Pull requests / #2380

#2380 setup: calibrate a dual config whose gpu is the string 0,1

open · @MEsfits · 0 comments · View on GitHub

Setup & installServer & APIMulti-GPUAMD / HIPNVIDIA / CUDA

Description

Fixes #2378.

`hardware_key()` passed the config's raw `"gpu"` value to `gpu_info()`, which matches a single card by number — so `./setup.sh --calibrate` on a dual config with `"gpu": "0,1"` ran every measurement round, then died with `there is no GPU 0,1` at the save step (and left tier-measurement flags half-written into the user's file).

- New `gpu_sel()` normalizes `0`, `[0, 2]` and `"0,2"` (the string form `serve` already splits); `hardware_key()` uses it for the string case, keeping `None` flowing to `hip_config_cards` where it means "default card".
- `recommend_remote_expert_opt()` uses it too instead of silently skipping string-form duals with a helper cache.
- Regression test `test_nvidia_string_dual_keys_like_the_list` in `CalibrationKey`.

Verified: string `"0,1"` yields the identical key as `[0, 1]`; single-int, single-string, and missing-key forms unchanged; all 36 `test_setup_amd.py` tests pass; full `test_setup_*.py` (583 tests) shows only the same 10 pre-existing environment failures as unmodified code (stash A/B).

Tested live on 2x RTX 3090 (24GB): the previously crashing save path now completes.

Related on strata.com

Editorial links to help you install, pick models, or read release notes — not part of the upstream thread.