ayrisdev/crawl4ai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Feat/llm config (#724)

Browse Source 
* feature: Add LlmConfig to easily configure and pass LLM configs to different strategies

* pulled in next branch and resolved conflicts

* feat: Add gemini and deepseek providers. Make ignore_cache in llm content filter to true by default to avoid confusions

* Refactor: Update LlmConfig in LLMExtractionStrategy class and deprecate old params

* updated tests, docs and readme
This commit is contained in:
Aravind
2025-02-21 13:11:37 +05:30
committed by GitHub
parent 3cb28875c3
commit 2af958e12c
25 changed files with 420 additions and 240 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
docs/examples/extraction_strategies_examples.py
View File

@@ -11,6 +11,7 @@ import asyncio
import os
from crawl4ai import AsyncWebCrawler, BrowserConfig, CrawlerRunConfig, CacheMode
from crawl4ai.async_configs import LlmConfig
from crawl4ai.extraction_strategy import (
LLMExtractionStrategy,
JsonCssExtractionStrategy,
@@ -60,22 +61,19 @@ async def main():
# 1. LLM Extraction with different input formats
markdown_strategy = LLMExtractionStrategy(
provider="openai/gpt-4o-mini",
api_token=os.getenv("OPENAI_API_KEY"),
llmConfig = LlmConfig(provider="openai/gpt-4o-mini", api_token=os.getenv("OPENAI_API_KEY")),
instruction="Extract product information including name, price, and description",
)
html_strategy = LLMExtractionStrategy(
input_format="html",
provider="openai/gpt-4o-mini",
api_token=os.getenv("OPENAI_API_KEY"),
llmConfig=LlmConfig(provider="openai/gpt-4o-mini", api_token=os.getenv("OPENAI_API_KEY")),
instruction="Extract product information from HTML including structured data",
)
fit_markdown_strategy = LLMExtractionStrategy(
input_format="fit_markdown",
provider="openai/gpt-4o-mini",
api_token=os.getenv("OPENAI_API_KEY"),
llmConfig=LlmConfig(provider="openai/gpt-4o-mini",api_token=os.getenv("OPENAI_API_KEY")),
instruction="Extract product information from cleaned markdown",
)

4
docs/examples/llm_extraction_openai_pricing.py
View File

@@ -1,3 +1,4 @@
from crawl4ai.async_configs import LlmConfig
from crawl4ai.extraction_strategy import *
from crawl4ai.crawler_strategy import *
import asyncio
@@ -25,8 +26,7 @@ async def main():
word_count_threshold=1,
extraction_strategy=LLMExtractionStrategy(
# provider= "openai/gpt-4o", api_token = os.getenv('OPENAI_API_KEY'),
provider="groq/llama-3.1-70b-versatile",
api_token=os.getenv("GROQ_API_KEY"),
llmConfig=LlmConfig(provider="groq/llama-3.1-70b-versatile", api_token=os.getenv("GROQ_API_KEY")),
schema=OpenAIModelFee.model_json_schema(),
extraction_type="schema",
instruction="From the crawled content, extract all mentioned model names along with their "

7
docs/examples/llm_markdown_generator.py
View File

@@ -1,6 +1,7 @@
import os
import asyncio
from crawl4ai import AsyncWebCrawler, BrowserConfig, CrawlerRunConfig, CacheMode
from crawl4ai.async_configs import LlmConfig
from crawl4ai.content_filter_strategy import LLMContentFilter
async def test_llm_filter():
@@ -22,8 +23,7 @@ async def test_llm_filter():
# Initialize LLM filter with focused instruction
filter = LLMContentFilter(
provider="openai/gpt-4o",
api_token=os.getenv('OPENAI_API_KEY'),
llmConfig=LlmConfig(provider="openai/gpt-4o", api_token=os.getenv('OPENAI_API_KEY')),
instruction="""
Focus on extracting the core educational content about Python classes.
Include:
@@ -43,8 +43,7 @@ async def test_llm_filter():
)
filter = LLMContentFilter(
provider="openai/gpt-4o",
api_token=os.getenv('OPENAI_API_KEY'),
llmConfig=LlmConfig(provider="openai/gpt-4o",api_token=os.getenv('OPENAI_API_KEY')),
chunk_token_threshold=2 ** 12 * 2, # 2048 * 2
ignore_cache = True,
instruction="""

5
docs/examples/quickstart_async.config.py
View File

@@ -1,5 +1,7 @@
import os, sys
from crawl4ai.async_configs import LlmConfig
sys.path.append(
os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
)
@@ -209,8 +211,7 @@ async def extract_structured_data_using_llm(
word_count_threshold=1,
page_timeout=80000,
extraction_strategy=LLMExtractionStrategy(
provider=provider,
api_token=api_token,
llmConfig=LlmConfig(provider=provider,api_token=api_token),
schema=OpenAIModelFee.model_json_schema(),
extraction_type="schema",
instruction="""From the crawled content, extract all mentioned model names along with their fees for input and output tokens.

8
docs/examples/quickstart_async.py
View File

@@ -1,5 +1,7 @@
import os, sys
from crawl4ai.async_configs import LlmConfig
# append parent directory to system path
sys.path.append(
os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
@@ -145,8 +147,7 @@ async def extract_structured_data_using_llm(
url="https://openai.com/api/pricing/",
word_count_threshold=1,
extraction_strategy=LLMExtractionStrategy(
provider=provider,
api_token=api_token,
llmConfig=LlmConfig(provider=provider,api_token=api_token),
schema=OpenAIModelFee.model_json_schema(),
extraction_type="schema",
instruction="""From the crawled content, extract all mentioned model names along with their fees for input and output tokens.
@@ -569,8 +570,7 @@ async def generate_knowledge_graph():
relationships: List[Relationship]
extraction_strategy = LLMExtractionStrategy(
provider="openai/gpt-4o-mini", # Or any other provider, including Ollama and open source models
api_token=os.getenv("OPENAI_API_KEY"), # In case of Ollama just pass "no-token"
llmConfig=LlmConfig(provider="openai/gpt-4o-mini", api_token=os.getenv("OPENAI_API_KEY")), # In case of Ollama just pass "no-token"
schema=KnowledgeGraph.model_json_schema(),
extraction_type="schema",
instruction="""Extract entities and relationships from the given text.""",

9
docs/examples/quickstart_sync.py
View File

@@ -1,5 +1,6 @@
import os
import time
from crawl4ai.async_configs import LlmConfig
from crawl4ai.web_crawler import WebCrawler
from crawl4ai.chunking_strategy import *
from crawl4ai.extraction_strategy import *
@@ -178,7 +179,7 @@ def add_llm_extraction_strategy(crawler):
result = crawler.run(
url="https://www.nbcnews.com/business",
extraction_strategy=LLMExtractionStrategy(
provider="openai/gpt-4o", api_token=os.getenv("OPENAI_API_KEY")
llmConfig = LlmConfig(provider="openai/gpt-4o", api_token=os.getenv("OPENAI_API_KEY"))
),
)
cprint(
@@ -197,8 +198,7 @@ def add_llm_extraction_strategy(crawler):
result = crawler.run(
url="https://www.nbcnews.com/business",
extraction_strategy=LLMExtractionStrategy(
provider="openai/gpt-4o",
api_token=os.getenv("OPENAI_API_KEY"),
llmConfig=LlmConfig(provider="openai/gpt-4o",api_token=os.getenv("OPENAI_API_KEY")),
instruction="I am interested in only financial news",
),
)
@@ -210,8 +210,7 @@ def add_llm_extraction_strategy(crawler):
result = crawler.run(
url="https://www.nbcnews.com/business",
extraction_strategy=LLMExtractionStrategy(
provider="openai/gpt-4o",
api_token=os.getenv("OPENAI_API_KEY"),
llmConfig=LlmConfig(provider="openai/gpt-4o",api_token=os.getenv("OPENAI_API_KEY")),
instruction="Extract only content related to technology",
),
)

23
docs/md_v2/api/parameters.md
View File

@@ -245,11 +245,32 @@ run_config = CrawlerRunConfig(
)
```
## 3. Putting It All Together
# 3. **LlmConfig** - Setting up LLM providers
LlmConfig is useful to pass LLM provider config to strategies and functions that rely on LLMs to do extraction, filtering, schema generation etc. Currently it can be used in the following -
1. LLMExtractionStrategy
2. LLMContentFilter
3. JsonCssExtractionStrategy.generate_schema
4. JsonXPathExtractionStrategy.generate_schema
## 3.1 Parameters
| **Parameter** | **Type / Default** | **What It Does** |
|-----------------------|----------------------------------------|---------------------------------------------------------------------------------------------------------------------------------------|
| **`provider`** | `"ollama/llama3","groq/llama3-70b-8192","groq/llama3-8b-8192", "openai/gpt-4o-mini" ,"openai/gpt-4o","openai/o1-mini","openai/o1-preview","openai/o3-mini","openai/o3-mini-high","anthropic/claude-3-haiku-20240307","anthropic/claude-3-opus-20240229","anthropic/claude-3-sonnet-20240229","anthropic/claude-3-5-sonnet-20240620","gemini/gemini-pro","gemini/gemini-1.5-pro","gemini/gemini-2.0-flash","gemini/gemini-2.0-flash-exp","gemini/gemini-2.0-flash-lite-preview-02-05","deepseek/deepseek-chat"`<br/>*(default: `"openai/gpt-4o-mini"`)* | Which LLM provoder to use.
| **`api_token`** |1.Optional. When not provided explicitly, api_token will be read from environment variables based on provider. For example: If a gemini model is passed as provider then,`"GEMINI_API_KEY"` will be read from environment variables <br/> 2. API token of LLM provider <br/> eg: `api_token = "gsk_1ClHGGJ7Lpn4WGybR7vNWGdyb3FY7zXEw3SCiy0BAVM9lL8CQv"` <br/> 3. Environment variable - use with prefix "env:" <br/> eg:`api_token = "env: GROQ_API_KEY"` | API token to use for the given provider
| **`base_url`** |Optional. Custom API endpoint | If your provider has a custom endpoint
## 3.2 Example Usage
```python
llmConfig = LlmConfig(provider="openai/gpt-4o-mini", api_token=os.getenv("OPENAI_API_KEY"))
```
## 4. Putting It All Together
- **Use** `BrowserConfig` for **global** browser settings: engine, headless, proxy, user agent.
- **Use** `CrawlerRunConfig` for each crawl’s **context**: how to filter content, handle caching, wait for dynamic elements, or run JS.
- **Pass** both configs to `AsyncWebCrawler` (the `BrowserConfig`) and then to `arun()` (the `CrawlerRunConfig`).
- **Use** `LlmConfig` for LLM provider configurations that can be used across all extraction, filtering, schema generation tasks. Can be used in - `LLMExtractionStrategy`, `LLMContentFilter`, `JsonCssExtractionStrategy.generate_schema` & `JsonXPathExtractionStrategy.generate_schema`
```python
# Create a modified copy with the clone() method

6
docs/md_v2/api/strategies.md
View File

@@ -131,6 +131,7 @@ OverlappingWindowChunking(
```python
from pydantic import BaseModel
from crawl4ai.extraction_strategy import LLMExtractionStrategy
from crawl4ai.async_configs import LlmConfig
# Define schema
class Article(BaseModel):
@@ -140,7 +141,7 @@ class Article(BaseModel):
# Create strategy
strategy = LLMExtractionStrategy(
provider="ollama/llama2",
llmConfig = LlmConfig(provider="ollama/llama2"),
schema=Article.schema(),
instruction="Extract article details"
)
@@ -197,6 +198,7 @@ result = await crawler.arun(
```python
from crawl4ai.chunking_strategy import OverlappingWindowChunking
from crawl4ai.async_configs import LlmConfig
# Create chunking strategy
chunker = OverlappingWindowChunking(
@@ -206,7 +208,7 @@ chunker = OverlappingWindowChunking(
# Use with extraction strategy
strategy = LLMExtractionStrategy(
provider="ollama/llama2",
llmConfig = LlmConfig(provider="ollama/llama2"),
chunking_strategy=chunker
)

77
docs/md_v2/core/browser-crawler-config.md
View File

@@ -1,9 +1,10 @@
# Browser & Crawler Configuration (Quick Overview)
# Browser, Crawler & LLM Configuration (Quick Overview)
Crawl4AI’s flexibility stems from two key classes:
1. **`BrowserConfig`** – Dictates **how** the browser is launched and behaves (e.g., headless or visible, proxy, user agent).
2. **`CrawlerRunConfig`** – Dictates **how** each **crawl** operates (e.g., caching, extraction, timeouts, JavaScript code to run, etc.).
2. **`CrawlerRunConfig`** – Dictates **how** each **crawl** operates (e.g., caching, extraction, timeouts, JavaScript code to run, etc.).
3. **`LlmConfig`** - Dictates **how** LLM providers are configured. (model, api token, base url, temperature etc.)
In most examples, you create **one** `BrowserConfig` for the entire crawler session, then pass a **fresh** or re-used `CrawlerRunConfig` whenever you call `arun()`. This tutorial shows the most commonly used parameters. If you need advanced or rarely used fields, see the [Configuration Parameters](../api/parameters.md).
@@ -234,13 +235,37 @@ The `clone()` method:
---
## 3. Putting It All Together
In a typical scenario, you define **one** `BrowserConfig` for your crawler session, then create **one or more** `CrawlerRunConfig` depending on each call’s needs:
## 3. LlmConfig Essentials
### Key fields to note
1. **`provider`**:
- Which LLM provoder to use.
- Possible values are `"ollama/llama3","groq/llama3-70b-8192","groq/llama3-8b-8192", "openai/gpt-4o-mini" ,"openai/gpt-4o","openai/o1-mini","openai/o1-preview","openai/o3-mini","openai/o3-mini-high","anthropic/claude-3-haiku-20240307","anthropic/claude-3-opus-20240229","anthropic/claude-3-sonnet-20240229","anthropic/claude-3-5-sonnet-20240620","gemini/gemini-pro","gemini/gemini-1.5-pro","gemini/gemini-2.0-flash","gemini/gemini-2.0-flash-exp","gemini/gemini-2.0-flash-lite-preview-02-05","deepseek/deepseek-chat"`<br/>*(default: `"openai/gpt-4o-mini"`)*
2. **`api_token`**:
- Optional. When not provided explicitly, api_token will be read from environment variables based on provider. For example: If a gemini model is passed as provider then,`"GEMINI_API_KEY"` will be read from environment variables
- API token of LLM provider <br/> eg: `api_token = "gsk_1ClHGGJ7Lpn4WGybR7vNWGdyb3FY7zXEw3SCiy0BAVM9lL8CQv"`
- Environment variable - use with prefix "env:" <br/> eg:`api_token = "env: GROQ_API_KEY"`
3. **`base_url`**:
- If your provider has a custom endpoint
```python
llmConfig = LlmConfig(provider="openai/gpt-4o-mini", api_token=os.getenv("OPENAI_API_KEY"))
```
## 4. Putting It All Together
In a typical scenario, you define **one** `BrowserConfig` for your crawler session, then create **one or more** `CrawlerRunConfig` & `LlmConfig` depending on each call’s needs:
```python
import asyncio
from crawl4ai import AsyncWebCrawler, BrowserConfig, CrawlerRunConfig, CacheMode
from crawl4ai import AsyncWebCrawler, BrowserConfig, CrawlerRunConfig, CacheMode, LlmConfig
from crawl4ai.extraction_strategy import JsonCssExtractionStrategy
async def main():
@@ -262,8 +287,39 @@ async def main():
}
extraction = JsonCssExtractionStrategy(schema)
# 3) Crawler run config: skip cache, use extraction
# 3) Example LLM content filtering
gemini_config = LlmConfig(
provider="gemini/gemini-1.5-pro"
api_token = "env:GEMINI_API_TOKEN"
)
# Initialize LLM filter with specific instruction
filter = LLMContentFilter(
llmConfig=gemini_config, # or your preferred provider
instruction="""
Focus on extracting the core educational content.
Include:
- Key concepts and explanations
- Important code examples
- Essential technical details
Exclude:
- Navigation elements
- Sidebars
- Footer content
Format the output as clean markdown with proper code blocks and headers.
""",
chunk_token_threshold=500, # Adjust based on your needs
verbose=True
)
md_generator = DefaultMarkdownGenerator(
content_filter=filter,
options={"ignore_links": True}
# 4) Crawler run config: skip cache, use extraction
run_conf = CrawlerRunConfig(
markdown_generator=md_generator,
extraction_strategy=extraction,
cache_mode=CacheMode.BYPASS,
)
@@ -283,11 +339,11 @@ if __name__ == "__main__":
---
## 4. Next Steps
## 5. Next Steps
For a **detailed list** of available parameters (including advanced ones), see:
- [BrowserConfig and CrawlerRunConfig Reference](../api/parameters.md)
- [BrowserConfig, CrawlerRunConfig & LlmConfig Reference](../api/parameters.md)
You can explore topics like:
@@ -298,11 +354,12 @@ You can explore topics like:
---
## 5. Conclusion
## 6. Conclusion
**BrowserConfig** and **CrawlerRunConfig** give you straightforward ways to define:
**BrowserConfig**, **CrawlerRunConfig** and **LlmConfig** give you straightforward ways to define:
- **Which** browser to launch, how it should run, and any proxy or user agent needs.
- **How** each crawl should behave—caching, timeouts, JavaScript code, extraction strategies, etc.
- **Which** LLM provider to use, api token, temperature and base url for custom endpoints
Use them together for **clear, maintainable** code, and when you need more specialized behavior, check out the advanced parameters in the [reference docs](../api/parameters.md). Happy crawling!

5
docs/md_v2/core/content-selection.md
View File

@@ -211,7 +211,7 @@ if __name__ == "__main__":
import asyncio
import json
from pydantic import BaseModel, Field
from crawl4ai import AsyncWebCrawler, CrawlerRunConfig
from crawl4ai import AsyncWebCrawler, CrawlerRunConfig, LlmConfig
from crawl4ai.extraction_strategy import LLMExtractionStrategy
class ArticleData(BaseModel):
@@ -220,8 +220,7 @@ class ArticleData(BaseModel):
async def main():
llm_strategy = LLMExtractionStrategy(
provider="openai/gpt-4",
api_token="sk-YOUR_API_KEY",
llmConfig = LlmConfig(provider="openai/gpt-4",api_token="sk-YOUR_API_KEY")
schema=ArticleData.schema(),
extraction_type="schema",
instruction="Extract 'headline' and a short 'summary' from the content."

5
docs/md_v2/core/markdown-generation.md
View File

@@ -175,14 +175,13 @@ prune_filter = PruningContentFilter(
For intelligent content filtering and high-quality markdown generation, you can use the **LLMContentFilter**. This filter leverages LLMs to generate relevant markdown while preserving the original content's meaning and structure:
```python
from crawl4ai import AsyncWebCrawler, BrowserConfig, CrawlerRunConfig
from crawl4ai import AsyncWebCrawler, BrowserConfig, CrawlerRunConfig, LlmConfig
from crawl4ai.content_filter_strategy import LLMContentFilter
async def main():
# Initialize LLM filter with specific instruction
filter = LLMContentFilter(
provider="openai/gpt-4o", # or your preferred provider
api_token="your-api-token", # or use environment variable
llmConfig = LlmConfig(provider="openai/gpt-4o",api_token="your-api-token"), #or use environment variable
instruction="""
Focus on extracting the core educational content.
Include:

18
docs/md_v2/core/quickstart.md
View File

@@ -128,6 +128,7 @@ Crawl4AI can also extract structured data (JSON) using CSS or XPath selectors. B
```python
from crawl4ai.extraction_strategy import JsonCssExtractionStrategy
from crawl4ai.async_configs import LlmConfig
# Generate a schema (one-time cost)
html = "<div class='product'><h2>Gaming Laptop</h2><span class='price'>$999.99</span></div>"
@@ -135,15 +136,13 @@ html = "<div class='product'><h2>Gaming Laptop</h2><span class='price'>$999.99</
# Using OpenAI (requires API token)
schema = JsonCssExtractionStrategy.generate_schema(
html,
provider="openai/gpt-4o", # Default provider
api_token="your-openai-token" # Required for OpenAI
llmConfig = LlmConfig(provider="openai/gpt-4o",api_token="your-openai-token") # Required for OpenAI
)
# Or using Ollama (open source, no token needed)
schema = JsonCssExtractionStrategy.generate_schema(
html,
provider="ollama/llama3.3", # Open source alternative
api_token=None # Not needed for Ollama
llmConfig = LlmConfig(provider="ollama/llama3.3", api_token=None) # Not needed for Ollama
)
# Use the schema for fast, repeated extractions
@@ -212,7 +211,7 @@ import os
import json
import asyncio
from pydantic import BaseModel, Field
from crawl4ai import AsyncWebCrawler, CrawlerRunConfig
from crawl4ai import AsyncWebCrawler, CrawlerRunConfig, LlmConfig
from crawl4ai.extraction_strategy import LLMExtractionStrategy
class OpenAIModelFee(BaseModel):
@@ -242,8 +241,7 @@ async def extract_structured_data_using_llm(
word_count_threshold=1,
page_timeout=80000,
extraction_strategy=LLMExtractionStrategy(
provider=provider,
api_token=api_token,
llmConfig = LlmConfig(provider=provider,api_token=api_token),
schema=OpenAIModelFee.model_json_schema(),
extraction_type="schema",
instruction="""From the crawled content, extract all mentioned model names along with their fees for input and output tokens.
@@ -259,12 +257,6 @@ async def extract_structured_data_using_llm(
print(result.extracted_content)
if __name__ == "__main__":
# Use ollama with llama3.3
# asyncio.run(
# extract_structured_data_using_llm(
# provider="ollama/llama3.3", api_token="no-token"
# )
# )
asyncio.run(
extract_structured_data_using_llm(

10
docs/md_v2/extraction/llm-strategies.md
View File

@@ -71,8 +71,7 @@ Below is an overview of important LLM extraction parameters. All are typically s
```python
extraction_strategy = LLMExtractionStrategy(
provider="openai/gpt-4",
api_token="YOUR_OPENAI_KEY",
llmConfig = LlmConfig(provider="openai/gpt-4", api_token="YOUR_OPENAI_KEY"),
schema=MyModel.model_json_schema(),
extraction_type="schema",
instruction="Extract a list of items from the text with 'name' and 'price' fields.",
@@ -97,7 +96,7 @@ import asyncio
import json
from pydantic import BaseModel, Field
from typing import List
from crawl4ai import AsyncWebCrawler, BrowserConfig, CrawlerRunConfig, CacheMode
from crawl4ai import AsyncWebCrawler, BrowserConfig, CrawlerRunConfig, CacheMode, LlmConfig
from crawl4ai.extraction_strategy import LLMExtractionStrategy
class Product(BaseModel):
@@ -107,9 +106,8 @@ class Product(BaseModel):
async def main():
# 1. Define the LLM extraction strategy
llm_strategy = LLMExtractionStrategy(
provider="openai/gpt-4o-mini", # e.g. "ollama/llama2"
api_token=os.getenv('OPENAI_API_KEY'),
schema=Product.schema_json(), # Or use model_json_schema()
llmConfig = LlmConfig(provider="openai/gpt-4o-mini", api_token=os.getenv('OPENAI_API_KEY')),
schema=Product.schema_json(), # Or use model_json_schema()
extraction_type="schema",
instruction="Extract all product objects with 'name' and 'price' from the content.",
chunk_token_threshold=1000,

9
docs/md_v2/extraction/no-llm-strategies.md
View File

@@ -415,6 +415,7 @@ The schema generator is available as a static method on both `JsonCssExtractionS
```python
from crawl4ai.extraction_strategy import JsonCssExtractionStrategy, JsonXPathExtractionStrategy
from crawl4ai.async_configs import LlmConfig
# Sample HTML with product information
html = """
@@ -433,17 +434,15 @@ html = """
# Option 1: Using OpenAI (requires API token)
css_schema = JsonCssExtractionStrategy.generate_schema(
html,
schema_type="css", # This is the default
provider="openai/gpt-4o", # Default provider
api_token="your-openai-token" # Required for OpenAI
schema_type="css",
llmConfig = LlmConfig(provider="openai/gpt-4o",api_token="your-openai-token")
)
# Option 2: Using Ollama (open source, no token needed)
xpath_schema = JsonXPathExtractionStrategy.generate_schema(
html,
schema_type="xpath",
provider="ollama/llama3.3", # Open source alternative
api_token=None # Not needed for Ollama
llmConfig = LlmConfig(provider="ollama/llama3.3", api_token=None) # Not needed for Ollama
)
# Use the generated schema for fast, repeated extractions