v0.0.6a

[0.0.6a] - 2026-04-08

Added

• Model Inventory panel on the HF Download tab, auto-loading on tab open
• Inventory table with columns: Model, Dtype, Params, Arch, Size, Reasoning, Script, Modalities
• Dtype detection from config.json (torch_dtype, quantization_config) with name-token fallback covering FP4, FP8, BF16, FP16, FP32, INT4, INT8, GGUF
• MoE detection via config keys (num_experts, num_experts_per_tok), known model_type values, architecture string, and name-token fallback including XB-AYB active-param notation
• Reasoning detection via is_thinking config key and name-token fallback (r1, qwq, thinking, cot, o1, thinker, reflect, deepthink)
• Modality detection (Text, Image, Audio, Embedding) from config keys (vision_config, audio_config), architecture string, and name-token fallback
• Parameter count from safetensors/PyTorch index metadata with name-regex fallback
• Script cross-reference against ~/SGLang/ and ~/vLLM/ launch scripts; shows engine badge if found
• Custom directory manager — add extra scan paths, shown as separate inventory tables
• Auto-tracking of custom local_dir paths used during downloads
• GET /api/hf/inventory — scans default HF cache and all saved custom directories
• POST /api/hf/inventory/dirs — add a custom scan directory
• DELETE /api/hf/inventory/dirs — remove a custom scan directory