2025.01.15
03:47:00.000
Llama 3.1 8B Instruct
Meta AI · Open Weights Trajectory
RTX 3090 24GB
Ollama v0.4.0
Q4_K_M compression tensor
MMLU-Pro
GPQA
IFEval
14.2GB proper mass. 47c/60 tok/s asymptotic velocity. Six-hour proper time to singularity. Quantization tidal forces below elastic limit. Information preserved.
2025.01.12
22:15:00.000
Qwen2.5 14B Instruct
Alibaba DAMO · Cross-lingual geodesic
RTX 3090 + 64GB host
llama.cpp b4000
Q5_K_M · Mixed topology
HumanEval
MBPP
MultiPL-E
LiveCodeBench
Eight layers in ergosphere, CPU-frame dragged. 23c/60 tok/s with redshift degradation. Code-generation frame-dragging exceptional in C++/Python sectors. Rust trajectory truncated by event horizon diameter, not curvature singularity.
2025.01.08
17:33:00.000
Mistral Small 24B Instruct
Mistral AI · European reference frame
Dual RTX 3090 NVLink
vLLM 0.6.5
BF16 · naked singularity
MATH
GSM8K
BBH
24B parameters without clothing — no compression event horizon. 48GB ergosphere fully utilized. 89c/60 tok/s — maximum stable orbit achieved. Chain-of-thought Penrose extraction exceeds all compressed competitors. NVLink wormhole throat negligible.