LucidityAI/Astral-4B
LoRA-style vs Qwen/Qwen3-4B. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 36 of 36 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/145, MLP 72/108, Attention 144/144
3 derivatives checked
- mradermacher/Astral-4B-GGUF: Same weights, converted
- mradermacher/Astral-4B-i1-GGUF: Quantization
- LucidityAI/Astral-4B-Coder: LoRA-style