Hugging Face's logo

Spaces:
fhueni
/
on-device-vs-cloud-llm-inference
Configuration error

App Files Community
1
on-device-vs-cloud-llm-inference / src

Commit History

fix: fixed qwen3 check to check first if service.getModelName() is defined
b3e6f6c

nicolasfwyss commited on

fix: now the csv parsing ignores comma delimiters if they are inside quotes, to ensure all context is parsed correctly
1901675

Philip Kehl commited on

fix: minor typos
024b1f0

Philip Kehl commited on

Merge remote-tracking branch 'origin/main'
be04135

fhueni commited on

feat: remove f1 from csv download
4fdda35

fhueni commited on

feat: changed maxTokens for cloud and on-device from 500 to 50
72519d5

Philip Kehl commited on

Merge branch 'main' of github.com:fabianHueni/UNIBE_MSGAI_HS25_Project
6b11ecf

Philip Kehl commited on

feat: disable thinking for qwen model
b9991b1

Philip Kehl commited on

feat: add dataset item id to logs and csv result exports
6c50dd9

fhueni commited on

feat: enhance download of csv after experiment completion
b7f03a5

fhueni commited on

fix: re-add reload dataset method to scheduler
8d0c95c

fhueni commited on

chore: added comments
0dd78af

Philip Kehl commited on

feat: added stripping method for thinking tokens, increased maxNewTokens to 500, set enableThinking to false in onDeviceService
e05ac95

Philip Kehl commited on

fix: made csv parsing more robust, changed prompt
b550d9a

Philip Kehl commited on

feat: wait until all items of experiment are run to download stats, split downloads in stats and raw data
fe2aa13

fhueni commited on

chore: renamed interarrival time method
cda31a7

fhueni commited on

feat: add granit models and finish model selection
1d22bd8

fhueni commited on

feat: add option to specify model quantization in on device model selection
383a122

fhueni commited on

Merge branch 'main' of https://github.com/fabianHueni/UNIBE_MSGAI_HS25_Project
8f96daa

Nicolas Wyss commited on

feat: added a button to start experiments with 1000 requests
7937989

Nicolas Wyss commited on

feat: add queues for on device and cloud processing to measure queue time
94662d5

fhueni commited on

feat: track queueing time, inference time, and total latency separately. the timestamp for start inference should be taken somewhere else
ebf4772

Philip Kehl commited on

fix: changed prompt
eb00e61

Nicolas Wyss commited on

Merge branch 'main' of https://github.com/fabianHueni/UNIBE_MSGAI_HS25_Project
b06cf72

Nicolas Wyss commited on

fix: changed the prompt to include the passage/context as well and moved the full prompt to scheduler.js file
e4a0d67

Nicolas Wyss commited on

feat: added additional models
7f224da

fhueni commited on

fix: correct shown device stats and add message template to on device service
8b3b676

fhueni commited on

feat: implement CSV download and converted logs to a table
d720b4b

fhueni commited on

feat: optimize console logging and fix import of transfromer.js library
7a38188

fhueni commited on

feat: fix prediction output object for evaluation
b92a978

fhueni commited on

feat: implement OpenRouter as cloud model provider, optimize UI, fix some issues
9674cf0

fhueni commited on

feat: model download progress bar and added metrics f1WordLevel, Rouge1, TotalTokens, TokensPerSecond
b1ed689

Philip Kehl commited on

feat: add on device model parameter and logging outputs
6ac6471

fhueni commited on

xenova transformers implemented
f8c42c9

Philip Kehl commited on

feat: add dataset, adjust scheduler to run full dataset, implement statistic download, add model load button
5c8fb6b

fhueni commited on

feat: setup base code structure for pipeline with todos
bfaf968

fhueni commited on