Docs / Guides
Inspect and benchmark
Read the released container directly and measure decode performance on the active machine.
Inspect a model artifact
shell
fermion inspect
fermion inspect --layer 0 --projection q
fermion inspect --sha256 --jsoninspect reports geometry, parameters by tensor class, projection weight-state occupancy, byte use, per-layer occupancy spread, and KV-cache cost. --json makes the result machine-readable.
| Flag | Effect |
|---|---|
--layer N | Restrict inspection to one layer and skip the rest of the file. |
--projection TYPE | Select q, k, v, o, gate, up, or down. |
--sha256 | Add a full-file digest. |
--json | Return structured output. |
Benchmark the current machine
shell
fermion bench
fermion bench --tokens 128 --warmup 2
fermion bench --draft autoThe benchmark prints detected hardware, model size, load time, wall throughput, and steady-state decode throughput separately. With a draft, it adds a drafted row and an output-identity check.
| Flag | Default | Purpose |
|---|---|---|
--tokens | 32 | Greedy generation window. |
--prompt | fixed prompt | Replace the benchmark prompt. |
--warmup | 1 | Warm-up generations before measurement. |
--draft | none | Measure speculative decoding too. |