This release introduces a completely redesigned Hugging Face model downloader, live trending discovery, fast quantization filters, download resume support, and UI performance optimizations:
What’s Changed:
- Modern Visual Model Cards & Badges: Redesigned model browser list using custom two-row micro-cards with bold elided typography, publisher badges, download counts, like counts, and gated status indicators.
- Live Trending Discovery: Integrated Hugging Face Hub
trendingScorequery with automatic discovery on initial dialog open, ensuring active community models appear without manual queries. - Smart GGUF Filtering & Recommendations: Fast quantization category filters (
All,Q4 Recommended,Q5,Q8,Small Q2-Q3), instant file search box, and high-contrast recommended badges for balanced quantizations (Q4_K_M,Q5_K_M). - Seamless Download Workflow: Built-in Selection Summary Panel showing active model and file details, allowing instant downloads directly from the Search tab without unwanted automatic tab switching.
- Download Resume & Core Optimization: Large 1 MB chunk buffers and 150ms GUI progress signal throttling preventing UI thread lockups during gigabit downloads; HTTP
Rangeresume support via temporary.partfiles preserving partial downloads on cancel or interruption. - Gated Model & Token Support: Secure password-masked Hugging Face token input with show/hide toggle and private file permissions (
0600) saved to~/.llamatray/hf_token, allowing authorized exploration and download of gated models. - High-Contrast Theme Engine: Dynamic palette luminance detection (
is_dark_theme) adapting badges, status indicators, borders, and text colors for WCAG-compliant readability on dark and light desktop environments. - Comprehensive Test Suite: 28 targeted unit tests in
tests/test_hf_downloader.pyvalidating quant matching, formatting, token permissions, theme luminance, and local mock HTTP server Range resume and 401 handling.