TemplateLLM API Reference

class lmitf.templete_llm.TemplateLLM(template_path: str, api_key: str = None, base_url: str = None)[source]

Bases: BaseLLM

__init__(template_path: str, api_key: str = None, base_url: str = None)[source]

初始化 LLM 客户端

Parameters:

  • api_key (str, optional) – OpenAI API 密钥。如果未提供，将从环境变量 OPENAI_API_KEY 读取

  • base_url (str, optional) – API 基础URL。如果未提供，将从环境变量 OPENAI_BASE_URL 读取

call(**kwargs)[source]

调用LLM，生成响应

Returns:

LLM的响应内容

Return type:

str

Overview

The TemplateLLM class extends BaseLLM to provide template-based prompt management. It allows you to define reusable prompt templates with variables, making it easier to maintain consistent AI interactions across your application.

Class Reference

TemplateLLM

class lmitf.templete_llm.TemplateLLM(template_path: str, api_key: str = None, base_url: str = None)[source]

Bases: BaseLLM

__init__(template_path: str, api_key: str = None, base_url: str = None)[source]

初始化 LLM 客户端

Parameters:

  • api_key (str, optional) – OpenAI API 密钥。如果未提供，将从环境变量 OPENAI_API_KEY 读取

  • base_url (str, optional) – API 基础URL。如果未提供，将从环境变量 OPENAI_BASE_URL 读取

call(**kwargs)[source]

调用LLM，生成响应

Returns:

LLM的响应内容

Return type:

str

Key Features

  • Template System: Define reusable prompt templates with variables

  • Python-based Templates: Templates are Python files for maximum flexibility

  • Variable Substitution: Support for complex variable substitution patterns

  • Inheritance: Extends all BaseLLM functionality

  • Batch Processing: Process multiple inputs with the same template

Template File Format

Templates are Python files that define prompt structures:

# example_template.py
template = """
Task: {task_description}

Input: {input_text}

Please provide a detailed analysis following these guidelines:
1. {guideline_1}
2. {guideline_2}
3. {guideline_3}

Output format: {output_format}
"""

# Optional: Define default values
defaults = {
    "output_format": "JSON",
    "task_description": "Analysis Task"
}

# Optional: Define validation rules
required_vars = ["input_text", "guideline_1", "guideline_2", "guideline_3"]

Usage Examples

Basic Template Usage

from lmitf import TemplateLLM

# Load a template
template_llm = TemplateLLM("path/to/template.py")

# Use the template
response = template_llm.call_template(
    input_text="Your text here",
    guideline_1="Be comprehensive",
    guideline_2="Use specific examples", 
    guideline_3="Provide actionable insights"
)
print(response)

Pre-built Templates

LMITF includes several pre-built templates:

# Text to triples extraction
from lmitf.datasets.llm_prompts.text2triples import template

template_llm = TemplateLLM("lmitf/datasets/llm_prompts/text2triples.py")
result = template_llm.call_template(
    text="John works at Google as a software engineer."
)
print(result)  # Extracts relationship triples

Custom Template Creation

Create your own template file:

# my_template.py
template = """
You are a {role} expert. 

Task: {task}

Context: {context}

Requirements:
{requirements}

Please provide your {output_type}.
"""

defaults = {
    "role": "AI assistant",
    "output_type": "response"
}

required_vars = ["task", "context", "requirements"]

Use the custom template:

template_llm = TemplateLLM("my_template.py")

response = template_llm.call_template(
    role="data science",
    task="Analyze this dataset",
    context="E-commerce sales data from 2023",
    requirements="Focus on trends and anomalies",
    output_type="detailed report"
)

Batch Processing

Process multiple inputs with the same template:

inputs = [
    {"text": "First document content"},
    {"text": "Second document content"},
    {"text": "Third document content"}
]

results = []
for input_data in inputs:
    result = template_llm.call_template(**input_data)
    results.append(result)

Method Reference

call_template()

Execute a template with provided variables.

Parameters:

  • **kwargs: Template variables as keyword arguments

Returns:

  • str: Generated response using the template

reload_template()

Reload the template file (useful during development).

# After modifying the template file
template_llm.reload_template()

Template Best Practices

1. Clear Variable Names

# Good
template = "Analyze {dataset_name} for {analysis_type} patterns"

# Avoid
template = "Analyze {x} for {y} patterns"

2. Provide Defaults

defaults = {
    "tone": "professional",
    "format": "markdown",
    "max_length": "500 words"
}

3. Use Required Variables

required_vars = ["input_text", "task_type"]
# Will raise error if these aren't provided

4. Template Documentation

"""
Template: Content Analysis
Description: Analyzes content for sentiment, topics, and key insights
Variables:
  - content (required): Text content to analyze
  - focus_areas (optional): Specific areas to focus on
  - output_format (optional): JSON, markdown, or plain text
"""

Advanced Features

Conditional Logic in Templates

template = """
{%- if analysis_type == "detailed" %}
Please provide a comprehensive analysis including:
1. Summary
2. Key themes
3. Recommendations
{%- else %}
Please provide a brief summary.
{%- endif %}

Content: {content}
"""

Template Inheritance

# base_template.py
base_template = """
System: You are a {role}.
Task: {task}
"""

# specific_template.py
from .base_template import base_template

template = base_template + """
Additional instructions: {instructions}
Input: {input_data}
"""

Built-in Templates

text2triples

Extracts relationship triples from text:

template_llm = TemplateLLM("lmitf/datasets/llm_prompts/text2triples.py")
result = template_llm.call_template(
    text="Apple was founded by Steve Jobs in Cupertino."
)
# Output: [("Apple", "founded_by", "Steve Jobs"), ("Apple", "located_in", "Cupertino")]

Error Handling

try:
    template_llm = TemplateLLM("nonexistent_template.py")
except FileNotFoundError:
    print("Template file not found")

try:
    result = template_llm.call_template(missing_required_var="value")
except KeyError as e:
    print(f"Missing required variable: {e}")