Neural Membrane Testing

Home inference as living tissue — models grown, not merely executed

Signal Transduction Log

2025.01.15	Llama 3.1 8B Instruct Meta AI · metamembrane lineage	RTX 3090 24GB Ollama v0.4.0 · Q4_K_M pruning	MMLU-Pro GPQA IFEval	View signal	14.2GB activation peak. 47 impulses/second sustained. Six-hour depolarization without refractory failure. Quantization synaptic noise below perceptual threshold.
2025.01.12	Qwen2.5 14B Instruct Alibaba Cloud · cross-lingual graft	RTX 3090 + 64GB host llama.cpp b4000 · Q5_K_M	HumanEval MBPP MultiPL-E LiveCodeBench	View signal	Eight layers offloaded to systemic circulation. 23 impulses/second with delayed propagation. Code-generation receptor density remarkably high in C++ and Python domains. Rust task failure: context length exhaustion, not binding-site deficiency.
2025.01.08	Mistral Small 24B Mistral AI · European consortium	Dual RTX 3090 NVLink vLLM 0.6.5 · BF16 native	MATH GSM8K BBH	View signal	Native 24B architecture, no synaptic pruning required. 48GB aggregate membrane fully utilized. 89 impulses/second — highest recorded. Chain-of-thought long-term potentiation significantly exceeds quantized competitors.

Secondary Ganglion

Roleplaying potentials, creative excitation patterns, character consistency oscillations, and exploratory stimulus-response mappings that resist quantitative synaptic weight analysis.

Propagate signal →