qingy2024/Pristine-Mini-8B-A1B-Base
LoRA-style (fewer layers) vs inclusionAI/Ring-mini-2.0. qingy2024/Pristine-Mini-8B-A1B-Base matches part of inclusionAI/Ring-mini-2.0 (7479 of its 14813 tensors; the rest of inclusionAI/Ring-mini-2.0 has no counterpart). On the shared part: Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of dense, down_proj, gate_proj, query_key_value, up_proj changed, in 19 of 20 layers; norms, biases and embeddings are untouched.
Tensors changed: MLP 1393/1457, lm_head 0/1, Attention 5/40, Norms 0/81, Embeddings 0/1
1 derivatives checked
- mradermacher/Pristine-Mini-8B-A1B-Base-GGUF: Same weights, converted