lemuralabs/lemura-arabic-asr-qwen3
LoRA-style vs Qwen/Qwen3-ASR-1.7B. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 28 of 28 layers; norms, biases and embeddings are untouched.
Tensors changed: other 0/11, MLP 84/180, Norms 0/211, Attention 112/304, Embeddings 0/1
1 derivatives checked
- mradermacher/lemura-arabic-asr-qwen3-GGUF: Same weights, converted (fewer layers)