Skip to content

Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.

C++

GPU roles: two identical cards recommend the highest number as Main, so the engine order stays [0, 1]; a model that keeps one card keeps setup's old pick; golden back to [0, 1] with roles only

Niko1221890fee81b0ae
diff --git a/apps/host/src/components/GpuRoles.tsx b/apps/host/src/components/GpuRoles.tsx
index b2aa282e4..1badb9a97 100644
--- a/apps/host/src/components/GpuRoles.tsx
+++ b/apps/host/src/components/GpuRoles.tsx
@@ -16,7 +16,7 @@ export function recommendationLine(cards: RoleCard[], vendor?: string | null): s
const byi = new Map(cards.map(c => [c.index, c]));
const m = byi.get(rec.main)!;
if (cards.length === 1) return tr('Strata will use your {0}.', [shortName(m)]);
- const why = { same: 'the cards are alike, so the lower number is Main', memory: 'it has more memory', faster: 'it is faster', only: 'it is the only card', none: '' }[rec.why];
+ const why = { same: 'the cards are alike', memory: 'it has more memory', faster: 'it is faster', only: 'it is the only card', none: '' }[rec.why];
let s = tr('{0} is Main: {1}.', [shortName(m), tr(why)]);
const helpers = Object.keys(rec.roles).map(Number).filter(i => rec.roles[i] === 'helper').map(i => shortName(byi.get(i)!));
if (helpers.length) s += ' ' + tr(helpers.length > 1 ? '{0} help.' : '{0} helps.', [helpers.join(', ')]);
diff --git a/apps/host/src/lib/gpu-roles.ts b/apps/host/src/lib/gpu-roles.ts
index 8ba20e054..374e017c3 100644
--- a/apps/host/src/lib/gpu-roles.ts
+++ b/apps/host/src/lib/gpu-roles.ts
@@ -79,6 +79,7 @@ export function pickMain(cs: RoleCard[]): { main: RoleCard | null; why: MainWhy
const byMem = (a: RoleCard, b: RoleCard) => rnd(mem(b)) - rnd(mem(a)) || a.index - b.index;
let main = [...group].sort(byMem)[0];
let why: MainWhy = group.length === 1 ? 'faster' : 'same';
+ if (group.length > 1 && new Set(group.map(c => `${c.name}|${rnd(mem(c))}`)).size === 1) main = [...group].sort((a, b) => b.index - a.index)[0]; // identical cards: the HIGHEST number is Main
if (top <= 0 && new Set(pool.map(c => rnd(mem(c)))).size > 1) why = 'memory';
const bigger = pool.filter(c => rnd(mem(c)) > rnd(mem(main)) && (top <= 0 || speed(c) >= MEMORY_GUARD_SPEED * top));
if (bigger.length) { main = [...bigger].sort(byMem)[0]; why = 'memory'; }
@@ -238,7 +239,7 @@ export function sentence(cards: RoleCard[] | null | undefined, vendor?: string |
const byi = new Map(cs.map(c => [c.index, c]));
const m = byi.get(rec.main)!;
if (cs.length === 1) return `Strata will use your ${shortName(m)}.`;
- const why = { same: 'the cards are alike, so the lower number is Main', memory: 'it has more memory', faster: 'it is faster', only: 'it is the only card', none: '' }[rec.why];
+ const why = { same: 'the cards are alike', memory: 'it has more memory', faster: 'it is faster', only: 'it is the only card', none: '' }[rec.why];
let s = `${shortName(m)} is Main: ${why}.`;
const helpers = byNumber(rec.roles).filter(i => rec.roles[i] === 'helper').map(i => shortName(byi.get(i)!));
if (helpers.length) s += ` ${helpers.join(' and ')} ${helpers.length > 1 ? 'help' : 'helps'}.`;
diff --git a/apps/host/src/locales/de.json b/apps/host/src/locales/de.json
index fa6c50540..d181908ce 100644
--- a/apps/host/src/locales/de.json
+++ b/apps/host/src/locales/de.json
@@ -3471,7 +3471,7 @@
"s8625": "Es wurde keine Grafikkarte gefunden, die Strata nutzen kann.",
"s8626": "Strata verwendet Ihre {0}.",
"s8627": "{0} ist die Hauptkarte: {1}.",
- "s8628": "die Karten sind gleich, daher ist die niedrigere Nummer die Hauptkarte",
+ "s8628": "die Karten sind gleich",
"s8629": "sie hat mehr Speicher",
"s8630": "sie ist schneller",
"s8631": "sie ist die einzige Karte",
diff --git a/apps/host/src/locales/en.json b/apps/host/src/locales/en.json
index 2609d43d4..f1ecc9800 100644
--- a/apps/host/src/locales/en.json
+++ b/apps/host/src/locales/en.json
@@ -3471,7 +3471,7 @@
"s8625": "No graphics card that Strata can use was found.",
"s8626": "Strata will use your {0}.",
"s8627": "{0} is Main: {1}.",
- "s8628": "the cards are alike, so the lower number is Main",
+ "s8628": "the cards are alike",
"s8629": "it has more memory",
"s8630": "it is faster",
"s8631": "it is the only card",
diff --git a/apps/host/src/locales/es.json b/apps/host/src/locales/es.json
index 37158a49e..905102726 100644
--- a/apps/host/src/locales/es.json
+++ b/apps/host/src/locales/es.json
@@ -3471,7 +3471,7 @@
"s8625": "No se encontró ninguna tarjeta gráfica que Strata pueda usar.",
"s8626": "Strata usará tu {0}.",
"s8627": "{0} es la principal: {1}.",
- "s8628": "las tarjetas son iguales, así que la de menor número es la principal",
+ "s8628": "las tarjetas son iguales",
"s8629": "tiene más memoria",
"s8630": "es más rápida",
"s8631": "es la única tarjeta",
diff --git a/apps/host/src/locales/ja.json b/apps/host/src/locales/ja.json
index d5d0caac1..eeddc975a 100644
--- a/apps/host/src/locales/ja.json
+++ b/apps/host/src/locales/ja.json
@@ -3471,7 +3471,7 @@
"s8625": "Strata が使えるグラフィックカードが見つかりません。",
"s8626": "Strata は {0} を使います。",
"s8627": "{0} がメインです:{1}。",
- "s8628": "カードが同じなので番号の小さい方がメインです",
+ "s8628": "カードが同じため",
"s8629": "メモリが多いため",
"s8630": "より高速なため",
"s8631": "唯一のカードのため",
diff --git a/apps/host/src/locales/ko.json b/apps/host/src/locales/ko.json
index b103fa278..ea1aded1a 100644
--- a/apps/host/src/locales/ko.json
+++ b/apps/host/src/locales/ko.json
@@ -3471,7 +3471,7 @@
"s8625": "Strata가 사용할 수 있는 그래픽 카드를 찾지 못했습니다.",
"s8626": "Strata가 {0}을(를) 사용합니다.",
"s8627": "{0}이(가) 메인입니다: {1}.",
- "s8628": "카드가 같아서 번호가 낮은 쪽이 메인입니다",
+ "s8628": "카드가 같습니다",
"s8629": "메모리가 더 많습니다",
"s8630": "더 빠릅니다",
"s8631": "유일한 카드입니다",
diff --git a/apps/host/src/locales/pt.json b/apps/host/src/locales/pt.json
index 4af869165..7787a542c 100644
--- a/apps/host/src/locales/pt.json
+++ b/apps/host/src/locales/pt.json
@@ -3471,7 +3471,7 @@
"s8625": "Nenhuma placa de vídeo que o Strata possa usar foi encontrada.",
"s8626": "O Strata usará a sua {0}.",
"s8627": "{0} é a principal: {1}.",
- "s8628": "as placas são iguais, por isso a de menor número é a principal",
+ "s8628": "as placas são iguais",
"s8629": "tem mais memória",
"s8630": "é mais rápida",
"s8631": "é a única placa",
diff --git a/apps/host/src/locales/ru.json b/apps/host/src/locales/ru.json
index 621a867cf..ed0d4358d 100644
--- a/apps/host/src/locales/ru.json
+++ b/apps/host/src/locales/ru.json
@@ -3471,7 +3471,7 @@
"s8625": "Не найдено видеокарт, которые может использовать Strata.",
"s8626": "Strata будет использовать ваш {0}.",
"s8627": "{0} — основная: {1}.",
- "s8628": "карты одинаковые, поэтому основной считается карта с меньшим номером",
+ "s8628": "карты одинаковые",
"s8629": "у неё больше памяти",
"s8630": "она быстрее",
"s8631": "это единственная карта",
diff --git a/apps/host/src/locales/zh.json b/apps/host/src/locales/zh.json
index bef3dc374..f7a00a9df 100644
--- a/apps/host/src/locales/zh.json
+++ b/apps/host/src/locales/zh.json
@@ -3471,7 +3471,7 @@
"s8625": "未找到 Strata 可用的显卡。",
"s8626": "Strata 将使用你的 {0}。",
"s8627": "{0} 是主卡:{1}。",
- "s8628": "两张卡相同,所以编号较小的是主卡",
+ "s8628": "两张卡相同",
"s8629": "它的显存更大",
"s8630": "它更快",
"s8631": "它是唯一的显卡",
diff --git a/docs/MULTI_GPU.md b/docs/MULTI_GPU.md
index b043bff11..5ebae7fe7 100644
--- a/docs/MULTI_GPU.md
+++ b/docs/MULTI_GPU.md
@@ -56,7 +56,7 @@ the picture reader. (Older versions of this page said "the first is the main one
**Recommended roles** (the same rule in the app, `strata launch` and setup; `strata_control/gpu_roles.py`): the faster card
is Main (speed is multiprocessors x max clock, the engine's own measure; cards within 5% are the same card, and the one with
-more memory, then the lower number, wins; a card with clearly more memory beats a faster one that is at least half as fast).
+more memory wins, then the lower number; two identical cards: the highest number is Main, so the engine order stays 0, 1 as it always was; a card with clearly more memory beats a faster one that is at least half as fast).
Every other card starts as Helper, except a card with **8 GB or less** (Off), built-in graphics beside a card with memory of
its own (Off) and a card the engine cannot run (Off). Every role is one click (the app), one key (`strata launch`: M, H, O;
R goes back to the recommendation) or one line (below).
diff --git a/setup.py b/setup.py
index 3d486fd32..94cd0fc3e 100644
--- a/setup.py
+++ b/setup.py
@@ -1634,6 +1634,14 @@ def choose_gpus(a, found) -> list:
return choose_roles(a, found)
+def fallback_card(a, gpu, chosen):
+ """A model that cannot share its memory across cards keeps ONE card: the one setup always chose (the most VRAM, then the lower number),
+ not the Main card of the roles - unless the person named the roles (--gpu-roles), then their Main card."""
+ if getattr(a, "_roles_explicit", False) or not getattr(a, "_roles", None):
+ return gpu
+ return min(chosen, key=lambda g: (-round(g["vram_gb"]), g["index"]))
+
+
def card_speeds() -> dict:
"""{card number: SMs x max clock} of the NVIDIA cards (strata_control/gpu_roles.py); {} when it cannot say."""
from strata_control import gpu_roles
@@ -7616,6 +7624,7 @@ def main() -> int:
ap.error("--gpu-roles: setup needs at least one card that is Main or Helper")
mains = [i for i, r in named.items() if r == "main"] or [sorted(on)[-1]]
helpers = sorted(i for i in on if i != mains[0])
+ a._roles_explicit = True
a._roles = {**{i: "off" for i in named}, **{i: "helper" for i in helpers}, mains[0]: "main"}
order = helpers + [mains[0]]
if len(order) > 1:
@@ -8017,7 +8026,8 @@ def main() -> int:
warn(f"installing {model} with {ram:.0f} GB of RAM, as you chose")
if in_budget:
budget = budget_choice(model, ram, a.resident_budget_gib)
- if multi and not unsloth_together(a, model, ram, gpu, chosen):
+ if multi and not unsloth_together(a, model, ram, fallback_card(a, gpu, chosen), chosen):
+ gpu = fallback_card(a, gpu, chosen)
multi, sel, chosen = [], [gpu["index"]], [gpu]
q4_split = bool(multi) and not resident_split() # #498: on several GPUs without the RAM budget (#642: kept)
if not q4_split:
@@ -8033,7 +8043,8 @@ def main() -> int:
"the low-RAM mode")
low_ram = low_ram_wanted(model, ram, a.low_ram)
# #642: the engines from RESIDENT_SPLIT_ENGINE run the low-RAM mode on the chosen cards as on one (decided below)
- if low_ram and multi and not resident_split() and not low_ram_together(a, model, ram, gpu, chosen):
+ if low_ram and multi and not resident_split() and not low_ram_together(a, model, ram, fallback_card(a, gpu, chosen), chosen):
+ gpu = fallback_card(a, gpu, chosen)
multi, sel, chosen = [], [gpu["index"]], [gpu]
# (the low-RAM mode's variant is decided once the context is known, below; on several GPUs it is the mapped one
# before RESIDENT_SPLIT_ENGINE, #642)
diff --git a/strata_control/gpu_roles.py b/strata_control/gpu_roles.py
index 8b2d5e1ee..33df9fe43 100644
--- a/strata_control/gpu_roles.py
+++ b/strata_control/gpu_roles.py
@@ -40,6 +40,10 @@ def _speed(c: dict) -> float:
return float(v) if isinstance(v, (int, float)) and not isinstance(v, bool) and v > 0 else 0.0
+def rnd_(x: float) -> int:
+ return round(x)
+
+
def _shares(c: dict) -> bool:
return gpu_fit._shares_memory(c)
@@ -80,7 +84,7 @@ def _eligible(cs: list[dict]) -> list[dict]:
def pick_main(cs: list[dict]) -> tuple[dict | None, str]:
"""(the Main card, why) why: only | faster | same | memory. The faster card, cards within 5% being the same card (the one with more memory,
- then the lower number); a card with clearly more memory wins when it is at least half as fast (#1576)."""
+ then, for identical cards, the HIGHEST number - so the engine order stays [0, 1] - else the lower number); a card with clearly more memory wins when it is at least half as fast (#1576)."""
pool = _eligible(cs)
if not pool:
return None, "none"
@@ -90,6 +94,8 @@ def pick_main(cs: list[dict]) -> tuple[dict | None, str]:
group = [c for c in pool if _speed(c) * SAME_SPEED >= top] if top > 0 else list(pool)
main = sorted(group, key=lambda c: (-round(_mem(c)), _idx(c)))[0]
why = "faster" if len(group) == 1 else "same"
+ if len(group) > 1 and len({(str(c.get("name")), rnd_(_mem(c))) for c in group}) == 1:
+ main = max(group, key=_idx) # identical cards: the HIGHEST number is Main, so the engine order stays [0, 1] as it always was
if top <= 0 and len({round(_mem(c)) for c in pool}) > 1:
why = "memory" # speeds unknown: the card with the most memory
bigger = [c for c in pool if round(_mem(c)) > round(_mem(main)) and (top <= 0 or _speed(c) >= MEMORY_GUARD_SPEED * top)]
@@ -338,7 +344,7 @@ def sentence(cards: list[dict], vendor: str | None = None) -> str:
if len(cs) == 1:
return f"Strata will use your {short_name(m)}."
helpers = [byi[i] for i, r in rec["roles"].items() if r == "helper"]
- why = {"same": "the cards are alike, so the lower number is Main", "memory": "it has more memory", "faster": "it is faster", "only": "it is the only card"}[rec["why"]]
+ why = {"same": "the cards are alike", "memory": "it has more memory", "faster": "it is faster", "only": "it is the only card"}[rec["why"]]
s = f"{short_name(m)} is Main: {why}."
if helpers:
s += " " + " and ".join(short_name(h) for h in helpers) + (" help." if len(helpers) > 1 else " helps.")
diff --git a/tests/data/gpu_roles_vectors.json b/tests/data/gpu_roles_vectors.json
index 41dd5ed8f..45c2eff12 100644
--- a/tests/data/gpu_roles_vectors.json
+++ b/tests/data/gpu_roles_vectors.json
@@ -15,7 +15,7 @@
"recommend": [
{"hw": "one", "roles": {"0": "main"}, "main": 0, "why": "only"},
{"hw": "two", "roles": {"0": "helper", "1": "main"}, "main": 1, "why": "faster"},
- {"hw": "equal", "roles": {"0": "main", "1": "helper"}, "main": 0, "why": "same"},
+ {"hw": "equal", "roles": {"0": "helper", "1": "main"}, "main": 1, "why": "same"},
{"hw": "three", "roles": {"0": "helper", "1": "main", "2": "off"}, "main": 1, "why": "faster"},
{"hw": "igpu", "roles": {"0": "off", "1": "main"}, "main": 1, "why": "only"},
{"hw": "igpu_only", "roles": {"0": "main"}, "main": 0, "why": "only"},
diff --git a/tests/terminal/test_launch.py b/tests/terminal/test_launch.py
index 733fa1bbb..a4d6acc64 100644
--- a/tests/terminal/test_launch.py
+++ b/tests/terminal/test_launch.py
@@ -585,7 +585,7 @@ class AdoptRules(unittest.TestCase):
# which cards: named ones win (one card too: --gpu 1 on a box that has them), else the fit rule on a box with several cards
two = catalog(HW_2X16)
self.assertEqual(cards_for_launch(S.LaunchSettings(family="qwen", model="IQ3_XXS", gpus=[1, 0]), two), [1, 0])
- self.assertEqual(cards_for_launch(S.LaunchSettings(family="qwen", model="IQ3_XXS"), two), [1, 0]) # roles: card 0 Main (last), card 1 Helper
+ self.assertEqual(cards_for_launch(S.LaunchSettings(family="qwen", model="IQ3_XXS"), two), [0, 1]) # roles: card 0 Main (last), card 1 Helper
one = catalog(HW_12)
self.assertEqual(cards_for_launch(S.LaunchSettings(family="qwen", model="IQ3_XXS", gpus=[1]), one), [1])
self.assertIsNone(cards_for_launch(S.LaunchSettings(family="qwen", model="IQ3_XXS"), one))
@@ -614,7 +614,7 @@ class AdoptRules(unittest.TestCase):
self.assertIsNone(installed_cards({**chosen, "gpus_asked": False})) # never asked: setup's default, not a choice
self.assertIsNone(installed_cards({"gpu": 1}))
self.assertIsNone(cards_for_launch(plain, two, base=chosen)) # the installed choice stays
- self.assertEqual(cards_for_launch(plain, two, base={**chosen, "gpus_asked": False}), [1, 0]) # no choice recorded: the recommended roles
+ self.assertEqual(cards_for_launch(plain, two, base={**chosen, "gpus_asked": False}), [0, 1]) # no choice recorded: the recommended roles
self.assertEqual(cards_for_launch(S.LaunchSettings(family="qwen", model="IQ3_XXS", gpus=[0]), two, base=chosen), [0]) # --gpu 0 wins
if True:
installed = h.base / "strata-iq3_xxs.json"
diff --git a/tests/terminal/test_launch_tui.py b/tests/terminal/test_launch_tui.py
index f579870bb..9ed3d990f 100644
--- a/tests/terminal/test_launch_tui.py
+++ b/tests/terminal/test_launch_tui.py
@@ -866,7 +866,7 @@ class CardsStep(Base):
self.assertIn("Which graphics cards should it use?", t)
self.assertIn("Step 2 of 10", t)
flat = " ".join(t.split())
- self.assertIn("[ Main ]", flat) # equal cards: the lower number is Main
+ self.assertIn("[ Main ]", flat) # equal cards: the highest number is Main (the engine order stays [0, 1])
self.assertIn("[ Helper ]", flat)
self.assertIn("recommended", flat)
self.assertIn("M Main", flat)
@@ -880,14 +880,14 @@ class CardsStep(Base):
def test_m_h_o_set_the_role_of_the_card_under_the_cursor_and_r_goes_back(self):
async def case(app, pilot, ctx):
await press(pilot, "enter", "enter")
- await press(pilot, "down", "o") # card 2 Off
+ await press(pilot, "o") # card 1 (a Helper) Off
flat = " ".join(text_of(app).split())
self.assertIn("recommended: helper", flat) # the muted tag says what was recommended
- await press(pilot, "m") # Main on card 2: card 1 becomes a Helper, one quiet line says so
+ await press(pilot, "m") # Main on card 1: card 2 becomes a Helper, one quiet line says so
flat = " ".join(text_of(app).split())
self.assertIn("is now a Helper.", flat)
await press(pilot, "enter")
- self.assertEqual(ctx.settings.gpu_roles, {"0": "helper", "1": "main"})
+ self.assertEqual(ctx.settings.gpu_roles, {"0": "main", "1": "helper"})
self.assertIsNone(ctx.settings.gpus)
self.run_app(case, ctx=make_ctx(HW_2X16))
diff --git a/tools/test_setup_golden.json b/tools/test_setup_golden.json
index 73725ecc7..430fc2992 100644
--- a/tools/test_setup_golden.json
+++ b/tools/test_setup_golden.json
@@ -295,13 +295,13 @@
"cwd": "<T>",
"exe": "<T>/engine/<EXE>",
"gpu": [
- 1,
- 0
+ 0,
+ 1
],
"gpu_order": "as_given",
"gpu_roles": {
- "0": "main",
- "1": "helper"
+ "0": "helper",
+ "1": "main"
},
"gpus_asked": true,
"layer_split": "auto",
@@ -355,13 +355,13 @@
"cwd": "<T>",
"exe": "<T>/engine/<EXE>",
"gpu": [
- 1,
- 0
+ 0,
+ 1
],
"gpu_order": "as_given",
"gpu_roles": {
- "0": "main",
- "1": "helper"
+ "0": "helper",
+ "1": "main"
},
"gpus_asked": true,
"layer_split": "auto",
@@ -415,13 +415,13 @@
"cwd": "<T>",
"exe": "<T>/engine/<EXE>",
"gpu": [
- 1,
- 0
+ 0,
+ 1
],
"gpu_order": "as_given",
"gpu_roles": {
- "0": "main",
- "1": "helper"
+ "0": "helper",
+ "1": "main"
},
"gpus_asked": true,
"layer_split": "auto",
@@ -475,13 +475,13 @@
"cwd": "<T>",
"exe": "<T>/engine/<EXE>",
"gpu": [
- 1,
- 0
+ 0,
+ 1
],
"gpu_order": "as_given",
"gpu_roles": {
- "0": "main",
- "1": "helper"
+ "0": "helper",
+ "1": "main"
},
"gpus_asked": true,
"layer_split": "auto",
@@ -590,13 +590,13 @@
"cwd": "<T>",
"exe": "<T>/engine/<EXE>",
"gpu": [
- 1,
- 0
+ 0,
+ 1
],
"gpu_order": "as_given",
"gpu_roles": {
- "0": "main",
- "1": "helper"
+ "0": "helper",
+ "1": "main"
},
"gpus_asked": true,
"layer_split": "auto",
@@ -649,13 +649,13 @@
"cwd": "<T>",
"exe": "<T>/engine/<EXE>",
"gpu": [
- 1,
- 0
+ 0,
+ 1
],
"gpu_order": "as_given",
"gpu_roles": {
- "0": "main",
- "1": "helper"
+ "0": "helper",
+ "1": "main"
},
"gpus_asked": true,
"layer_split": "auto",
@@ -707,13 +707,13 @@
"cwd": "<T>",
"exe": "<T>/engine/<EXE>",
"gpu": [
- 1,
- 0
+ 0,
+ 1
],
"gpu_order": "as_given",
"gpu_roles": {
- "0": "main",
- "1": "helper"
+ "0": "helper",
+ "1": "main"
},
"gpus_asked": true,
"layer_split": "auto",
@@ -766,13 +766,13 @@
"cwd": "<T>",
"exe": "<T>/engine/<EXE>",
"gpu": [
- 1,
- 0
+ 0,
+ 1
],
"gpu_order": "as_given",
"gpu_roles": {
- "0": "main",
- "1": "helper"
+ "0": "helper",
+ "1": "main"
},
"gpus_asked": true,
"layer_split": "auto",
diff --git a/tools/test_setup_golden.py b/tools/test_setup_golden.py
index 2d5ea6186..eb940cee6 100644
--- a/tools/test_setup_golden.py
+++ b/tools/test_setup_golden.py
@@ -180,7 +180,7 @@ class Golden(unittest.TestCase):
lowram_multi = self.golden["32GB-2x24GB qwen IQ3_XXS"]["config"] # #364 -> #642: low-RAM, two cards -> both
self.assertIn("--resident-experts" if "--resident-experts" in lowram_multi["args"] else "--mmap-experts",
lowram_multi["args"])
- self.assertEqual(lowram_multi.get("gpu"), [1, 0]) # 0.1.40 (#642): resident on both cards; roles: Main 0 goes last
+ self.assertEqual(lowram_multi.get("gpu"), [0, 1]) # 0.1.40 (#642): resident on both cards; roles: Main 1 goes last, as before
def test_yes(self):
for key, ram, found, family, model in self.cases():
diff --git a/tools/test_setup_gpu_fit.py b/tools/test_setup_gpu_fit.py
index 20e8e1b3a..efbbc837e 100644
--- a/tools/test_setup_gpu_fit.py
+++ b/tools/test_setup_gpu_fit.py
@@ -44,12 +44,12 @@ class SetupFitRule(unittest.TestCase):
self.assertEqual(cfg["gpu_roles"], {"0": "helper", "1": "main"})
self.assertEqual(cfg["layer_split"], "auto")
- def test_two_equal_cards_start_as_main_and_helper_with_the_lower_number_main(self):
+ def test_two_equal_cards_start_with_the_highest_number_as_main(self):
found = [card(0, "NVIDIA GeForce RTX 3060", 12.0, "86"), card(1, "NVIDIA GeForce RTX 3060", 12.0, "86")]
code, out, cfg, asked = run(26.0, found, "IQ3_XXS")
self.assertEqual(code, 0, out)
- self.assertEqual(cfg["gpu"], [1, 0])
- self.assertEqual(cfg["gpu_roles"], {"0": "main", "1": "helper"})
+ self.assertEqual(cfg["gpu"], [0, 1])
+ self.assertEqual(cfg["gpu_roles"], {"0": "helper", "1": "main"})
self.assertEqual(cfg["gpu_order"], "as_given")
self.assertEqual(cfg["layer_split"], "auto")
diff --git a/tools/test_setup_risk.py b/tools/test_setup_risk.py
index b3562e2c6..99d6cfe30 100644
--- a/tools/test_setup_risk.py
+++ b/tools/test_setup_risk.py
@@ -241,7 +241,7 @@ class LowRamGpus(unittest.TestCase):
answers={"Low-RAM mode: which GPUs?": "2"}) # the roles are the default (Main last); the low-RAM question still defaults to one card
self.assertEqual(code, 0, out)
self.assertTrue(any("Low-RAM mode: which GPUs? [1]" in q for q in asked), asked)
- self.assertEqual(cfg["gpu"], [1, 0])
+ self.assertEqual(cfg["gpu"], [0, 1])
self.assertIn("--mmap-experts", cfg["args"])
self.assertIn("1) GPU 0 (NVIDIA GeForce RTX 3090, 24 GB) only", out)
@@ -278,7 +278,7 @@ class LowRamGpus(unittest.TestCase):
code, out, cfg, asked = install(ram, found, ["--family", "qwen", "--model", "Q2_0", "--no-start"],
answers={"Use these roles?": "y"}) # the recommended roles: Main 0 (last), Helper 1
self.assertEqual(code, 0, out)
- self.assertEqual(cfg["gpu"], [1, 0])
+ self.assertEqual(cfg["gpu"], [0, 1])
self.assertIn("--resident-experts", cfg["args"])
self.assertFalse(any("Low-RAM mode" in q for q in asked), asked)
self.assertIn("stay in RAM", out)
diff --git a/tools/test_setup_unsloth.py b/tools/test_setup_unsloth.py
index f6cc3d180..2d32c1397 100644
--- a/tools/test_setup_unsloth.py
+++ b/tools/test_setup_unsloth.py
@@ -647,7 +647,7 @@ class LayerSplit(unittest.TestCase):
self.assertIn("--resident-budget-gib", cfg["args"])
code, out, cfg, asked = install(165.0, self.found, argv, answers={f"{M}: which GPUs?": "2"})
self.assertEqual(code, 0, out)
- self.assertEqual(cfg["gpu"], [1, 0])
+ self.assertEqual(cfg["gpu"], [0, 1])
self.assertEqual(cfg["layer_split"], "auto")
self.assertNotIn("--resident-budget-gib", cfg["args"])