Font Parsing Error in Llama Parse API Causing Garbled Output

[seymaalan]

Describe the bug

When parsing specific CV files (one 3 pages and another 2 pages) in Turkish and English using the Llama Parse API, the output contains gibberish, resembling Chinese characters. The issue seems related to font identification errors, as shown in the job logs. This bug has resulted in incorrect parsing, consuming significant credits without producing usable results.

Files

The PDF files contain private personal information and cannot be shared directly through this platform. However, I can email them to the bugfix team if necessary.

Job ID

The following job IDs are associated with this issue:

• 85114183-dd6c-4eb2-aa55-7f5fde82d30a
• 601cec22-e659-4f74-b1c1-62535b632ffd
• 9fa6155a-aacf-44e3-8ae3-030f20454d49
• bec57eb5-60f8-4de4-ab25-89757b2fea85

Note that, they consumed significant credits:

• 85114183-dd6c-4eb2-aa55-7f5fde82d30a - 45 credits
• 601cec22-e659-4f74-b1c1-62535b632ffd - 82 credits
• 9fa6155a-aacf-44e3-8ae3-030f20454d49 - 45 credits
• bec57eb5-60f8-4de4-ab25-89757b2fea85 - 82 credits

Client

• Frontend
• API

Additional context

The parsing requests were made using the Python API in “accurate mode,” with no further custom settings or options. The issue resulted in high credit consumption (254 credits in total), and a refund is requested for the affected credits.

The logs indicate that the problem is likely caused by unrecognized fonts:

Page Errors and Warning details:

• Page 1: Fail to identify 45 glyphs from fonts such as FreeMono, TeXGyreSchola-Regular, OpenSymbol, DejaVuSans, etc.
• Page 2: Fail to identify 81 glyphs, with similar fonts involved.
• Additional pages exhibit similar font identification errors.

Here is a part of job logs, it continues similar.
Job logs Page Errors and Warning details Page 1 [error] - FONT_ERROR : Fail to identify 45 glyphs on page 1 from font : GAAAAA+FreeMono, IAAAAA+TeXGyreSchola-Regular, JAAAAA+OpenSymbol, HAAAAA+DejaVuSans, DAAAAA+FreeSans, FAAAAA+FreeSerif, NAAAAA+NimbusMonoPSRegular, UAAAAA+C059Roman, VAAAAA+Purisa, WAAAAA+TeXGyreCursor-Regular Page 2 [error] - FONT_ERROR : Fail to identify 81 glyphs on page 2 from font : NAAAAA+NimbusMonoPSRegular, FAAAAA+FreeSerif, GAAAAA+FreeMono, YAAAAA+DroidSansFallback, IAAAAA+TeXGyreSchola-Regular, JAAAAA+OpenSymbol, VAAAAA+Purisa, WAAAAA+TeXGyreCursor-Regular, HAAAAA+DejaVuSans, ABAAAA+Gentium, DAAAAA+FreeSans, EBAAAA+FreeSerif Page 3 [error] - FONT_ERROR : Fail to identify 31 glyphs on page 3 from font : IAAAAA+TeXGyreSchola-Regular, DAAAAA+FreeSans, JAAAAA+OpenSymbol, EBAAAA+FreeSerif, NAAAAA+NimbusMonoPSRegular, HAAAAA+DejaVuSans, WAAAAA+TeXGyreCursor-Regular Page 4 [error] - FONT_ERROR : Fail to identify 70 glyphs on page 4 from font : HAAAAA+DejaVuSans, IAAAAA+TeXGyreSchola-Regular, JAAAAA+OpenSymbol, VAAAAA+Purisa, EBAAAA+FreeSerif, DAAAAA+FreeSans, PBAAAA+FreeSans, WAAAAA+TeXGyreCursor-Regular, NAAAAA+NimbusMonoPSRegular Page 5 [error] - FONT_ERROR : Fail to identify 75 glyphs on page 5 from font : IAAAAA+TeXGyreSchola-Regular, WAAAAA+TeXGyreCursor-Regular, JAAAAA+OpenSymbol, WBAAAA+LMSansQuot8-Regular, HAAAAA+DejaVuSans, PBAAAA+FreeSans, NAAAAA+NimbusMonoPSRegular, XBAAAA+DejaVuSans, EBAAAA+FreeSerif, DAAAAA+FreeSans, YAAAAA+DroidSansFallback, VAAAAA+Purisa ...

Here is an excerpt of the garbled output:

坥手楴䙯牭䉯畮摡特硃慌䵸佐䅧䱳丸畓 ഊ 䍯湴敮琭䑩獰潳楴楯渺潲洭摡瑡㬠湡浥㴢晩汥∻楬敮慭攽 汯戢 ≢ ഊ 䍯湴敮琭呹灥㨠慰灬楣慴楯港灤昍ਥ偄䘭㜊 㠠潢樊㰼 呹灥 偡来 偡牥湴‱ ⁒䱡獴䵯摩晩敤䐺㔱 ㌧ 剥 ◢ ‰ ⬰ j 獯畲捥猠㈠删敤楡䉯砠嬰 㤵㜶㠴㠹崠牯灂潸 ⸰ 㔹⁛ ‵㔮‰ 㘰 ⸰ ‸ 㐱㤰そ 䉬敥摂潸 㔹 ⸰ ⁛ ‰ 㔮㘰 㐱㤰そ 呲業䉯 ‸ 砠嬰 㤵㜶㠴㠹崠牴䉯砠嬰㤵㜶㠴㠹崠潮瑥湴猠㤠删潴慴攠牯異 ⸰ 㰠⸰ 祰攠‼ ‵ ‵牯異 匠牡湳灡牥湣 礠匠敶楣敒䝂 㸠娠ㄠ㸾ⴭ湤潢樊㤠潢樊㰼楬瑥爠污瑥䑥捯摥 䱥湧瑨 㸠獴牥慭ⴭ鳭巍狣㘒蚮纈 ൯ 魔礰 рრ▵盬碓 ‾ ″ 剉癊轆㼞䮲ⴭ ⧷ ꓑ)傦摊陧䖙麢 Ӿ"㹴翨渴㤡锪蒿 埸㪅洏跭㱠ᄍ㠺 ୶ 輶 ᕙ 敶 蟟铽 ៌举 ⋌ ⴭⴭ绡 ⴴ 䥹偐 അ 菁 ⪌ 句 ⴭ厂猝冝怭 ᙳ ⴭ ⴭⴭⴭⴭⴭḸ 㮵欋喚 ㆀⴭⴭ 坹諆 ⴭ ⴭ ⴭ ⴭⴭⴭⴭ ⴭⴭこ哈䡨 ⴭ蘯䇲 碗䣜ⴭ揁 虋醕獹䓘㝊舷夏ⴭ 珣銓 ⴭ ⴭ 񣐀 񣐀 񣐀 幻㠏 ᘿ 骣

This bug has severely impacted our workflow, causing wasted credits and time. We request an investigation into this issue and a resolution to avoid similar problems in the future.