Markdown/Plain Text Output ?
#3
by lokit2356 - opened
I tested the model using llama server and the response is always in HTML. How do i switch to Markdown or Plain Text output?
./llama-server \
-m /apps/models/chandra-gguf/chandra-Q8_0.gguf \
--mmproj /apps/models/chandra-gguf/chandra-mmproj-f16.gguf \
--device none \
-ngl 0 \
--no-mmproj-offload \
-t 40 \
-tb 40 \
--numa distribute \
-c 8192 \
-b 1024 \
-ub 512 \
-np 1 \
--mmap \
--mlock \
--no-warmup \
--log-timestamps \
--verbose \
--host 0.0.0.0 \
--port 9000
@lokit2356
In a way like that, Chandra always provides the output in HTML format, such as HTML markdowns, OCR, and JSON, for more accurate reconstruction of the input document page.
Thanks. I just discovered that there is a Markdown parser in their output module of chandra's official Python package. This could imply that the model only outputs HTML, and the Markdown is simply a parsed representation of that HTML.