Simplify voice_args: remove redundant fallback chain, require explicit name

Remove silence-generated default voice; use cloned voice as default when available

Generate default voice from silence on startup; fix voice fallback priority

Fix: remove duplicate voiceSel declaration breaking UI

Revert "Generate default voice from silence when no voices available"

Voice selector: inline popup with groups (builtin/unseen/manual), delete cloned voices, outside-click close

Generate default voice from silence when no voices available

Fix: voice clone mic null when not in call mode

Fix: create model/speaker dir on demand for voice clone

Fix voice clone, path defaults, README rewrite; add pydub dep

Remove hardcoded '请描述这张图片' from camera prompt

Revert "Remove '请描述这张图片' from camera prompt; img_tokens before user text"

Revert "Camera: inject scene text instead of raw img_tokens to avoid description mode"

Revert "Fix scene refresh starving conversation: non-blocking lock + SCENE_LOCK guard"

Fix scene refresh starving conversation: non-blocking lock + SCENE_LOCK guard

Camera: inject scene text instead of raw img_tokens to avoid description mode

Remove '请描述这张图片' from camera prompt; img_tokens before user text

Revert camera prompt changes; add NaN guard and CUDA sync; add engineering challenges doc

Add suppression prefix before image tokens

Camera context improvements: image tokens before user text + system prompt

Remove hardcoded '请描述这张图片'; image tokens alone suffice

Fix camera: don't force 'describe this image' when user has text input

Move audio/vision encoder .to(device) after model init for both HF and .pth paths

Default to HF format loading in omni_o_call.py

Convert omni-o.pth to HF format + add from_pretrained support

Replace web_demo.html with omni_o_web.html: self-contained frontend for omni-o call

Rewrite omni_o_terminal.py: add real-time interrupt/barge-in

Add omni_o_terminal.py: terminal voice chat script

Add omni_o_call.py: real-time voice call server for omni-o checkpoint

Fix eval_vam.py: update paths to checkpoint/ dirs, fix VAMConfig attn heads, remove open_thinking kwarg

feat: add interactive mode to eval_vlm.py with auto <image> insertion, update README

feat: add VLM SFT mini configs, resume training, dataset streaming, and eval fixes

rename minimind→omni in model_type, eval/convert scripts, and docs

fix eval_llm: add --tokenizer_path/--native flags, auto-detect num_hidden_layers from checkpoint; update README with correct inference commands

Flatten src layout: drop omni namespace, modules live directly under src/

Restructure repo: rename classes, split datasets, organize trainers by modality, add YAML config + root trainer entry points