Cerberus149/remembrance-soundscapes

Keep GPU duration at 60s for both functions

Restore GPU durations (240s/180s) for paid GPU tier

Update README.md

Reduce @spaces.GPU duration to 60s — ZeroGPU max exceeded with 240s/180s

Add error logging to _pipeline_inner Step 1 to expose GPU allocation failures

Remove gr.Progress (was blocking pipeline) + add step logging

Update repo references: sonic-runs-assets → remembrance-soundscapes

Add gr.Progress phase updates to pipeline

Retry upload_to_github on 409 SHA mismatch

Remove archive tab from app.py

switching to zerogpu

Fix README — move switch instructions out of frontmatter (YAML comments not supported by HF)

Rewrite README to reflect current state of the project

Add inline switch instructions to README frontmatter

Make switching modes a single README edit

Add one-command switch scripts for ZeroGPU vs paid GPU

Bump gradio sdk_version to 5.25.0 to fix boolean JSON schema bug

Fix gradio HfFolder import error

Pin torchaudio==2.9.1 to match HF base torch, drop unpinned torch

Remove flash-attn — cannot build without torch pre-installed in Gradio SDK

Switch to Gradio SDK for ZeroGPU support

Use Copenhagen timezone (Europe/Copenhagen) for timestamps

Fix counter: derive next ID from records.csv instead of counter.txt

Increase GPU timeout: vision 60s→240s, audio 60s→180s

Add pending.json tracking and explicit queue for multi-user support

Bind Gradio to 0.0.0.0 so HF Space can reach it

Remove app_file from Docker SDK space frontmatter

Remove conflicting flash-attn wheel from requirements.txt

fix: add pre-built flash-attn wheel to avoid source compilation

fix: remove pandas>=2.1.0 pin — not needed on Python 3.10, conflicts with stable-audio-tools

fix: install packaging + ninja before flash-attn build

fix: use nvidia/cuda:12.1.0-cudnn8-devel-ubuntu22.04 (valid tag)

fix: Dockerfile for persistent GPU — install torch before flash-attn

fix: add setuptools + pandas>=2.1.0 for Python 3.13 persistent GPU build

Reduce GPU quota: duration 120→60s, steps 150→100

Rewrite archive to read from records.csv instead of local files

Sync records.csv and counter.txt from GitHub on startup

Replace torchaudio.save with soundfile; remove torchcodec dependency

Fix torchaudio.save: force soundfile backend (torchcodec/FFmpeg unavailable)

Use local ./sonic_runs instead of /data for storage

Remove analysis prompt input from image-to-soundscape interface

Add customizable analysis prompts for image metadata generation

Refactor archive table: improve sorting logic and enhance data handling for better integration with DataFrame

Refactor archive management: enhance file listing with a sortable table and selection functionality

Enhance file management: add filtering options for file listing and update related functions for improved usability

Refactor archive_choices: update to return list of (label, value) tuples for checkbox display

Refactor archive_choices and delete_selected_files: update to return relative paths and improve path handling

Refactor archive management: streamline file handling and update UI interactions for better usability

Implement model caching and update file management UI for improved performance

Enhance file management UI: update functions to return checkbox states for file selection