Disable the debugging mode.

Update the deps.

Load the tokenizer right away.

Change the default model.

Remove useless import and arguments.

Remove the logging of imports.

Upgrade psaiops, which hopefully fixes the space issues.

Log each successful import.

Add a simple logging isntruction.

Relax the deps constraints even further.

TIdy up.

Try Qwen 3.5...

Try using gemma 3 instead.

Try setting the default config explicitely... Desperate tests...

Fix typos.

Retry the community kernels.

Enable the debugging mode.

Use the same configuration when frontloading the model.

Try the flex attention...

Next, try setting everything to auto...

Try solving the CUDA errors with dtype float16...

Try to solve the CUDA errors by moving back to SDPA attention.

Disable the mxfp4 quantization.

Attempt to fix the dtype to solve the CUDA issues.

Update the deps to solve the issues with the class `GptOssForCausalLM`.

Update deps.

Do not access the GPU outside of the spaces wrapper.

Load custom CSS styles to give a visual feedback on button click.

Fix typo

Add a factory for the tabulation

Fix dep conflicts + decorate computation with GPU handler + use latest APIs

Fix missing import

Use the latest gradio + psaiops APIs

Use the latest versions

Use the psai-ops boilerplate with the matching app

initial commit