wyrdsekai/companion-3.5-4b

wyrdsekai companion 3.5-4B v10 (full precision)

Full-precision safetensors of the wyrdsekai voice model, v10 — the source the deployment conversions come from: GGUF Q4_K_M for llama.cpp, MLX 4-bit for Apple Silicon. Published for re-quantization to other formats, further fine-tuning, and full-precision evaluation.

See the GGUF card for intended use and limitations. Training data: synthetic corpora only — no end-user or household conversation data. SFT corpus for the drive line is published at wyrdsekai/drive-sft-corpus.