mradermacher/gemma-4-E4B-it-SDFT_Heretic_RP-GGUF

Same weights, converted (fewer layers) vs Ilya626/gemma-4-E4B-it-SDFT_Heretic_RP. mradermacher/gemma-4-E4B-it-SDFT_Heretic_RP-GGUF matches part of Ilya626/gemma-4-E4B-it-SDFT_Heretic_RP (677 of its 2130 tensors; the rest of Ilya626/gemma-4-E4B-it-SDFT_Heretic_RP has no counterpart). On the shared part: Same weights stored in another dtype (F16 → F16/F32), plus 43 new tensors.

Tensors changed: Embeddings 0/2, Norms 0/296, MLP 0/126, other 0/85, Attention 0/168

On the Hub · Downloads on Model Pulse