mradermacher/GhostShell-4B-GGUF

Same weights, converted (fewer layers) vs DuoNeural/GhostShell-4B. mradermacher/GhostShell-4B-GGUF matches part of DuoNeural/GhostShell-4B (623 of its 1148 tensors; the rest of DuoNeural/GhostShell-4B has no counterpart). On the shared part: Same weights stored in another dtype (BF16 → F16/BF16), plus 43 new tensors.

Tensors changed: Embeddings 0/2, Norms 0/278, MLP 0/126, other 0/85, Attention 0/132

On the Hub · Downloads on Model Pulse