Cross-reference graph
18,818 resolved references between 6,943 non-source records. This is the graph the repo actually encodes: a benchmark points at its engine and its accelerators, a gotcha points at what it affects, a FLOP class points at the hardware it is sensitive to. Sources are excluded here — everything cites them, so they would be one hub.
Neighbourhood
Type a record id or name to draw its immediate neighbourhood. This is the same data as the "Cross-references" and "Referenced by" panels on every record page, drawn instead of listed.
The most-connected records:
| Record | Type | Connections |
|---|---|---|
| llama.cpp engines/llama-cpp | engines | 3787 |
| GGUF Q4_K_M (4-bit K-quant, medium) quantization/gguf-q4-k-m | quantization | 1668 |
| PCI Express 4.0 interconnect/pcie-gen4 | interconnect | 1332 |
| GGUF Q4_0 (legacy block-4 symmetric) quantization/gguf-q4-0 | quantization | 701 |
| AMD Radeon 8060S (Ryzen AI Max+ integrated GPU) accelerators/amd-radeon-8060s-ryzen-ai-max | accelerators | 617 |
| vLLM engines/vllm | engines | 583 |
| PCI Express 5.0 interconnect/pcie-gen5 | interconnect | 493 |
| NVIDIA H100 SXM accelerators/nvidia-h100-sxm | accelerators | 443 |
| NVIDIA A100 80GB SXM4 accelerators/nvidia-a100-80gb-sxm4 | accelerators | 321 |
| GGUF MXFP4_MOE (llama.cpp MoE-specific ftype: MXFP4 for experts, Q8_0 for everything else) quantization/gguf-mxfp4-moe | quantization | 287 |
| AMD Instinct MI300X accelerators/amd-instinct-mi300x | accelerators | 230 |
| NVIDIA GeForce RTX 4090 accelerators/nvidia-rtx-4090 | accelerators | 225 |
| NVIDIA A100 40GB SXM4 accelerators/nvidia-a100-40gb-sxm4 | accelerators | 198 |
| NVIDIA B200 accelerators/nvidia-b200 | accelerators | 195 |
Whole-corpus map
A force-directed layout of all non-source records. Each record is one dot, each resolved reference one line. It is dense — that is the point: the corpus is a graph, not a list.
Hover a dot for its name; click to open the record. Dots with a red ring are contested records.