LlamaParse error parsing MSFT DOCX and PPTX files - suspect LlamaParseReader is applying wrong MIME Type

[ajpanyteam]

Describe the bug:

LLamaParse is failing for PPTX and DOCX files. But works for PDF and XLSX files.

Using the llamaIndex TS SDK v0.6.17.

I suspect this is related to #1007

Write a concise description of what the bug is:

Using the following code returns “Error while parsing the file: Failed to parse the file:"

const file = storage.bucket(process.env.GCP_BUCKET).file(fileId);
const [buffer] = await file.download();

const getMetadata = await file.getMetadata();
console.log(getMetadata[0].contentType)

const reader = new LlamaParseReader({
   resultType: "markdown",
   skipDiagonalText: true,
   verbose: true,
});

const unt8Array = new Uint8Array(buffer); 
const documents = await reader.loadDataAsContent(unt8Array); // -> Error while parsing the file: Failed to parse the file: c8e8b079-f3aa-4786-bfe3-e9b3981812cf, status: ERROR

Additional context:

• Digging into the SDK, I added a breakpoint on LlamaParseReader function loadDataAsContent which led to createJob. In this screenshot, the PPTX file MimeType applied is application/vnd.oasis.opendocument.spreadsheet, this does not sound right.
  

Files:

• wget "https://meetings.wmo.int/Cg-19/PublishingImages/SitePages/FINAC-43/7%20-%20EC-77-Doc%205%20Financial%20Statements%20for%202022%20(FINAC).pptx" -O data/presentation.ppt
  ppx source file from llamaParse github

Job ID:

• ac0c1b4a-915f-4d00-9340-25506d085f33
• f8275764-361f-4fd9-ac6d-634193c556b2
• c8e8b079-f3aa-4786-bfe3-e9b3981812cf
• Above jobIds are from my console output. My https://cloud.llamaindex.ai/parse history tab does not have these jobs.

[danielbank]

For certain files, this example code does not fail as an error but instead produces a garbage parse based on data which is not at all related to the original file. See example repo for reproducing

[himself65]

It seems a bug in llama parse side, also I remove the file type detect in sdk side beacuse llama parse will handle that correctly

[jsmusgrave]

@himself65 I just validate docx upload via the website. So, assuming they're both using the same API, it seems unlikely this is on the llama parse side.