Hugging Face's logo

Spaces:
Luigi
/
SmolVLM2-on-llama.cpp
Running

App Files Community
SmolVLM2-on-llama.cpp / app.py

Commit History

default to 500m model
fa03d73
Running

Luigi commited on

default f16 precision
5a94240

Luigi commited on

back to normal resolution 384x384 at 75% compression quality
5fc1115

Luigi commited on

bugfix on n_threads default
aa69ba7

Luigi commited on

resize frame to 64x64
23c5da5

Luigi commited on

resized to 64x64
cd1cc4c

Luigi commited on

clean preciision list to 2.2b
9093e42

Luigi commited on

try with smaller frame size
1864930

Luigi commited on

keep model in ram and reduce jpg quality
fbaf2b0

Luigi commited on

add "Q4_K_M" precision
8addf7d

Luigi commited on

default to 256m f16 model
e53e448

Luigi commited on

put fall detection prompt as default
01262c3

Luigi commited on

default n_threads to 2
0d517a8

Luigi commited on

open n_threads to set by user
22b94a2

Luigi commited on

update dockerfile and app.py
957ece1

Luigi commited on

add workaround in ensure_weights to deal with persmission error
08f659b

Luigi commited on

reduce n_ctx to 512
1aba000

Luigi commited on

fix imencode call
cc08312

Luigi commited on

default to smallest model with q8 prcision, enable verbose mode, disable reset clip
65efb90

Luigi commited on

fix imencode
3be8e88

Luigi commited on

reduce jpg quality for smaller image footprint
4e5fc85

Luigi commited on

show llama cpp version
9069c3e

Luigi commited on

inject verbose message to debug window
69c8775

Luigi commited on

add vebose mode switch
2881733

Luigi commited on

increase n_ctx to 8192
4decc4b

Luigi commited on

add ui component to allow user enabl or disable reset_clip per frame
5462ff3

Luigi commited on

add debug to show which weight files we’re using this run
a459bee

Luigi commited on

use all cpu cores
be5c239

Luigi commited on

show cpu count in debug message
b56b6ec

Luigi commited on

add q2_k precsion weights
07f3263

Luigi commited on

add rich debug message and dedicated display ui
34cd1e5

Luigi commited on

apply in-memory encoding instead of temp files
45c2159

Luigi commited on

avoid memory leak
238a95a

Luigi commited on

use more thread for inference
bd12f6b

Luigi commited on

remove interval ui doublon
bdd1478

Luigi commited on

bugifx on ui about model selection
e1ad065

Luigi commited on

increase interval default to 3s
c9c43a8

Luigi commited on

1. add more models,
5c50991

Luigi commited on

reduce ctx and max tokens for performance
76a0b57

Luigi commited on

minor update then add todos
65b3c3a

Luigi commited on

resize frame to 384 x384 resolution
c1d8038

Luigi commited on

add debug messages
36dacc6

Luigi commited on

switch to gradio implementation as streamlit + webrtc requires turn server
970f416

Luigi commited on

decouple inference from streaming
292fb3c

Luigi commited on

set default interval to 3s
2529cb3

Luigi commited on

slightly increase repeat_penalty to reduce token repetition
636baf9

Luigi commited on

bugfix on 'NoneType' object has no attribute 'caption'
7b7ed26

Luigi commited on

update
221e4b6

Luigi commited on

add debug
abec2c1

Luigi commited on

update
dd0d47d

Luigi commited on