AMAImedia/NOESIS-Qwen3-TTS-1.7B-Talker-Combo-LLM2-ASR2-BF16
LoRA-style (fewer layers) vs FINAL-Bench/Darwin-TTS-1.7B-Cross. AMAImedia/NOESIS-Qwen3-TTS-1.7B-Talker-Combo-LLM2-ASR2-BF16 matches part of FINAL-Bench/Darwin-TTS-1.7B-Cross (480 of its 976 tensors; the rest of FINAL-Bench/Darwin-TTS-1.7B-Cross has no counterpart). On the shared part: Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, up_proj changed, in 28 of 28 layers; norms, biases and embeddings are untouched.
Tensors changed: other 0/79, lm_head 0/15, Embeddings 0/17, Norms 0/134, MLP 84/103, Attention 0/132
1 derivatives checked
- mradermacher/NOESIS-Qwen3-TTS-1.7B-Talker-Combo-LLM2-ASR2-BF16-GGUF: Same weights, converted (fewer layers)