2 Commits

Include renames

Author	SHA1	Message	Date
filliptmandClaude Opus 4.6	9c0c7a2375	Fix weight loading, GQA, codec decoder, and tokenizer for Voxtral TTS ... Major fixes: - Embeddings: use actual checkpoint size (9088 entries with padded codebooks) - GQA: fix KV head repetition dimension (was interleaving wrong, corrupting all attention computations in backbone and acoustic transformer) - Codec decoder: restructure to flat decoder_blocks[0-7] matching checkpoint, use nn.utils.parametrizations.weight_norm, disable conv bias, fix dtype - Tokenizer: use official mistral_common SpeechRequest API for correct prompt format (BOS, BEGIN_AUDIO, AUDIO×N, /INST, text, INST, BEGIN_AUDIO) - Pipeline: add initial AUDIO token feed before generation loop, fix voice embedding injection positions - Loader: handle folder_paths KeyError gracefully Status: model loads, generates coherent speech on MPS, codec decodes to 24kHz audio. Text conditioning not yet working correctly — model produces plausible speech but doesn't follow the input text. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	2026-03-26 21:12:10 -05:00
filliptmandClaude Opus 4.6	60f6897263	Initial implementation of Voxtral-4B TTS direct inference for ComfyUI ... Custom node pack implementing Mistral's Voxtral-4B-TTS-2603 model with direct PyTorch inference (no vLLM dependency), enabling native MPS/Mac support alongside CUDA. Three-component architecture: - Mistral LLM backbone (26 layers, 3072 dim, GQA, RoPE) - Flow-matching acoustic transformer (3 layers, 8-step Euler ODE, CFG) - Neural audio codec decoder (4-stage ALiBi transformer + conv upsampling) Features: - 20 preset voices across 9 languages - 24 kHz mono output - BF16/FP16/FP32 dtype support - HuggingFace Hub auto-download with caching - ComfyUI AUDIO output format Nodes: - FL_VoxtralTTS_ModelLoader: download and load model - FL_VoxtralTTS_Generate: text + voice → audio Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	2026-03-26 20:09:20 -05:00