Keoir

disk-resident MoE · stall-rate first

connecting… ·
Tokens
0
Tok/s
0
Stall rate
0%
Bytes/token
0
Prefetch recall
0%
Prefetch precision
0%

Per-layer stall rate headline

Tokens / sec

Cache tier hit rates

Layer × expert access heatmap

No heatmap data yet — generate a few tokens.

I/O + prefetch detail

pinned0
cache0
speculative0
cold stalls0
bytes read0
reads / batches0 / 0
max queue depth0
prefetch issued / used0 / 0

Loading bundles…

Converter

Live conversion over HTTP is not wired yet. This form builds the keoir convert CLI command for you.

keoir convert --input /path/to/model.gguf --output ./mymodel.keoir --dtype f16