienceMerged = data.experience_merged;
|
||||
|
||||
console.log(`experience_merged=${experienceMerged}`);
|
||||
} catch (error) {
|
||||
console.error('Error:', error);
|
||||
}
|
||||
}
|
||||
|
||||
callRetriever();
|
||||
```
|
||||
</details>
|
||||
|
||||
### 💾 Dump Experiences From Vector Store
|
||||
Dump the experience with workspace_id from the vector store into the {path}/{workspace_id}.jsonl file.
|
||||
|
||||
Export and backup your valuable experience data for archival, analysis, or migration purposes. This operation:
|
||||
|
||||
- **Extracts** all experiences from the specified workspace in the vector store
|
||||
- **Saves** them to a structured JSONL file at `{path}/{workspace_id}.jsonl`
|
||||
- **Preserves** complete experience metadata and embeddings for future restoration
|
||||
|
||||
<details open>
|
||||
<summary><b>Python</b></summary>
|
||||
|
||||
```python
|
||||
response = requests.post(url=base_url + "vector_store", json={
|
||||
"workspace_id": workspace_id,
|
||||
import requests
|
||||
|
||||
response = requests.post(url="http://0.0.0.0:8001/vector_store", json={
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "dump",
|
||||
"path": "./",
|
||||
})
|
||||
print(response.json())
|
||||
```
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>curl</b></summary>
|
||||
|
||||
```bash
|
||||
curl -X POST "http://0.0.0.0:8001/vector_store" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "dump",
|
||||
"path": "./"
|
||||
}'
|
||||
```
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>Node.js</b></summary>
|
||||
|
||||
```javascript
|
||||
const fetch = require('node-fetch');
|
||||
// or: import fetch from 'node-fetch';
|
||||
|
||||
async function dumpExperiences() {
|
||||
try {
|
||||
const response = await fetch('http://0.0.0.0:8001/vector_store', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
workspace_id: "test_workspace",
|
||||
action: "dump",
|
||||
path: "./"
|
||||
})
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
console.log(data);
|
||||
} catch (error) {
|
||||
console.error('Error:', error);
|
||||
}
|
||||
}
|
||||
|
||||
dumpExperiences();
|
||||
```
|
||||
</details>
|
||||
|
||||
### 📥 Load Experiences To Vector Store
|
||||
Load the {path}/{workspace_id}.jsonl file into the vector store, workspace_id={workspace_id}.
|
||||
|
||||
Import and restore previously exported experience data to populate your workspace with existing knowledge. This operation:
|
||||
|
||||
- **Reads** experience data from the JSONL file located at `{path}/{workspace_id}.jsonl`
|
||||
- **Reconstructs** the vector embeddings and indexes them in the specified workspace
|
||||
- **Enables** immediate access to imported experiences for retrieval and decision-making
|
||||
|
||||
<details open>
|
||||
<summary><b>Python</b></summary>
|
||||
|
||||
```python
|
||||
response = requests.post(url=base_url + "vector_store", json={
|
||||
"workspace_id": workspace_id,
|
||||
import requests
|
||||
|
||||
response = requests.post(url="http://0.0.0.0:8001/vector_store", json={
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "load",
|
||||
"path": "./",
|
||||
})
|
||||
|
||||
print(response.json())
|
||||
```
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>curl</b></summary>
|
||||
|
||||
```bash
|
||||
curl -X POST "http://0.0.0.0:8001/vector_store" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "load",
|
||||
"path": "./"
|
||||
}'
|
||||
```
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>Node.js</b></summary>
|
||||
|
||||
```javascript
|
||||
const fetch = require('node-fetch');
|
||||
// or: import fetch from 'node-fetch';
|
||||
|
||||
async function loadExperiences() {
|
||||
try {
|
||||
const response = await fetch('http://0.0.0.0:8001/vector_store', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
workspace_id: "test_workspace",
|
||||
action: "load",
|
||||
path: "./"
|
||||
})
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
console.log(data);
|
||||
} catch (error) {
|
||||
console.error('Error:', error);
|
||||
}
|
||||
}
|
||||
|
||||
loadExperiences();
|
||||
```
|
||||
</details>
|
||||
|
||||
💡 **Need More Advanced Operations?** For additional workspace management features(e.g. delete_workspace,
|
||||
copy_workspace), advanced configuration options, and troubleshooting guidance, check out our
|
||||
comprehensive [Quick Start Guide](./cookbook/simple_demo/quick_start.md).
|
||||
|
||||
🎭 **Want to See It in Action?** We've prepared a [simple react agent](./cookbook/simple_demo/simple_demo.py) that demonstrates how to enhance agent capabilities by integrating summarizer and retriever components, achieving significantly better performance.
|
||||
|
||||
|
|
@ -259,15 +499,60 @@ Coming Soon! Stay tuned for comprehensive evaluation results.
|
|||
|
||||
## 🏪 Ready-made Experience Store
|
||||
|
||||
Pre-built experience collections for common domains and use cases are coming soon. This will include ready-to-use experiences for web automation, data processing, API interactions, and more.
|
||||
ExperienceMaker provides pre-built experience libraries to jumpstart your agent's capabilities.
|
||||
You can directly load these curated experiences into your workspace and start benefiting from accumulated knowledge
|
||||
immediately.
|
||||
|
||||
### 📦 Available Experience Libraries
|
||||
|
||||
- **`appworld_v1.jsonl`**: Comprehensive experiences from Appworld agent interactions, covering complex task planning
|
||||
and execution patterns
|
||||
- **`bfcl_v1.jsonl`**: Function calling experiences from Berkeley Function-Calling Leaderboard tasks
|
||||
|
||||
### 🚀 Quick Start with Pre-built Experiences
|
||||
|
||||
Here's how to load and use the Appworld experience library:
|
||||
|
||||
#### Step 1: Load Pre-built Experiences
|
||||
|
||||
```python
|
||||
import requests
|
||||
|
||||
# Load Appworld experiences into your workspace
|
||||
response = requests.post(url="http://0.0.0.0:8001/vector_store", json={
|
||||
"workspace_id": "appworld_v1",
|
||||
"action": "load",
|
||||
"path": "./experience_library/",
|
||||
})
|
||||
|
||||
print(f"loading result result={response.json()}")
|
||||
```
|
||||
|
||||
#### Step 2: Retrieve Relevant Experiences
|
||||
|
||||
Now you can query the loaded experiences to get contextual guidance for your tasks:
|
||||
|
||||
```python
|
||||
import requests
|
||||
|
||||
# Query for app interaction experiences
|
||||
response = requests.post(url="http://0.0.0.0:8001/retriever", json={
|
||||
"workspace_id": "appworld_v1",
|
||||
"query": "How to navigate to settings and update user profile information?",
|
||||
"top_k": 1,
|
||||
})
|
||||
|
||||
experience_merged = response.json()["experience_merged"]
|
||||
print(f"Retrieved experiences: {experience_merged}")
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 📚 Additional Resources
|
||||
|
||||
- **[Quick Start](./cookbook/simple_demo/quick_start.md)**: This guide will help you get started with ExperienceMaker quickly using practical examples.
|
||||
- **[Vector Store Setup](./doc/vector_store_setup.md)**: Complete production deployment guide
|
||||
- **[Configuration Guide](./doc/configuration_guide.md)**: Describes all available command-line parameters for ExperienceMaker Service
|
||||
- **[Advanced Guide](./doc/advanced_guide.md)**: Custom pipelines, operation parameters, and advanced configuration methods
|
||||
- **[Operations Documentation](./doc/operations_documentation.md)**: Comprehensive operations configuration reference
|
||||
- **[Example Collection](./cookbook)**: Practical examples and use cases
|
||||
- **[Future RoadMap](./doc/future_roadmap.md)**: Our vision and upcoming features
|
||||
|
|
|
|||
559
cookbook/simple_demo/quick_start.md
Normal file
559
cookbook/simple_demo/quick_start.md
Normal file
|
|
@ -0,0 +1,559 @@
|
|||
# ExperienceMaker Quick Start Guide
|
||||
This guide will help you get started with ExperienceMaker quickly using practical examples.
|
||||
|
||||
## 🚀 What You'll Learn
|
||||
- How to set up ExperienceMaker service
|
||||
- Run an agent and generate experiences
|
||||
- Retrieve and apply experiences to new tasks
|
||||
- Build experience-enhanced agents
|
||||
|
||||
## 📋 Prerequisites
|
||||
- Python 3.12+
|
||||
- LLM API access (OpenAI or compatible)
|
||||
- Embedding model API access
|
||||
|
||||
## 🛠️ Installation
|
||||
|
||||
### Option 1: Install from PyPI (Recommended)
|
||||
|
||||
```bash
|
||||
pip install experiencemaker
|
||||
```
|
||||
|
||||
### Option 2: Install from Source
|
||||
|
||||
```bash
|
||||
git clone https://github.com/modelscope/ExperienceMaker.git
|
||||
cd ExperienceMaker
|
||||
pip install .
|
||||
```
|
||||
|
||||
## ⚙️ Environment Setup
|
||||
Create a `.env` file in your project directory:
|
||||
|
||||
```bash
|
||||
# Required: LLM API configuration
|
||||
LLM_API_KEY="sk-xxx"
|
||||
LLM_BASE_URL="https://xxx.com/v1"
|
||||
|
||||
# Required: Embedding model configuration
|
||||
EMBEDDING_MODEL_API_KEY="sk-xxx"
|
||||
EMBEDDING_MODEL_BASE_URL="https://xxx.com/v1"
|
||||
|
||||
# Optional: Elasticsearch configuration (if using Elasticsearch backend)
|
||||
|
||||
```
|
||||
|
||||
## 🚀 Start the Service
|
||||
For testing, use the `local_file` backend:
|
||||
```bash
|
||||
experiencemaker \
|
||||
http_service.port=8001 \
|
||||
llm.default.model_name=qwen3-32b \
|
||||
embedding_model.default.model_name=text-embedding-v4 \
|
||||
vector_store.default.backend=local_file
|
||||
```
|
||||
The service will start on `http://localhost:8001`
|
||||
|
||||
### Elasticsearch Backend
|
||||
```bash
|
||||
experiencemaker \
|
||||
http_service.port=8001 \
|
||||
llm.default.model_name=qwen3-32b \
|
||||
embedding_model.default.model_name=text-embedding-v4 \
|
||||
vector_store.default.backend=elasticsearch
|
||||
```
|
||||
|
||||
**Setup Elasticsearch:**
|
||||
```bash
|
||||
export ES_HOSTS="http://localhost:9200"
|
||||
# Quick setup using Elastic's official script
|
||||
curl -fsSL https://elastic.co/start-local | sh
|
||||
```
|
||||
|
||||
📖 **Need Help?** Refer to [Vector Store Setup](../../doc/vector_store_setup.md) for comprehensive deployment guidance.
|
||||
|
||||
## 📝 Your First ExperienceMaker Script
|
||||
|
||||
Here's how to get started!
|
||||
Note the `workspace_id` serves as your experience storage namespace. Experiences in different workspaces remain
|
||||
completely isolated and cannot access each other.
|
||||
|
||||
### 📊 Call Summarizer Examples
|
||||
|
||||
Transform conversation trajectories into valuable experiences using batch summarization. Each trajectory contains:
|
||||
|
||||
- **Message**: Complete conversation history between user and agent
|
||||
- **Score**: Performance rating (0-1 scale, where 0=failure, 1=success)
|
||||
|
||||
The summarizer analyzes these trajectories to extract actionable insights and patterns for future interactions.
|
||||
|
||||
<details open>
|
||||
<summary><b>Python</b></summary>
|
||||
|
||||
```python
|
||||
import requests
|
||||
|
||||
response = requests.post(url="http://0.0.0.0:8001/summarizer", json={
|
||||
"workspace_id": "test_workspace",
|
||||
"traj_list": [
|
||||
{"messages": [{"role": "user", "content": "hello world"}], "score": 1.0}
|
||||
]
|
||||
})
|
||||
|
||||
experience_list = response.json()["experience_list"]
|
||||
for experience in experience_list:
|
||||
print(experience)
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>curl</b></summary>
|
||||
|
||||
```bash
|
||||
curl -X POST "http://0.0.0.0:8001/summarizer" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"workspace_id": "test_workspace",
|
||||
"traj_list": [
|
||||
{
|
||||
"messages": [{"role": "user", "content": "hello world"}],
|
||||
"score": 1.0
|
||||
}
|
||||
]
|
||||
}'
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>Node.js</b></summary>
|
||||
|
||||
```javascript
|
||||
const fetch = require('node-fetch');
|
||||
// or: import fetch from 'node-fetch';
|
||||
|
||||
async function callSummarizer() {
|
||||
try {
|
||||
const response = await fetch('http://0.0.0.0:8001/summarizer', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
workspace_id: "test_workspace",
|
||||
traj_list: [
|
||||
{
|
||||
messages: [{ role: "user", content: "hello world" }],
|
||||
score: 1.0
|
||||
}
|
||||
]
|
||||
})
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
const experienceList = data.experience_list;
|
||||
|
||||
experienceList.forEach(experience => {
|
||||
console.log(experience);
|
||||
});
|
||||
} catch (error) {
|
||||
console.error('Error:', error);
|
||||
}
|
||||
}
|
||||
|
||||
callSummarizer();
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
### 🔍 Call Retriever Examples
|
||||
|
||||
Intelligently search and retrieve the most relevant experiences from your workspace to enhance decision-making. The retriever:
|
||||
|
||||
- **Finds** the top-k most similar experiences based on semantic similarity to your query
|
||||
- **Returns** pre-assembled context ready for immediate use, or raw experience data for custom processing
|
||||
- **Leverages** your workspace's accumulated knowledge to provide contextually relevant insights
|
||||
|
||||
<details open>
|
||||
<summary><b>Python</b></summary>
|
||||
|
||||
```python
|
||||
import requests
|
||||
|
||||
response = requests.post(url="http://0.0.0.0:8001/retriever", json={
|
||||
"workspace_id": "test_workspace",
|
||||
"query": "what is the meaning of life?",
|
||||
"top_k": 1,
|
||||
})
|
||||
|
||||
experience_merged: str = response.json()["experience_merged"]
|
||||
print(f"experience_merged={experience_merged}")
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>curl</b></summary>
|
||||
|
||||
```bash
|
||||
curl -X POST "http://0.0.0.0:8001/retriever" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"workspace_id": "test_workspace",
|
||||
"query": "what is the meaning of life?",
|
||||
"top_k": 1
|
||||
}'
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>Node.js</b></summary>
|
||||
|
||||
```javascript
|
||||
const fetch = require('node-fetch');
|
||||
// or: import fetch from 'node-fetch';
|
||||
|
||||
async function callRetriever() {
|
||||
try {
|
||||
const response = await fetch('http://0.0.0.0:8001/retriever', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
workspace_id: "test_workspace",
|
||||
query: "what is the meaning of life?",
|
||||
top_k: 1
|
||||
})
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
const experienceMerged = data.experience_merged;
|
||||
|
||||
console.log(`experience_merged=${experienceMerged}`);
|
||||
} catch (error) {
|
||||
console.error('Error:', error);
|
||||
}
|
||||
}
|
||||
|
||||
callRetriever();
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
### 💾 Dump Experiences From Vector Store
|
||||
|
||||
Export and backup your valuable experience data for archival, analysis, or migration purposes. This operation:
|
||||
|
||||
- **Extracts** all experiences from the specified workspace in the vector store
|
||||
- **Saves** them to a structured JSONL file at `{path}/{workspace_id}.jsonl`
|
||||
- **Preserves** complete experience metadata and embeddings for future restoration
|
||||
|
||||
<details open>
|
||||
<summary><b>Python</b></summary>
|
||||
|
||||
```python
|
||||
import requests
|
||||
|
||||
response = requests.post(url="http://0.0.0.0:8001/vector_store", json={
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "dump",
|
||||
"path": "./",
|
||||
})
|
||||
print(response.json())
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>curl</b></summary>
|
||||
|
||||
```bash
|
||||
curl -X POST "http://0.0.0.0:8001/vector_store" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "dump",
|
||||
"path": "./"
|
||||
}'
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>Node.js</b></summary>
|
||||
|
||||
```javascript
|
||||
const fetch = require('node-fetch');
|
||||
// or: import fetch from 'node-fetch';
|
||||
|
||||
async function dumpExperiences() {
|
||||
try {
|
||||
const response = await fetch('http://0.0.0.0:8001/vector_store', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
workspace_id: "test_workspace",
|
||||
action: "dump",
|
||||
path: "./"
|
||||
})
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
console.log(data);
|
||||
} catch (error) {
|
||||
console.error('Error:', error);
|
||||
}
|
||||
}
|
||||
|
||||
dumpExperiences();
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
### 📥 Load Experiences To Vector Store
|
||||
|
||||
Import and restore previously exported experience data to populate your workspace with existing knowledge. This operation:
|
||||
|
||||
- **Reads** experience data from the JSONL file located at `{path}/{workspace_id}.jsonl`
|
||||
- **Reconstructs** the vector embeddings and indexes them in the specified workspace
|
||||
- **Enables** immediate access to imported experiences for retrieval and decision-making
|
||||
|
||||
<details open>
|
||||
<summary><b>Python</b></summary>
|
||||
|
||||
```python
|
||||
import requests
|
||||
|
||||
response = requests.post(url="http://0.0.0.0:8001/vector_store", json={
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "load",
|
||||
"path": "./",
|
||||
})
|
||||
|
||||
print(response.json())
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>curl</b></summary>
|
||||
|
||||
```bash
|
||||
curl -X POST "http://0.0.0.0:8001/vector_store" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "load",
|
||||
"path": "./"
|
||||
}'
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>Node.js</b></summary>
|
||||
|
||||
```javascript
|
||||
const fetch = require('node-fetch');
|
||||
// or: import fetch from 'node-fetch';
|
||||
|
||||
async function loadExperiences() {
|
||||
try {
|
||||
const response = await fetch('http://0.0.0.0:8001/vector_store', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
workspace_id: "test_workspace",
|
||||
action: "load",
|
||||
path: "./"
|
||||
})
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
console.log(data);
|
||||
} catch (error) {
|
||||
console.error('Error:', error);
|
||||
}
|
||||
}
|
||||
|
||||
loadExperiences();
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
|
||||
### 🗑️ Delete Workspace
|
||||
|
||||
Permanently remove a workspace and all its associated experience data when it's no longer needed. This operation:
|
||||
|
||||
- **Removes** all experiences, embeddings, and metadata from the specified workspace
|
||||
- **Frees up** storage space and computational resources
|
||||
- **Cannot be undone** - ensure you've backed up important data before deletion
|
||||
|
||||
<details open>
|
||||
<summary><b>Python</b></summary>
|
||||
|
||||
```python
|
||||
import requests
|
||||
|
||||
response = requests.post(url="http://0.0.0.0:8001/vector_store", json={
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "delete"
|
||||
})
|
||||
|
||||
print(response.json())
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>curl</b></summary>
|
||||
|
||||
```bash
|
||||
curl -X POST "http://0.0.0.0:8001/vector_store" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "delete"
|
||||
}'
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>Node.js</b></summary>
|
||||
|
||||
```javascript
|
||||
const fetch = require('node-fetch');
|
||||
// or: import fetch from 'node-fetch';
|
||||
|
||||
async function deleteWorkspace() {
|
||||
try {
|
||||
const response = await fetch('http://0.0.0.0:8001/vector_store', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
workspace_id: "test_workspace",
|
||||
action: "delete"
|
||||
})
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
console.log(data);
|
||||
} catch (error) {
|
||||
console.error('Error:', error);
|
||||
}
|
||||
}
|
||||
|
||||
deleteWorkspace();
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
### 📋 Copy Workspace
|
||||
|
||||
Duplicate an existing workspace to create a new one with identical experience data, perfect for experimentation or branching. This operation:
|
||||
|
||||
- **Clones** all experiences and embeddings from the source workspace
|
||||
- **Creates** a new independent workspace with the copied data
|
||||
- **Preserves** original workspace while enabling safe testing and modifications in the copy
|
||||
|
||||
<details open>
|
||||
<summary><b>Python</b></summary>
|
||||
|
||||
```python
|
||||
import requests
|
||||
|
||||
response = requests.post(url="http://0.0.0.0:8001/vector_store", json={
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "copy",
|
||||
"src_workspace_id": "src_workspace"
|
||||
})
|
||||
|
||||
print(response.json())
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>curl</b></summary>
|
||||
|
||||
```bash
|
||||
curl -X POST "http://0.0.0.0:8001/vector_store" \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"workspace_id": "test_workspace",
|
||||
"action": "copy",
|
||||
"src_workspace_id": "src_workspace"
|
||||
}'
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
<details>
|
||||
<summary><b>Node.js</b></summary>
|
||||
|
||||
```javascript
|
||||
const fetch = require('node-fetch');
|
||||
// or: import fetch from 'node-fetch';
|
||||
|
||||
async function copyWorkspace() {
|
||||
try {
|
||||
const response = await fetch('http://0.0.0.0:8001/vector_store', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
workspace_id: "test_workspace",
|
||||
action: "copy",
|
||||
src_workspace_id: "src_workspace"
|
||||
})
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
console.log(data);
|
||||
} catch (error) {
|
||||
console.error('Error:', error);
|
||||
}
|
||||
}
|
||||
|
||||
copyWorkspace();
|
||||
```
|
||||
|
||||
</details>
|
||||
|
||||
🎭 **Want to See It in Action?** We've prepared a [simple react agent](../../cookbook/simple_demo/simple_demo.py) that
|
||||
demonstrates how to enhance agent capabilities by integrating summarizer and retriever components, achieving
|
||||
significantly better performance.
|
||||
|
||||
## 🐛 Common Issues
|
||||
|
||||
### Service Won't Start
|
||||
- Check if port 8001 is available
|
||||
- Verify your API keys in `.env` file
|
||||
- Ensure Python version is 3.12+
|
||||
|
||||
### No Experiences Retrieved
|
||||
- Make sure you've run the summarizer first
|
||||
- Check if workspace_id matches between operations
|
||||
- Verify vector store backend is properly configured
|
||||
|
||||
### API Connection Errors
|
||||
- Confirm LLM_BASE_URL and API keys are correct
|
||||
- Test API access independently
|
||||
- Check network connectivity
|
||||
|
||||
---
|
||||
|
||||
🎯 **You're all set!** You now have a working ExperienceMaker setup that can learn from interactions and improve over time.
|
||||
|
|
@ -64,8 +64,8 @@ def run_retriever(query: str):
|
|||
|
||||
|
||||
def run_agent_with_experience(query_first: str, query_second: str, dump_experience: bool = True):
|
||||
# messages = run_agent(query=query_second)
|
||||
# run_summary(messages, dump_experience)
|
||||
messages = run_agent(query=query_second)
|
||||
run_summary(messages, dump_experience)
|
||||
experience_merged = run_retriever(query_first)
|
||||
messages = run_agent(query=f"{experience_merged}\n\nUser Question:\n{query_first}")
|
||||
return messages
|
||||
|
|
@ -103,7 +103,7 @@ if __name__ == "__main__":
|
|||
query1 = "Analyze Xiaomi Corporation"
|
||||
query2 = "Analyze the company Tesla."
|
||||
|
||||
# run_agent(query=query1, dump_messages=True)
|
||||
# run_agent_with_experience(query_first=query1, query_second=query2)
|
||||
# dump_experience()
|
||||
run_agent(query=query1, dump_messages=True)
|
||||
run_agent_with_experience(query_first=query1, query_second=query2)
|
||||
dump_experience()
|
||||
load_experience()
|
||||
|
|
|
|||
|
|
@ -1,310 +0,0 @@
|
|||
# ExperienceMaker Advanced Configuration Guide
|
||||
|
||||
This guide covers advanced configuration topics including custom pipelines, operation parameters, and configuration
|
||||
methods.
|
||||
|
||||
## 🏗️ Configuration Architecture
|
||||
|
||||
ExperienceMaker uses a layered configuration system with the following priority order:
|
||||
|
||||
1. **Default Configuration** (lowest priority)
|
||||
2. **YAML Configuration File**
|
||||
3. **Command Line Arguments** (highest priority)
|
||||
|
||||
## 📁 Configuration Structure
|
||||
|
||||
```yaml
|
||||
# Service Configuration
|
||||
http_service:
|
||||
host: "0.0.0.0"
|
||||
port: 8001
|
||||
timeout_keep_alive: 600
|
||||
limit_concurrency: 64
|
||||
|
||||
# Pipeline Definitions
|
||||
api:
|
||||
retriever: recall_experience_op->rerank_experience_op->rewrite_experience_op
|
||||
summarizer: trajectory_preprocess_op->[success_extraction_op|failure_extraction_op]->experience_validation_op
|
||||
vector_store: vector_store_action_op
|
||||
|
||||
# Operation Configurations
|
||||
op:
|
||||
operation_name:
|
||||
backend: operation_backend
|
||||
llm: default # Optional: reference to LLM config
|
||||
embedding_model: default # Optional: reference to embedding config
|
||||
vector_store: default # Optional: reference to vector store config
|
||||
params: # Operation-specific parameters
|
||||
param1: value1
|
||||
param2: value2
|
||||
|
||||
# Resource Configurations
|
||||
llm:
|
||||
default:
|
||||
backend: openai_compatible
|
||||
model_name: qwen3-32b
|
||||
params:
|
||||
temperature: 0.6
|
||||
|
||||
embedding_model:
|
||||
default:
|
||||
backend: openai_compatible
|
||||
model_name: text-embedding-v4
|
||||
params:
|
||||
dimensions: 1024
|
||||
|
||||
vector_store:
|
||||
default:
|
||||
backend: local_file
|
||||
embedding_model: default
|
||||
```
|
||||
|
||||
## 🔧 Pipeline Configuration
|
||||
|
||||
### Pipeline Syntax
|
||||
|
||||
Pipeline configurations use a special syntax to define operation flows:
|
||||
|
||||
- `->`: Sequential execution
|
||||
- `[]`: Parallel execution group
|
||||
- `|`: Alternative operations within parallel group
|
||||
|
||||
### Examples
|
||||
|
||||
```yaml
|
||||
# Sequential pipeline
|
||||
api:
|
||||
retriever: op1->op2->op3
|
||||
|
||||
# Parallel execution
|
||||
api:
|
||||
summarizer: op1->[op2|op3|op4]->op5
|
||||
|
||||
# Complex pipeline with nested parallel operations
|
||||
api:
|
||||
retriever: preprocess_op->[recall_op->rerank_op|backup_op]->merge_op
|
||||
```
|
||||
|
||||
## ⚙️ Custom Operation Parameters
|
||||
|
||||
### Operation Configuration Structure
|
||||
|
||||
```yaml
|
||||
op:
|
||||
custom_operation:
|
||||
backend: custom_backend_name
|
||||
llm: default # Reference to LLM configuration
|
||||
vector_store: default # Reference to vector store
|
||||
params: # Custom parameters for this operation
|
||||
retrieve_top_k: 15 # Number of top results to retrieve
|
||||
similarity_threshold: 0.8 # Similarity threshold for filtering
|
||||
enable_rerank: true # Enable reranking functionality
|
||||
custom_param: "custom_value" # Any custom parameter
|
||||
```
|
||||
|
||||
### Common Operation Parameters
|
||||
|
||||
**Retrieval Operations:**
|
||||
|
||||
```yaml
|
||||
recall_experience_op:
|
||||
params:
|
||||
retrieve_top_k: 15
|
||||
similarity_threshold: 0.5
|
||||
|
||||
rerank_experience_op:
|
||||
params:
|
||||
enable_llm_rerank: true
|
||||
enable_score_filter: false
|
||||
top_k: 5
|
||||
```
|
||||
|
||||
**Extraction Operations:**
|
||||
|
||||
```yaml
|
||||
success_extraction_op:
|
||||
params:
|
||||
extraction_mode: "detailed"
|
||||
include_context: true
|
||||
|
||||
experience_validation_op:
|
||||
params:
|
||||
validation_threshold: 0.5
|
||||
strict_mode: false
|
||||
```
|
||||
|
||||
## 🚀 Configuration Methods
|
||||
|
||||
### Method 1: Custom Configuration File
|
||||
|
||||
**Step 1:** Create your configuration file
|
||||
|
||||
```yaml
|
||||
# my_custom_config.yaml
|
||||
api:
|
||||
retriever: custom_recall_op->custom_rerank_op
|
||||
|
||||
op:
|
||||
custom_recall_op:
|
||||
backend: recall_experience_op
|
||||
params:
|
||||
retrieve_top_k: 20
|
||||
similarity_threshold: 0.7
|
||||
|
||||
llm:
|
||||
default:
|
||||
model_name: gpt-4
|
||||
params:
|
||||
temperature: 0.3
|
||||
```
|
||||
|
||||
**Step 2:** Use the custom configuration
|
||||
|
||||
```bash
|
||||
experiencemaker config_path=/path/to/my_custom_config.yaml
|
||||
```
|
||||
|
||||
### Method 2: Command Line Parameters
|
||||
|
||||
Override any configuration parameter using dot notation:
|
||||
|
||||
```bash
|
||||
# Basic parameter override
|
||||
experiencemaker \
|
||||
llm.default.model_name=gpt-4 \
|
||||
embedding_model.default.model_name=text-embedding-3-large
|
||||
|
||||
# Operation parameters
|
||||
experiencemaker \
|
||||
op.recall_experience_op.params.retrieve_top_k=20 \
|
||||
op.rerank_experience_op.params.top_k=8
|
||||
|
||||
# Service configuration
|
||||
experiencemaker \
|
||||
http_service.port=8080 \
|
||||
thread_pool.max_workers=32
|
||||
|
||||
# Pipeline configuration
|
||||
experiencemaker \
|
||||
api.retriever="custom_op1->custom_op2"
|
||||
```
|
||||
|
||||
### Method 3: Hybrid Approach
|
||||
|
||||
Combine configuration file with command line overrides:
|
||||
|
||||
```bash
|
||||
experiencemaker \
|
||||
config_path=/path/to/base_config.yaml \
|
||||
llm.default.model_name=gpt-4 \
|
||||
op.recall_experience_op.params.retrieve_top_k=25
|
||||
```
|
||||
|
||||
## 🎯 Practical Examples
|
||||
|
||||
### Example 1: High-Performance Configuration
|
||||
|
||||
```bash
|
||||
experiencemaker \
|
||||
http_service.port=8002 \
|
||||
thread_pool.max_workers=64 \
|
||||
op.recall_experience_op.params.retrieve_top_k=50 \
|
||||
op.rerank_experience_op.params.top_k=10 \
|
||||
llm.default.params.temperature=0.1
|
||||
```
|
||||
|
||||
### Example 2: Development Configuration
|
||||
|
||||
```yaml
|
||||
# dev_config.yaml
|
||||
http_service:
|
||||
port: 8003
|
||||
|
||||
api:
|
||||
retriever: recall_experience_op->rerank_experience_op
|
||||
|
||||
op:
|
||||
recall_experience_op:
|
||||
params:
|
||||
retrieve_top_k: 5 # Faster for development
|
||||
|
||||
rerank_experience_op:
|
||||
params:
|
||||
top_k: 3
|
||||
|
||||
llm:
|
||||
default:
|
||||
model_name: qwen-turbo
|
||||
params:
|
||||
temperature: 0.8
|
||||
```
|
||||
|
||||
```bash
|
||||
experiencemaker config_path=dev_config.yaml
|
||||
```
|
||||
|
||||
### Example 3: Multi-Backend Setup
|
||||
|
||||
```yaml
|
||||
# multi_backend_config.yaml
|
||||
llm:
|
||||
fast:
|
||||
backend: openai_compatible
|
||||
model_name: qwen-turbo
|
||||
params:
|
||||
temperature: 0.9
|
||||
|
||||
accurate:
|
||||
backend: openai_compatible
|
||||
model_name: gpt-4
|
||||
params:
|
||||
temperature: 0.1
|
||||
|
||||
op:
|
||||
quick_extraction_op:
|
||||
backend: success_extraction_op
|
||||
llm: fast
|
||||
|
||||
detailed_validation_op:
|
||||
backend: experience_validation_op
|
||||
llm: accurate
|
||||
params:
|
||||
validation_threshold: 0.8
|
||||
```
|
||||
|
||||
## 📋 Configuration Tips
|
||||
|
||||
1. **Start Simple**: Begin with the default configuration and override specific parameters
|
||||
2. **Use Environment Variables**: Set API keys and URLs in `.env` file
|
||||
3. **Parameter Validation**: Invalid parameters will cause startup errors with detailed messages
|
||||
4. **Performance Tuning**: Adjust `retrieve_top_k`, `top_k`, and `max_workers` based on your needs
|
||||
5. **Pipeline Testing**: Use simple pipelines first, then gradually add complexity
|
||||
|
||||
## 🔍 Troubleshooting
|
||||
|
||||
### Common Issues
|
||||
|
||||
**Configuration Not Loading:**
|
||||
|
||||
```bash
|
||||
# Check if config file exists and has correct YAML syntax
|
||||
experiencemaker config_path=/full/path/to/config.yaml
|
||||
```
|
||||
|
||||
**Parameter Override Not Working:**
|
||||
|
||||
```bash
|
||||
# Use exact parameter path from configuration structure
|
||||
experiencemaker op.operation_name.params.parameter_name=value
|
||||
```
|
||||
|
||||
**Pipeline Syntax Errors:**
|
||||
|
||||
- Check for balanced brackets `[]`
|
||||
- Ensure operation names exist in `op` section
|
||||
- Use `|` only within `[]` groups
|
||||
|
||||
---
|
||||
|
||||
🎯 **Advanced Configuration Mastery!** You can now create sophisticated ExperienceMaker setups tailored to your specific
|
||||
needs.
|
||||
|
|
@ -1,15 +1,9 @@
|
|||
# Services Params Documentation
|
||||
# Configuration Guide
|
||||
|
||||
This document describes all available command-line parameters for ExperienceMaker Service.
|
||||
This document describes all available parameters for ExperienceMaker Service.
|
||||
The application uses [OmegaConf](https://omegaconf.readthedocs.io/) for configuration management, supporting both YAML
|
||||
files and command-line overrides.
|
||||
|
||||
## Basic Usage
|
||||
|
||||
```bash
|
||||
experiencemaker [parameter1=value1] [parameter2=value2] ...
|
||||
```
|
||||
|
||||
## Configuration Loading Priority
|
||||
|
||||
1. Default values from `AppConfig` dataclass
|
||||
|
|
@ -17,7 +11,153 @@ experiencemaker [parameter1=value1] [parameter2=value2] ...
|
|||
3. Custom YAML file (if `config_path` is specified)
|
||||
4. Command-line overrides
|
||||
|
||||
## Basic Configuration Parameters
|
||||
## 🏗️ Configuration Architecture
|
||||
|
||||
ExperienceMaker uses a layered configuration system with the following priority order:
|
||||
|
||||
1. **Default Configuration** (lowest priority)
|
||||
2. **YAML Configuration File**
|
||||
3. **Command Line Arguments** (highest priority)
|
||||
|
||||
## Basic Bash Usage
|
||||
|
||||
```bash
|
||||
experiencemaker [parameter1=value1] [parameter2=value2] ...
|
||||
```
|
||||
|
||||
## 🧩 YAML Configuration Composition
|
||||
|
||||
The YAML configuration file follows a specific composition pattern that enables flexible and modular configuration:
|
||||
|
||||
### 1. Resource Declaration
|
||||
|
||||
First, you declare the three core resources that form the foundation of the system:
|
||||
|
||||
- **`llm`**: Language model configurations
|
||||
- **`embedding_model`**: Embedding model configurations
|
||||
- **`vector_store`**: Vector storage configurations
|
||||
|
||||
In these sections, `default` (or any custom name) represents a declared configuration object that can be referenced
|
||||
later:
|
||||
|
||||
```yaml
|
||||
llm:
|
||||
default: # This is a declared LLM configuration object
|
||||
backend: openai_compatible
|
||||
model_name: qwen3-32b
|
||||
|
||||
embedding_model:
|
||||
default: # This is a declared embedding model configuration object
|
||||
backend: openai_compatible
|
||||
model_name: text-embedding-v4
|
||||
|
||||
vector_store:
|
||||
default: # This is a declared vector store configuration object
|
||||
backend: local_file
|
||||
embedding_model: default
|
||||
```
|
||||
|
||||
### 2. Operation Backend Registration
|
||||
|
||||
In the `op` section, each operation declares its `backend` implementation. The backend names are registered through
|
||||
`@OP_REGISTRY.register()` decorator, typically converting camel-case class names to underscore format:
|
||||
|
||||
```yaml
|
||||
op:
|
||||
recall_experience_op:
|
||||
backend: recall_experience_op # Registered via @OP_REGISTRY.register()
|
||||
```
|
||||
|
||||
### 3. Resource References
|
||||
|
||||
Operations reference the previously declared resources using their names:
|
||||
|
||||
```yaml
|
||||
op:
|
||||
recall_experience_op:
|
||||
backend: recall_experience_op
|
||||
llm: default # References the declared LLM object
|
||||
embedding_model: default # References the declared embedding model object
|
||||
vector_store: default # References the declared vector store object
|
||||
```
|
||||
|
||||
### 4. Pipeline
|
||||
|
||||
Pipeline configurations use a special syntax to define operation flows:
|
||||
|
||||
- `->`: Sequential execution
|
||||
- `[]`: Parallel execution group
|
||||
- `|`: Alternative operations within parallel group
|
||||
|
||||
### Examples
|
||||
|
||||
```yaml
|
||||
# Sequential pipeline
|
||||
api:
|
||||
retriever: op1->op2->op3
|
||||
|
||||
# Parallel execution
|
||||
summarizer: op1->[op2|op3|op4]->op5
|
||||
|
||||
# Complex pipeline with nested parallel operations
|
||||
vector_store: preprocess_op->[recall_op->rerank_op|backup_op]->merge_op
|
||||
```
|
||||
|
||||
This compositional approach enables:
|
||||
|
||||
- **Modularity**: Declare resources once, reference everywhere
|
||||
- **Flexibility**: Mix and match different backends and configurations
|
||||
- **Complexity**: Build sophisticated processing chains through pipeline syntax
|
||||
|
||||
## 📁 Configuration Structure
|
||||
|
||||
```yaml
|
||||
# Service Configuration
|
||||
http_service:
|
||||
host: "0.0.0.0"
|
||||
port: 8001
|
||||
timeout_keep_alive: 600
|
||||
limit_concurrency: 64
|
||||
|
||||
# Pipeline Definitions
|
||||
api:
|
||||
retriever: recall_experience_op->rerank_experience_op->rewrite_experience_op
|
||||
summarizer: trajectory_preprocess_op->[success_extraction_op|failure_extraction_op]->experience_validation_op
|
||||
vector_store: vector_store_action_op
|
||||
|
||||
# Operation Configurations
|
||||
op:
|
||||
operation_name:
|
||||
backend: operation_name # Register through `@OP_REGISTRY.register()`, typically by converting camel-cased types into underscored names
|
||||
llm: default # Optional: reference to LLM config, Register through `@LLM_REGISTRY.register()`
|
||||
embedding_model: default # Optional: reference to embedding config, Register through `@EMBEDDING_MODEL_REGISTRY.register()`
|
||||
vector_store: default # Optional: reference to vector store config, Register through `@VECTOR_STORE_REGISTRY.register()`
|
||||
params: # Operation-specific parameters
|
||||
param1: value1
|
||||
param2: value2
|
||||
|
||||
# Resource Configurations
|
||||
llm:
|
||||
default:
|
||||
backend: openai_compatible
|
||||
model_name: qwen3-32b
|
||||
params:
|
||||
temperature: 0.6
|
||||
|
||||
embedding_model:
|
||||
default:
|
||||
backend: openai_compatible
|
||||
model_name: text-embedding-v4
|
||||
params:
|
||||
dimensions: 1024
|
||||
|
||||
vector_store:
|
||||
default:
|
||||
backend: local_file
|
||||
embedding_model: default
|
||||
```
|
||||
|
||||
## Detailed Configuration Parameters
|
||||
|
||||
| Parameter | Type | Default Value | Description | Example |
|
||||
|----------------------|--------|-----------------|----------------------------------------------------------------------|-------------------------------------------|
|
||||
|
|
@ -86,38 +226,112 @@ parameters:
|
|||
| `vector_store.{name}.embedding_model` | string | `""` | Reference to embedding model configuration | `vector_store.default.embedding_model=default` |
|
||||
| `vector_store.{name}.params.{param}` | any | `{}` | Vector store-specific parameters | `vector_store.default.params.store_dir=file_vector_store` |
|
||||
|
||||
## Complete Example
|
||||
|
||||
Here's a complete example showing how to configure the entire system:
|
||||
## 🎯 Practical Examples
|
||||
|
||||
### Example 1
|
||||
|
||||
```bash
|
||||
experiencemaker \
|
||||
http_service.port=8080 \
|
||||
thread_pool.max_workers=20 \
|
||||
llm.default.backend=openai_compatible \
|
||||
llm.default.model_name=qwen3-32b \
|
||||
llm.default.params.temperature=0.6 \
|
||||
embedding_model.default.backend=openai_compatible \
|
||||
embedding_model.default.model_name=text-embedding-v4 \
|
||||
embedding_model.default.params.dimensions=1024 \
|
||||
vector_store.default.backend=elasticsearch \
|
||||
vector_store.default.embedding_model=default \
|
||||
http_service.port=8002 \
|
||||
thread_pool.max_workers=64 \
|
||||
op.recall_experience_op.params.retrieve_top_k=50 \
|
||||
op.rerank_experience_op.params.top_k=10 \
|
||||
llm.default.params.temperature=0.1
|
||||
```
|
||||
|
||||
## Configuration File vs Command Line
|
||||
### Example 2
|
||||
|
||||
You can also create a YAML configuration file and override specific parameters:
|
||||
```yaml
|
||||
# dev_config.yaml
|
||||
http_service:
|
||||
port: 8003
|
||||
|
||||
1. Create a custom configuration file (`xxx/my_config.yaml`)
|
||||
2. Use it with command-line overrides:
|
||||
api:
|
||||
retriever: recall_experience_op->rerank_experience_op
|
||||
|
||||
op:
|
||||
recall_experience_op:
|
||||
params:
|
||||
retrieve_top_k: 5 # Faster for development
|
||||
|
||||
rerank_experience_op:
|
||||
params:
|
||||
top_k: 3
|
||||
|
||||
llm:
|
||||
default:
|
||||
model_name: qwen-turbo
|
||||
params:
|
||||
temperature: 0.8
|
||||
```
|
||||
|
||||
```bash
|
||||
experiencemaker config_path=xxx/my_config.yaml llm.default.model_name=qwen3-32b http_service.port=8080
|
||||
experiencemaker config_path=dev_config.yaml
|
||||
```
|
||||
|
||||
## Parameter Validation
|
||||
### Example 3: Multi-Backend Setup
|
||||
|
||||
- All parameters are validated according to their types
|
||||
- Referenced configurations (like `llm`, `embedding_model`, `vector_store`) must exist
|
||||
- Backend implementations must be registered in their respective registries
|
||||
- Nested parameters use dot notation for access
|
||||
```yaml
|
||||
# multi_backend_config.yaml
|
||||
llm:
|
||||
fast:
|
||||
backend: openai_compatible
|
||||
model_name: qwen-turbo
|
||||
params:
|
||||
temperature: 0.9
|
||||
|
||||
accurate:
|
||||
backend: openai_compatible
|
||||
model_name: gpt-4
|
||||
params:
|
||||
temperature: 0.1
|
||||
|
||||
op:
|
||||
quick_extraction_op:
|
||||
backend: success_extraction_op
|
||||
llm: fast
|
||||
|
||||
detailed_validation_op:
|
||||
backend: experience_validation_op
|
||||
llm: accurate
|
||||
params:
|
||||
validation_threshold: 0.8
|
||||
```
|
||||
|
||||
## 📋 Configuration Tips
|
||||
|
||||
1. **Start Simple**: Begin with the default configuration and override specific parameters
|
||||
2. **Use Environment Variables**: Set API keys and URLs in `.env` file
|
||||
3. **Parameter Validation**: Invalid parameters will cause startup errors with detailed messages
|
||||
4. **Performance Tuning**: Adjust `retrieve_top_k`, `top_k`, and `max_workers` based on your needs
|
||||
5. **Pipeline Testing**: Use simple pipelines first, then gradually add complexity
|
||||
|
||||
## 🔍 Troubleshooting
|
||||
|
||||
### Common Issues
|
||||
|
||||
**Configuration Not Loading:**
|
||||
|
||||
```bash
|
||||
# Check if config file exists and has correct YAML syntax
|
||||
experiencemaker config_path=/full/path/to/config.yaml
|
||||
```
|
||||
|
||||
**Parameter Override Not Working:**
|
||||
|
||||
```bash
|
||||
# Use exact parameter path from configuration structure
|
||||
experiencemaker op.operation_name.params.parameter_name=value
|
||||
```
|
||||
|
||||
**Pipeline Syntax Errors:**
|
||||
|
||||
- Check for balanced brackets `[]`
|
||||
- Ensure operation names exist in `op` section
|
||||
- Use `|` only within `[]` groups
|
||||
|
||||
---
|
||||
|
||||
🎯 **Advanced Configuration Mastery!** You can now create sophisticated ExperienceMaker setups tailored to your specific
|
||||
needs.
|
||||
|
|
@ -13,7 +13,6 @@ Just as financial analysts develop analytical frameworks, senior engineers estab
|
|||
- [ ] Coding
|
||||
- [ ] Education
|
||||
- [ ] Research
|
||||
- etc
|
||||
- [ ] Experience marketplace: community-driven experience sharing and exchange
|
||||
|
||||
## P0 - Support for Rich Experience Formats
|
||||
|
|
@ -24,6 +23,14 @@ Expert knowledge extends beyond text to include debugged code, fine-tuned toolch
|
|||
- [ ] **Tool Integration**: APIs, MCP configurations, and tool setups
|
||||
- [ ] **Pipeline Templates**: Agent execution pipelines and multi-step tool combinations
|
||||
|
||||
## P0 - MCP Integration
|
||||
|
||||
Modernize our API architecture by migrating three core APIs to the Model Context Protocol (MCP) standard for improved interoperability and standardization.
|
||||
|
||||
- [ ] Summarizer API
|
||||
- [ ] Retriever API
|
||||
- [ ] Vector Store API
|
||||
|
||||
## P1 - Experience Validation & Optimization
|
||||
|
||||
AI-powered analysis of experience usage patterns and effectiveness, with automatic quality optimization and cross-task validation feedback loops.
|
||||
|
|
@ -42,14 +49,26 @@ Transform valuable experience data from daily work into usable insights:
|
|||
|
||||
Enable AI to naturally become stronger through everyday work, rather than wasting real-world experience data due to format limitations.
|
||||
|
||||
## P2 - Open Source Experience Libraries
|
||||
|
||||
Democratize AI experience sharing by making curated experience libraries publicly available on Hugging Face, enabling the broader AI community to benefit from and contribute to professional experience repositories.
|
||||
|
||||
- [ ] **Hugging Face Integration**: Upload and maintain experience libraries on Hugging Face Hub
|
||||
- [ ] **Community Contributions**: Enable community-driven experience library improvements and additions
|
||||
- [ ] **Standardized Formats**: Establish standard formats for experience sharing across different domains
|
||||
- [ ] **Version Control**: Implement versioning system for experience library updates and improvements
|
||||
|
||||
## Current TODO
|
||||
|
||||
- [x] op dev & test ready
|
||||
- [ ] integrate into beyond-agent @jinli
|
||||
- [x] op dev & test ready @jiaji
|
||||
- [ ] cook_book-appworld code & readme @jiaji
|
||||
- [ ] cook_book-bfcl-v3 op @zouyin delay 0730
|
||||
- [ ] fix multi-process bug @jinli
|
||||
- [x] fix multi-process bug @jinli
|
||||
- [ ] logo optimize @jiaji
|
||||
- [ ] Ready-made Experience Store @jinli, add appworld/bfcl-v3 default experience store @jiaji
|
||||
- [x] Ready-made Experience Store @jinli, add appworld/bfcl-v3 default experience store @jiaji
|
||||
- [ ] op config make up @jiaji
|
||||
- [ ] config make up, easy to understand @jinli
|
||||
- [x] config make up, easy to understand @jinli
|
||||
- [x] refine readme @jinli
|
||||
|
||||
- [x] integrate into beyond-agent @jinli
|
||||
- [ ] rm workspace_id in code @jinli
|
||||
|
|
|
|||
|
|
@ -1,174 +0,0 @@
|
|||
# ExperienceMaker Quick Start Guide
|
||||
This guide will help you get started with ExperienceMaker quickly using practical examples.
|
||||
|
||||
## 🚀 What You'll Learn
|
||||
- How to set up ExperienceMaker service
|
||||
- Run an agent and generate experiences
|
||||
- Retrieve and apply experiences to new tasks
|
||||
- Build experience-enhanced agents
|
||||
|
||||
## 📋 Prerequisites
|
||||
- Python 3.12+
|
||||
- LLM API access (OpenAI or compatible)
|
||||
- Embedding model API access
|
||||
|
||||
## 🛠️ Installation
|
||||
|
||||
### Option 1: Install from PyPI (Recommended)
|
||||
|
||||
```bash
|
||||
pip install experiencemaker
|
||||
```
|
||||
|
||||
### Option 2: Install from Source
|
||||
|
||||
```bash
|
||||
git clone https://github.com/modelscope/ExperienceMaker.git
|
||||
cd ExperienceMaker
|
||||
pip install .
|
||||
```
|
||||
|
||||
## ⚙️ Environment Setup
|
||||
Create a `.env` file in your project directory:
|
||||
|
||||
```bash
|
||||
# Required: LLM API configuration
|
||||
LLM_API_KEY="sk-xxx"
|
||||
LLM_BASE_URL="https://xxx.com/v1"
|
||||
|
||||
# Required: Embedding model configuration
|
||||
EMBEDDING_MODEL_API_KEY="sk-xxx"
|
||||
EMBEDDING_MODEL_BASE_URL="https://xxx.com/v1"
|
||||
|
||||
# Optional: Elasticsearch configuration (if using Elasticsearch backend)
|
||||
|
||||
```
|
||||
|
||||
## 🚀 Start the Service
|
||||
For testing, use the `local_file` backend:
|
||||
```bash
|
||||
experiencemaker \
|
||||
http_service.port=8001 \
|
||||
llm.default.model_name=qwen3-32b \
|
||||
embedding_model.default.model_name=text-embedding-v4 \
|
||||
vector_store.default.backend=local_file
|
||||
```
|
||||
The service will start on `http://localhost:8001`
|
||||
|
||||
### Elasticsearch Backend
|
||||
```bash
|
||||
experiencemaker \
|
||||
llm.default.model_name=qwen3-32b \
|
||||
embedding_model.default.model_name=text-embedding-v4 \
|
||||
vector_store.default.backend=elasticsearch
|
||||
```
|
||||
|
||||
**Setup Elasticsearch:**
|
||||
```bash
|
||||
export ES_HOSTS="http://localhost:9200"
|
||||
# Quick setup using Elastic's official script
|
||||
curl -fsSL https://elastic.co/start-local | sh
|
||||
```
|
||||
📖 **Need Help?** Refer to [Vector Store Setup](./doc/vector_store_setup.md) for comprehensive deployment guidance.
|
||||
|
||||
## 📝 Your First ExperienceMaker Script
|
||||
|
||||
Here's how to get started!
|
||||
- The `load_dotenv()` function loads environment variables from your `.env` file, or you can manually export them.
|
||||
- The `base_url` points to your ExperienceMaker service.
|
||||
- The `workspace_id` serves as your experience storage namespace. Experiences in different workspaces remain completely
|
||||
isolated and cannot access each other.
|
||||
|
||||
```python
|
||||
import requests
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
base_url = "http://0.0.0.0:8001/"
|
||||
workspace_id = "test_workspace"
|
||||
```
|
||||
|
||||
### 📊 Call Summarizer Examples
|
||||
|
||||
Batch summarize the trajectory list, where each trajectory consists of a message and a score.
|
||||
|
||||
- The message is the conversation history.
|
||||
- The score represents the rating between 0 and 1, with 0 typically indicating failure and 1 indicating success.
|
||||
|
||||
```python
|
||||
response = requests.post(url=base_url + "summarizer", json={
|
||||
"workspace_id": workspace_id,
|
||||
"traj_list": [
|
||||
{"messages": messages, "score": 1.0}
|
||||
]
|
||||
})
|
||||
|
||||
response = response.json()
|
||||
experience_list = response["experience_list"]
|
||||
for experience in experience_list:
|
||||
print(experience)
|
||||
```
|
||||
|
||||
### 🔍 Call Retriever Examples
|
||||
Retrieve the top_k={top_k} experiences related to {query} in workspace=test_workspace, and finally accept the assembled context.
|
||||
Alternatively, you can also accept the raw experience_list parameter and assemble the context yourself.
|
||||
|
||||
```python
|
||||
response = requests.post(url=base_url + "retriever", json={
|
||||
"workspace_id": workspace_id,
|
||||
"query": query,
|
||||
"top_k": 1,
|
||||
})
|
||||
|
||||
response = response.json()
|
||||
experience_merged: str = response["experience_merged"]
|
||||
print(f"experience_merged={experience_merged}")
|
||||
```
|
||||
|
||||
### 💾 Dump Experiences From Vector Store
|
||||
Dump the experience with workspace_id from the vector store into the {path}/{workspace_id}.jsonl file.
|
||||
|
||||
```python
|
||||
response = requests.post(url=base_url + "vector_store", json={
|
||||
"workspace_id": workspace_id,
|
||||
"action": "dump",
|
||||
"path": "./",
|
||||
})
|
||||
print(response.json())
|
||||
```
|
||||
|
||||
### 📥 Load Experiences To Vector Store
|
||||
Load the {path}/{workspace_id}.jsonl file into the vector store, workspace_id={workspace_id}.
|
||||
|
||||
```python
|
||||
response = requests.post(url=base_url + "vector_store", json={
|
||||
"workspace_id": workspace_id,
|
||||
"action": "load",
|
||||
"path": "./",
|
||||
})
|
||||
|
||||
print(response.json())
|
||||
```
|
||||
|
||||
🎭 **Want to See It in Action?** We've prepared a [simple react agent](./cookbook/simple_demo/simple_demo.py) that demonstrates how to enhance agent capabilities by integrating summarizer and retriever components, achieving significantly better performance.
|
||||
|
||||
## 🐛 Common Issues
|
||||
|
||||
### Service Won't Start
|
||||
- Check if port 8001 is available
|
||||
- Verify your API keys in `.env` file
|
||||
- Ensure Python version is 3.12+
|
||||
|
||||
### No Experiences Retrieved
|
||||
- Make sure you've run the summarizer first
|
||||
- Check if workspace_id matches between operations
|
||||
- Verify vector store backend is properly configured
|
||||
|
||||
### API Connection Errors
|
||||
- Confirm LLM_BASE_URL and API keys are correct
|
||||
- Test API access independently
|
||||
- Check network connectivity
|
||||
|
||||
---
|
||||
|
||||
🎯 **You're all set!** You now have a working ExperienceMaker setup that can learn from interactions and improve over time.
|
||||
242
experience_library/appworld_v1.jsonl
Normal file
242
experience_library/appworld_v1.jsonl
Normal file
File diff suppressed because one or more lines are too long
|
|
@ -65,7 +65,7 @@ class FailureExtractionOp(BaseOp):
|
|||
experiences = []
|
||||
|
||||
for exp_data in experiences_data:
|
||||
experience = BaseExperience(
|
||||
experience: BaseExperience = TextExperience(
|
||||
workspace_id=self.context.request.workspace_id,
|
||||
when_to_use=exp_data.get("when_to_use", exp_data.get("condition", "")),
|
||||
content=exp_data.get("experience", ""),
|
||||
|
|
|
|||
|
|
@ -7,7 +7,7 @@ from loguru import logger
|
|||
from experiencemaker.enumeration.role import Role
|
||||
from experiencemaker.op import OP_REGISTRY
|
||||
from experiencemaker.op.base_op import BaseOp
|
||||
from experiencemaker.schema.experience import BaseExperience, ExperienceMeta
|
||||
from experiencemaker.schema.experience import BaseExperience, ExperienceMeta, TextExperience
|
||||
from experiencemaker.schema.message import Message, Trajectory
|
||||
from experiencemaker.schema.response import SummarizerResponse
|
||||
from experiencemaker.utils.op_utils import merge_messages_content, parse_json_experience_response, get_trajectory_context
|
||||
|
|
@ -59,12 +59,12 @@ class SuccessExtractionOp(BaseOp):
|
|||
outcome="successful"
|
||||
)
|
||||
|
||||
def parse_experiences(message: Message) -> List[BaseExperience]:
|
||||
def parse_experiences(message: Message) -> List[TextExperience]:
|
||||
experiences_data = parse_json_experience_response(message.content)
|
||||
experiences = []
|
||||
|
||||
for exp_data in experiences_data:
|
||||
experience = BaseExperience(
|
||||
experience = TextExperience(
|
||||
workspace_id=self.context.request.workspace_id,
|
||||
when_to_use=exp_data.get("when_to_use", exp_data.get("condition", "")),
|
||||
content=exp_data.get("experience", ""),
|
||||
|
|
|
|||
|
|
@ -1,7 +1,9 @@
|
|||
import datetime
|
||||
from abc import ABC
|
||||
from typing import List
|
||||
from uuid import uuid4
|
||||
|
||||
from loguru import logger
|
||||
from pydantic import BaseModel, Field
|
||||
|
||||
from experiencemaker.schema.vector_node import VectorNode
|
||||
|
|
@ -17,7 +19,7 @@ class ExperienceMeta(BaseModel):
|
|||
self.modified_time = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
|
||||
|
||||
|
||||
class BaseExperience(BaseModel):
|
||||
class BaseExperience(BaseModel, ABC):
|
||||
workspace_id: str = Field(default="")
|
||||
|
||||
experience_id: str = Field(default_factory=lambda: uuid4().hex)
|
||||
|
|
@ -101,8 +103,8 @@ def vector_node_to_experience(node: VectorNode) -> BaseExperience:
|
|||
return KnowledgeExperience.from_vector_node(node)
|
||||
|
||||
else:
|
||||
raise RuntimeError(f"experience type {experience_type} not supported")
|
||||
|
||||
logger.warning(f"experience type {experience_type} not supported")
|
||||
return TextExperience.from_vector_node(node)
|
||||
|
||||
if __name__ == "__main__":
|
||||
e1 = TextExperience(
|
||||
|
|
|
|||
|
|
@ -28,7 +28,9 @@ class BaseVectorStore(BaseModel, ABC):
|
|||
try:
|
||||
for line in tqdm(f, desc="load from path"):
|
||||
if line.strip():
|
||||
yield VectorNode(**json.loads(line.strip(), **kwargs))
|
||||
node = VectorNode(**json.loads(line.strip(), **kwargs))
|
||||
node.workspace_id = workspace_id
|
||||
yield node
|
||||
|
||||
finally:
|
||||
fcntl.flock(f, fcntl.LOCK_UN)
|
||||
|
|
@ -45,6 +47,7 @@ class BaseVectorStore(BaseModel, ABC):
|
|||
fcntl.flock(f, fcntl.LOCK_EX)
|
||||
try:
|
||||
for node in tqdm(nodes, desc="dump to path"):
|
||||
node.workspace_id = workspace_id
|
||||
f.write(json.dumps(node.model_dump(), ensure_ascii=ensure_ascii, **kwargs))
|
||||
f.write("\n")
|
||||
count += 1
|
||||
|
|
|
|||
|
|
@ -51,14 +51,15 @@ class EsVectorStore(BaseVectorStore):
|
|||
def _iter_workspace_nodes(self, workspace_id: str, **kwargs) -> Iterable[VectorNode]:
|
||||
response = self._client.search(index=workspace_id, body={"query": {"match_all": {}}})
|
||||
for doc in response['hits']['hits']:
|
||||
yield self.doc2node(doc)
|
||||
yield self.doc2node(doc, workspace_id)
|
||||
|
||||
def refresh(self, workspace_id: str):
|
||||
self._client.indices.refresh(index=workspace_id)
|
||||
|
||||
@staticmethod
|
||||
def doc2node(doc) -> VectorNode:
|
||||
def doc2node(doc, workspace_id: str) -> VectorNode:
|
||||
node = VectorNode(**doc["_source"])
|
||||
node.workspace_id = workspace_id
|
||||
node.unique_id = doc["_id"]
|
||||
if "_score" in doc:
|
||||
node.metadata["_score"] = doc["_score"] - 1
|
||||
|
|
@ -105,7 +106,7 @@ class EsVectorStore(BaseVectorStore):
|
|||
|
||||
nodes: List[VectorNode] = []
|
||||
for doc in response['hits']['hits']:
|
||||
nodes.append(self.doc2node(doc))
|
||||
nodes.append(self.doc2node(doc, workspace_id))
|
||||
|
||||
self.retrieve_filters.clear()
|
||||
return nodes
|
||||
|
|
@ -124,10 +125,10 @@ class EsVectorStore(BaseVectorStore):
|
|||
docs = [
|
||||
{
|
||||
"_op_type": "index",
|
||||
"_index": node.workspace_id,
|
||||
"_index": workspace_id,
|
||||
"_id": node.unique_id,
|
||||
"_source": {
|
||||
"workspace_id": node.workspace_id,
|
||||
"workspace_id": workspace_id,
|
||||
"content": node.content,
|
||||
"metadata": node.metadata,
|
||||
"vector": node.vector
|
||||
|
|
|
|||
|
|
@ -84,6 +84,7 @@ class FileVectorStore(BaseVectorStore):
|
|||
workspace_id=workspace_id,
|
||||
path=self.store_path,
|
||||
**kwargs)
|
||||
|
||||
logger.info(f"update workspace_id={workspace_id} nodes.size={len(nodes)} all.size={len(all_node_dict)} "
|
||||
f"update_cnt={update_cnt}")
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue