tencent/KaLM-Embedding-Gemma3-12B-2511
LoRA-style (fewer layers) vs google/gemma-3-12b-pt. tencent/KaLM-Embedding-Gemma3-12B-2511 matches part of google/gemma-3-12b-pt (626 of its 1065 tensors; the rest of google/gemma-3-12b-pt 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, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 48 of 48 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/289, MLP 144/144, Attention 192/192
3 derivatives checked
- mradermacher/KaLM-Embedding-Gemma3-12B-2511-GGUF: Quantization
- mradermacher/KaLM-Embedding-Gemma3-12B-2511-i1-GGUF: Quantization
- thedarkstar/KaLM-Embedding-Gemma3-12B-2511-mlx-4Bit: Quantization
1 other repos hold byte-for-byte identical weight files.