Modelpublic

glebkudr/Infinity-Parser2-Pro-Q8-GGUF

sourceHugging Faceapache-2.0updated 4mo agoView on Hugging Face

1likes560downloads

Model Card

Infinity-Parser2-Pro Q8_0 GGUF

This repository contains a local Q8_0 GGUF conversion of infly/Infinity-Parser2-Pro.

If you are looking for a reliable OCR API use sotaocr.com

Files

File	Size	SHA256
`Infinity-Parser2-Pro-Q8_0.gguf`	36,903,139,968 bytes	`4effd023d153254d676832df662da2613e0266f973b8208ed01a044b0a82a3fc`
`Infinity-Parser2-Pro-mmproj-Q8_0.gguf`	614,194,688 bytes	`2a4007556c20769f1f873fb1e02585988dec62e46822a5295afb3df075185637`
`Infinity-Parser2-Pro-mmproj-bf16.gguf`	902,822,528 bytes	`c94c47e850a5cfb59f7baceba397416eea9b8da38e3e76f1dd83271a866b8269`

Recommended projector

Use Infinity-Parser2-Pro-mmproj-bf16.gguf for document parsing and OCR tasks. It preserves the upstream BF16 vision projector weights more closely than the Q8_0 projector while adding only a small amount of storage overhead.

Source

—Source model: infly/Infinity-Parser2-Pro
—Source snapshot commit used for conversion: 1d070df7db5acca0ffa75596229070a047704f89
—Conversion toolchain: llama.cpp, local build from commit 5aa3a6459
—Main model quantization: Q8_0
—Recommended multimodal projector type: BF16

The main model was converted with MTP disabled for compatibility with the current GGUF loader metadata (qwen35moe.block_count=40).

Smoke Test

The converted model and multimodal projector were smoke-tested locally with llama-cli by loading both GGUF files and running a one-token image prompt.

License

The upstream model card states that infly/Infinity-Parser2-Pro is licensed under Apache-2.0. This converted GGUF distribution follows the same license.