`bench/prefill-decode.toml`

37 lines
# Prefill/Decode disaggregation benchmark config.
#
# Top-level keys are bench-wide. [prefill] and [decode] sections set per-side
# placement filters and (optionally) per-side model.
#
# Example:
#   uv run python bench/prefill_decode_bench.py --config bench/prefill-decode.toml

host = "james"
port = 52415
timeout = 7200.0
settle_timeout = 60.0

# Workload
pp = [4096]
tg = [512]
repeat = 1
warmup = 0

json_out = "bench/prefill_decode_results.json"

[prefill]
model = "mlx-community/gpt-oss-20b-MXFP4-Q8"
node = "mike"
instance_meta = "ring"
sharding = "pipeline"
min_nodes = 1
max_nodes = 1

[decode]
model = "mlx-community/gpt-oss-20b-MXFP4-Q8"
node = "james"
instance_meta = "ring"
sharding = "pipeline"
min_nodes = 1
max_nodes = 1