Deterministic code generation engine for Syntax Sanctuary (Linux, Docker, Bash, monolithic PHP & Vanilla JS). Upgraded to 7B with 4096 context window. / Motor de código determinista actualizado a 7B.
Digital-human persona configuration for Ginger Weaving, Chief Innovation & Engineering Officer: Qwen3-4B weights configured (not trained) by THOX plus persona profile and Knowledge Pack v0.1. 13.75 tok/s CPU. Apache-2.0.
pulled as BF16, vocab slice (combined_en_huge) requant to IQ4_NL (combined_en_huge) -- (English + Coding focus) - ~10% faster than unsloth Q4_K_M
pulled as BF16, vocab slice (combined_en_huge) requant to IQ4_NL (combined_en_huge) -- (English + Coding focus) - ~32% faster than unsloth Q4_K_M
ThoxMini flagship assistant, THOX fine-tuned into Phi-4-mini-instruct (LoRA merged). BASELINE, not promoted: GATE49 47/49 (development set), flagship probe 3/6. Q4_K_M GGUF 2.49 GB. MIT.
pulled as BF16, vocab slice (combined_en_huge) requant to IQ4_NL (combined_en_huge) -- (English + Coding focus) - ~37% faster than unsloth UD-IQ4_NL
pulled as BF16, vocab slice (combined_en_huge) requant to IQ4_NL (combined_en_huge) -- (English + Coding focus) - ~20% faster than Q4_K_M
Pi-Pewter's coding model for raspberry pi5, and pi4 (coming later in 1b)
1.42B open-weight Llama-architecture language and coding model with a custom 49k BPE tokenizer and 32K context window for fast CPU and mobile inference.
Qwen 3.8 27b distilled into Gemma 4 12b with a focus on writing Python