[Bug]: The LLM strategy always sends all tokens from the URL to the LLM server even the URL input is HTML content

[phamngocquy]

crawl4ai version

0.6.3

Expected Behavior

my example crawler:

llm_strategy = LLMExtractionStrategy(
    llm_config=self.llm_config,
    schema=PdfDoc.model_json_schema(),
    extraction_type="schema",
    instruction="""
    From the crawled content, extract data from html
    - data in html source include pdf file name and href url under data-pdf style attribute
    - One extracted post JSON format should look like this:

        {
            "name": "Volume 1 - 2024 CMD Rate Case - E-filing.pdf",
            "data_pdf": "/DMS/pdfview/``wportal`Documents`DMS`31`2494`/Volume 1 - 2024 CMD Rate Case - E-filing~pdf",
        }
    """,
    input_format="cleaned_html",
)

run_conf = CrawlerRunConfig(
    extraction_strategy=llm_strategy,
    cache_mode=CacheMode.DISABLED,
    target_elements=[
        ".btnOpenPdfFile",
        ".btn",
        ".btn-default",
        ".btn-sm",
    ],
    excluded_tags=[
        "header",
        "footer",
        "nav",
        "meta",
        "script",
        "style",
        "iframe",
        "li",
        "ul",
    ],
    prettiify=True,
)


resp = requests.get(url)
async with AsyncWebCrawler(config=self.browser_conf) as crawler:
    result = await crawler.arun(url=f"raw://{resp.text}", config=run_conf)
    assert isinstance(result, CrawlResultContainer)

Expected: only cleaned_html should be sent to LLM server

Current Behavior

The behavior of llm extract strategy is send both url and html content to llm server. I think we doesn't handle the case when url input as raw html. That leading to full raw html under url always sending to llm server. It's likely unexpected behavior, the token quota may leak due to it.

Is this reproducible?

Yes

Inputs Causing the Bug

Steps to Reproduce

Code snippets

OS

Linux

Python version

3.12.3

Browser

No response

Browser version

No response

Error logs & Screenshots (if applicable)

No response