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++

Recommend by the owner's memory tiers (RAM + VRAM): 40 GB Swift 1.5 IQ2_XS with Gyro-M named, 48 GB IQ3_XXS, 64 GB IQ3_S; MODELS.md Pick by memory

Niko1221aa82a88639c6
diff --git a/docs/INSTALL.md b/docs/INSTALL.md
index ec533147f..9330ccde4 100644
--- a/docs/INSTALL.md
+++ b/docs/INSTALL.md
@@ -22,7 +22,7 @@ There are three ways to install it. Take the first one unless you have a reason
| | |
| --- | --- |
| **GPU** | **NVIDIA** RTX 20, 30, 40 or 50 series, **12 GB VRAM or more** (8 GB runs, slowly). Measured on an RTX 5070 and an RTX 3090; RTX 20 (Turing, since 0.1.27) was tested by a contributor on an RTX 2070. **AMD** Radeon RX 7900 XT / XTX, RX 9070 / 9070 XT and Radeon AI PRO R9700 (validated), RX 7800 XT / 7700 XT and RX 9060 XT (validated by their owners), RX 6800 / 6900 series (community-reported), with 12 GB of VRAM or more. See [AMD cards](#amd-cards). |
-| **RAM** | Enough for the size you pick ([which model](MODELS.md#pick-by-ram)); **64 GB** runs every size. A big GPU makes up for less RAM - the [low-RAM mode](MODELS.md#a-big-graphics-card-and-little-ram). |
+| **RAM** | Enough for the size you pick ([which model](MODELS.md#pick-by-memory)); **64 GB** runs every size. A big GPU makes up for less RAM - the [low-RAM mode](MODELS.md#a-big-graphics-card-and-little-ram). |
| **CPU** | x86-64 with AVX2 (any Intel/AMD desktop CPU from the last ~8 years). AVX-512 (Ryzen 7000/9000) is a bit faster. Older CPUs without AVX2 are experimental and slow: [Older CPUs](#older-cpus-experimental). |
| **Disk** | The app and its engine take about 1-3 GB. ~70-80 GB free for the model, ~6 GB for the MTP layer (+1 GB with images). Q2_0 on an AVX-512 CPU also writes a one-time ~40 GB copy of its experts for the fast CPU kernel. On Linux with an AMD card, ROCm takes ~10 GB more when setup installs it. An NVMe SSD is strongly recommended: it makes the first start much faster. |
| **OS** | Windows 10/11, Linux (Ubuntu 22.04/24.04 and Debian 12/13 get everything installed automatically; the app also runs on Fedora 44), or a Mac with Apple Silicon (M1 or newer) and macOS 15 or newer ([Mac](MAC.md)). |
diff --git a/docs/MODELS.md b/docs/MODELS.md
index 601bbee23..fe4c0b7c3 100644
--- a/docs/MODELS.md
+++ b/docs/MODELS.md
@@ -6,23 +6,31 @@ makers' compressions). It also runs a smaller model, [Qwen3.6-35B-A3B](#qwen36-3
(`strata launch`) and the installer recommend one for your PC; this page explains the choice. Back to the
[README](../README.md).
-> **On this page:** [Pick by RAM](#pick-by-ram) · [Speed](#how-fast-is-each-size) · [The sizes](#the-sizes) ·
+> **On this page:** [Pick by memory](#pick-by-memory) · [Speed](#how-fast-is-each-size) · [The sizes](#the-sizes) ·
> [Will it fit?](#will-it-fit) · [The versions](#the-versions) · [Qwen3.6](#qwen36-35b-a3b) ·
> [Adding another model](#adding-or-switching-models)
-## Pick by RAM
+## Pick by memory
-| Your RAM | Take | Why |
+The tiers count RAM and VRAM (the graphics card's memory) together, the same as the [README](../README.md#the-other-sizes).
+`strata launch`, setup and the install tool of the MCP server pre-select by them; the desktop app does from 0.2.1. Every
+other size stays a choice. A PC with a card of 12 GB and 32 GB of RAM is in the 40 GB tier (32 + 12 = 44).
+
+| Your memory (RAM + VRAM) | Pre-selected | Notes |
| --- | --- | --- |
-| **32 GB** | **Coder** | it fits 32 GB; best at code, weaker at everything else ([why](#coder)). With a 24 GB card, Q2_0 and IQ2_XS run too ([low-RAM mode](#a-big-graphics-card-and-little-ram)) and are the better pick for general use |
-| **48 GB** | **IQ2_XS** (or Q2_0, the fastest) | the larger sizes do not fit |
-| **64 GB** | **IQ2_XS** (recommended), or IQ3_XXS / IQ3_S | every size fits (IQ3_S with little else open) |
-| **96 GB or more** | **IQ3_S**, or [Unsloth's UD-IQ4_XS or UD-Q4_K_XL](#unsloths-files-ud-iq4_xs-to-q8_0) (~4-bit) | room for the largest sizes |
-| **128-192 GB** | [Unsloth's UD-Q5_K_XL, UD-Q6_K_XL or Q8_0](#unsloths-files-ud-iq4_xs-to-q8_0) | every expert of the 5-8-bit files in RAM |
+| **32 GB** | **Gyro-S**, or the **Coder** for code | The Coder needs 32 GB of RAM and keeps half of the experts, so it is weaker outside code ([why](#coder)). Gyro-S has lower quality than the 2-3-bit sizes, and the 2-bit Q2_0 decodes faster |
+| **40 GB** | **Swift 1.5** IQ2_XS (39.2 GB) | Gyro-M (35 GB) is the other choice: close to IQ3_XXS in quality, but it writes answers 17-21% slower than Q2_0. Gyro runs on NVIDIA and AMD cards, not on Intel Arc or the Mac |
+| **48 GB** | **IQ3_XXS** (47.0 GB) | Slower, better |
+| **64 GB** | **IQ3_S** (54.8 GB), the original. Swift 1.5's IQ3_S is an equal choice | The slowest, the best: it matches the full model on the published tests |
+| **96 GB or more** | [Unsloth's UD-IQ4_XS to Q8_0](#unsloths-files-ud-iq4_xs-to-q8_0), 4 to 8 bit | UD-IQ4_XS keeps every expert in RAM from about 80 GB of RAM, Q8_0 from about 136 GB. Up to 192 GB, the 5-8-bit files fit in RAM. Setup still pre-selects IQ3_S here, except on a Strix Halo |
+
+With less memory than a row says, Strata's [low-RAM mode](#a-big-graphics-card-and-little-ram) starts by itself and the
+model runs slower, much slower with a small card. The figures above are RAM + VRAM; [Will it fit?](#will-it-fit) counts
+RAM alone, for the sizes that keep their experts in RAM.
-Not sure? Take **IQ2_XS**. The **Coder** is the one that fits a 32 GB PC, but it keeps only 256 of the 512 experts,
-chosen on code data, so it is weaker outside code and in languages other than English. For general use, or whenever
-your RAM allows, take a full model (Q2_0, IQ2_XS or IQ3_S).
+Not sure? Take what the installer pre-selects. The Coder fits a 32 GB PC, but it keeps only 256 of the 512 experts,
+chosen on code data, so it is weaker outside code and in languages other than English. For general use, take a full
+model (Q2_0, IQ2_XS or IQ3_S) whenever your memory allows.
## How fast is each size
@@ -68,7 +76,7 @@ in a pull request.
| Model | RAM+VRAM Requirements | Speed | Quality |
| --- | ---: | --- | --- |
| **Q2_0** | 37.6 GB | fastest | good |
-| **IQ2_XS** | 39.2 GB | fast | better (**recommended**) |
+| **IQ2_XS** | 39.2 GB | fast | better (**recommended** at 40 GB, as Swift 1.5) |
| **IQ3_XXS** | 47.0 GB | slower | great |
| **IQ3_S** | 54.8 GB | slowest | best: matches the full model on the published tests |
diff --git a/setup.py b/setup.py
index f54d89072..5fd9a0fdc 100644
--- a/setup.py
+++ b/setup.py
@@ -2550,10 +2550,40 @@ def mem_have(model: str, ram: float, vram: float = 0.0) -> float:
return ram + vram if MODELS[model].get("mem_total") else ram
+# 0.2.0, the owner's memory tiers (README "The other sizes"): RAM + VRAM together pick the size. The floors sit 2 GB
+# below the tiers' names (40 / 48 / 64 GB) because the OS reports a little less than the installed memory. 64 GB and
+# up: IQ3_S (the original; Swift 1.5's IQ3_S is an equal choice), 48 GB: IQ3_XXS, 40 GB: Swift 1.5's IQ2_XS, with Gyro-M
+# (35 GB) as the other choice. Below 40 GB the older rules stand (Gyro-S, or the Coder for code). The Mac has its own
+# rules (strata_control/mac_fit.py). tools/strata_mcp.py's recommend() holds the same floors (the app and `strata launch`).
+MEMORY_TIERS = ((62.0, "qwen", "IQ3_S"), (46.0, "qwen", "IQ3_XXS"), (38.0, "swift", "IQ2_XS"))
+
+
+def tier_pick(ram: float, vram: float = 0.0):
+ """(family, size) for this PC's RAM + VRAM by MEMORY_TIERS, or None below the lowest tier (then the older rules pick:
+ Gyro-S, or the Coder). With less RAM than the size's own figure the low-RAM mode starts by itself and the model is
+ slower. A recommendation: every size stays a choice."""
+ for floor, fam, m in MEMORY_TIERS:
+ if ram + vram >= floor:
+ return fam, m
+ return None
+
+
+def tier_note(ram: float, vram: float = 0.0, gyro: bool = True) -> str:
+ """The sentence under the model menu for the lowest tier: Gyro-M is the other choice there (NVIDIA and AMD only)."""
+ pick = tier_pick(ram, vram)
+ if not gyro or pick != ("swift", "IQ2_XS"):
+ return ""
+ return ("this PC's RAM + VRAM is in the 40 GB tier: Swift 1.5's IQ2_XS is the pick; Gyro-M (35 GB) is the other choice, "
+ "close to IQ3_XXS in quality but 17-21% slower at writing answers than Q2_0")
+
+
def recommended_size(family: str, names: list, ram: float, vram: float = 0.0) -> str:
- """The size the menu pre-selects (and --yes takes): the family's recommended size when this PC's RAM fits it
- (UD-Q4_K_XL before UD-IQ4_XS), else the largest size that fits, else the first. Recommend, never force: every
- other size stays a choice."""
+ """The size the menu pre-selects (and --yes takes): the memory tier's size (RAM + VRAM, MEMORY_TIERS) when this
+ family has it, else the family's recommended size when this PC's RAM fits it (UD-Q4_K_XL before UD-IQ4_XS), else
+ the largest size that fits, else the first. Recommend, never force: every other size stays a choice."""
+ tier = tier_pick(ram, vram)
+ if tier and tier[1] in names and family in ("qwen", "swift"):
+ return tier[1]
for m in RECOMMENDED_SIZES.get(family, ()):
if m in names and mem_have(m, ram, vram) >= MODELS[m]["ram_gb"]:
return m
@@ -6990,7 +7020,8 @@ def main() -> int:
if a.family:
family = a.family
else:
- rec_fam = STRIX_HALO_FAMILY if strix_halo_recommends(gpu, ram) and STRIX_HALO_FAMILY in fams else fams[0]
+ tier = tier_pick(ram, low_ram_vram(gpu))
+ rec_fam = STRIX_HALO_FAMILY if strix_halo_recommends(gpu, ram) and STRIX_HALO_FAMILY in fams else tier[0] if tier and tier[0] in fams else fams[0]
for i, f in enumerate(fams, 1):
d = FAMILIES[f]
say(f" {i}) {d['title']:20s} {d['by']} - {d['about']}" + (" [experimental]" if d.get("experimental") else "")
@@ -7000,6 +7031,8 @@ def main() -> int:
d = OWN_PATH_FAMILIES[f]
say(f" {j}) {d['title']:20s} {d['by']} - {d['about']} [new]")
allf = fams + list(OWN_PATH_FAMILIES)
+ if tier_note(ram, low_ram_vram(gpu)) and rec_fam == "swift":
+ say(" " + tier_note(ram, low_ram_vram(gpu)))
family = allf[int(ask("Which model?", [str(i) for i in range(1, len(allf) + 1)],
str(allf.index(rec_fam) + 1), a.yes)) - 1]
if family in OWN_PATH_FAMILIES: # Qwen3.6: its own engine path, its own steps 2-7
@@ -7048,7 +7081,7 @@ def main() -> int:
fit = (f" <- fits in the low-RAM mode (the GPU holds ~{100 * low_ram_gpu_share(m, low_ram_vram(gpu)):.0f}%, "
+ ("the rest in RAM)" if low_ram_resident(m, ram, low_ram_vram(gpu)) else "the rest from the SSD)"))
say(f" {i}) {m:8s}{size_tag(family, m, installed)} {d['about']}; download {d['download_gb']:.0f} GB, uses ~{d['arena_gb']:.0f} GB of RAM{fit}")
- rec = str(names.index(installed or recommended_size(family, names, ram, gpu["vram_gb"])) + 1)
+ rec = str(names.index(installed or recommended_size(family, names, ram, low_ram_vram(gpu))) + 1)
model = a.model or names[int(ask("Which size?", [str(i) for i in range(1, len(names) + 1)], rec, a.yes)) - 1]
budget, q4_split = None, False
in_budget = bool(MODELS[model].get("budget")) and unsloth_budget(model, ram, gpu, a.resident_budget_gib)
diff --git a/tests/terminal/test_launch.py b/tests/terminal/test_launch.py
index adbec8833..38441b557 100644
--- a/tests/terminal/test_launch.py
+++ b/tests/terminal/test_launch.py
@@ -99,7 +99,18 @@ class CatalogRules(unittest.TestCase):
for r in rows:
self.assertAlmostEqual(r.download_gb, models[r.model]["download_gb"])
rec = [r for r in rows if r.recommended]
- self.assertEqual([(r.family, r.model) for r in rec], [("qwen", "IQ3_XXS")])
+ self.assertEqual([(r.family, r.model) for r in rec], [("qwen", "IQ3_S")]) # 64 + 12 GB: the 64 GB tier
+
+ def test_the_recommendation_follows_the_memory_tiers(self):
+ # the owner's tiers count RAM + VRAM: 44 GB is the 40 GB tier (Swift 1.5 IQ2_XS, Gyro-M named), 60 GB the 48 GB tier
+ from tests.terminal.launch_support import HW_32GB_RAM
+ with Home():
+ rec = catalog(HW_32GB_RAM).recommendation(HW_32GB_RAM)
+ self.assertEqual((rec["family"], rec["model"]), ("swift", "IQ2_XS"))
+ self.assertIn("Gyro-M", rec["why"])
+ hw = {**HW_32GB_RAM, "ram_gb": 48.0}
+ rec = catalog(hw).recommendation(hw)
+ self.assertEqual((rec["family"], rec["model"]), ("qwen", "IQ3_XXS"))
def test_fit_on_matches_native_fit_ts(self):
# apps/host/src/lib/native-fit.ts: memory = experts + 1; GPU holds experts up to vram - 5; RAM holds up to ram - 10; the rest spills
@@ -668,7 +679,7 @@ class FlagsCommand(unittest.TestCase):
def test_defaults_come_from_the_recommendation(self):
p = self.plan()
s = p["settings"]
- self.assertEqual((s["family"], s["model"], s["context"], s["access"], s["tune"]), ("qwen", "IQ3_XXS", 32768, "local", "skip"))
+ self.assertEqual((s["family"], s["model"], s["context"], s["access"], s["tune"]), ("qwen", "IQ3_S", 32768, "local", "skip"))
self.assertTrue(any("picked" in w for w in p["warnings"]))
def test_family_and_kv_and_gpu_and_data_dir_and_set(self):
diff --git a/tools/strata_mcp.py b/tools/strata_mcp.py
index 21ef92b79..5ee96079a 100644
--- a/tools/strata_mcp.py
+++ b/tools/strata_mcp.py
@@ -117,6 +117,7 @@ FALLBACK_FAMILIES = {
"tag": "nvfp4-"},
}
OWN_PATH_ENGINE = (0, 1, 45) # the first engine that runs Qwen3.6-35B-A3B with pictures and parallel requests (setup.py MIN_ENGINE note)
+MEMORY_TIERS = ((62.0, "qwen", "IQ3_S"), (46.0, "qwen", "IQ3_XXS"), (38.0, "swift", "IQ2_XS")) # as setup.py's: RAM + VRAM floors of the owner's 64 / 48 / 40 GB tiers
GYRO_ROOM_GB = 4.0 # recommend() suggests a Gyro size only when VRAM + RAM exceed its memory figure by this much
FALLBACK_CONTEXTS = [8192, 32768, 65536, 131072, 204800, 262144, 393216, 524288]
BENCH_PROMPT = ("Write a short story (about 300 words) about a lighthouse keeper who finds a message in a bottle. "
@@ -1015,7 +1016,23 @@ class Strata:
except Exception: # noqa: BLE001
return False
notes = []
- if ram >= 60:
+ # 0.2.0, the owner's memory tiers (README "The other sizes"; setup.py MEMORY_TIERS holds the same floors): RAM + VRAM
+ # together. An APU's shared memory is the RAM itself, so only its BIOS carve-out counts as VRAM.
+ tvram = float(best.get("dedicated_gb", 0.0)) if best.get("uma") else float(vram)
+ tier = next(((f, n) for lo, f, n in MEMORY_TIERS if ram + tvram >= lo), None)
+ if tier and tier[1] in models and tier[0] in families:
+ fam, model = tier
+ total = f"{ram:.0f} GB of RAM + {tvram:.0f} GB of VRAM = {ram + tvram:.0f} GB"
+ if model == "IQ2_XS":
+ why = (f"{total}, the 40 GB tier: Swift 1.5's IQ2_XS (thinks for a shorter time)"
+ + ("; Gyro-M (35 GB) is the other choice there, close to IQ3_XXS in quality but 17-21% slower at "
+ "writing answers than Q2_0" if "Gyro-M" in models and best.get("vendor") in ("nvidia", "amd") else ""))
+ elif model == "IQ3_XXS":
+ why = f"{total}, the 48 GB tier: IQ3_XXS (better quality, slower)"
+ else:
+ why = (f"{total}, the 64 GB tier: IQ3_S (matches the full model, the slowest); Swift 1.5's IQ3_S is an "
+ "equal choice")
+ elif ram >= 60:
fam, model, why = "qwen", "IQ3_XXS", f"{ram:.0f} GB of RAM: setup's own pick from 60 GB (better quality)"
elif ram >= models["Q2_0"]["ram_gb"] - 4:
fam, model, why = "qwen", "Q2_0", f"{ram:.0f} GB of RAM: the full model's fastest size fits"
diff --git a/tools/test_setup_gyro.py b/tools/test_setup_gyro.py
index b9d372d10..08694fdec 100644
--- a/tools/test_setup_gyro.py
+++ b/tools/test_setup_gyro.py
@@ -58,12 +58,16 @@ class GyroCatalog(unittest.TestCase):
def test_mcp_recommends_gyro_only_where_nothing_else_fits(self):
s = M.Strata(ROOT)
- for ram, vram, family in ((64.0, 12.0, "qwen"), (48.0, 12.0, "qwen"), (32.0, 12.0, "coder")):
+ for ram, vram, family in ((64.0, 12.0, "qwen"), (48.0, 12.0, "qwen"), (32.0, 12.0, "swift"), (28.0, 8.0, "coder")):
self.assertEqual(s.recommend(hardware(ram, vram))["family"], family, (ram, vram))
got = s.recommend(hardware(20.0, 12.0)) # 32 GB in total, the Coder's 32 GB of RAM missing
self.assertEqual((got["family"], got["model"]), ("gyro", "Gyro-S"))
- got = s.recommend(hardware(24.0, 16.0)) # 40 GB in total: the Coder fits in the low-RAM mode, so it wins
- self.assertEqual((got["family"], got["model"]), ("coder", "IQ1_M"))
+ got = s.recommend(hardware(24.0, 16.0)) # 40 GB in total: the owner's 40 GB tier, Swift 1.5's IQ2_XS
+ self.assertEqual((got["family"], got["model"]), ("swift", "IQ2_XS"))
+ self.assertIn("Gyro-M", got["why"]) # named as the other choice (NVIDIA and AMD)
+ intel = hardware(24.0, 16.0)
+ intel["gpus"] = [{**g, "vendor": "intel"} for g in intel["gpus"]]
+ self.assertNotIn("Gyro", s.recommend(intel)["why"]) # never Intel Arc
self.assertIsNone(s.recommend(hardware(12.0, 8.0))["family"])
self.assertIsNone(s.recommend(hardware(16.0, 11.0))["family"]) # 27 GB: below Gyro-S's 27.6 + room
diff --git a/tools/test_setup_menu.py b/tools/test_setup_menu.py
index f6778a821..1f1768363 100644
--- a/tools/test_setup_menu.py
+++ b/tools/test_setup_menu.py
@@ -109,7 +109,11 @@ class RecommendedSizes(unittest.TestCase):
def test_without_the_ram_the_largest_that_fits_and_iq3_s_stays_marked(self):
n = self.names("qwen")
self.assertEqual(setup.recommended_size("qwen", n, 61.0), "IQ3_XXS") # IQ3_S needs 62 GB
- self.assertEqual(setup.recommended_size("qwen", n, 50.0), "IQ2_XS")
+ self.assertEqual(setup.recommended_size("qwen", n, 50.0), "IQ3_XXS") # 48 GB tier (RAM + VRAM)
+ self.assertEqual(setup.recommended_size("qwen", n, 40.0, 8.0), "IQ3_XXS") # 40 + 8 = 48 GB together
+ self.assertEqual(setup.recommended_size("qwen", n, 32.0, 12.0), "IQ2_XS") # 44 GB: the 40 GB tier
+ self.assertEqual(setup.recommended_size("swift", self.names("swift"), 32.0, 12.0), "IQ2_XS")
+ self.assertEqual(setup.recommended_size("qwen", n, 32.0, 32.0), "IQ3_S") # 64 GB together
self.assertEqual(setup.size_tag("qwen", "IQ3_S"), " (recommended)")
def test_coder_has_nothing_marked(self):
diff --git a/tools/test_strata_mcp.py b/tools/test_strata_mcp.py
index c0e87dfa7..9f2df0943 100644
--- a/tools/test_strata_mcp.py
+++ b/tools/test_strata_mcp.py
@@ -242,7 +242,7 @@ class Protocol(FakeRoot):
call = r[3]["result"]
self.assertFalse(call["isError"])
self.assertIn("families", call["structuredContent"])
- self.assertEqual(json.loads(call["content"][0]["text"])["recommendation"]["model"], "IQ3_XXS")
+ self.assertEqual(json.loads(call["content"][0]["text"])["recommendation"]["model"], "IQ3_S")
self.assertEqual(r[4]["result"], {})
self.assertNotIn(None, r) # the notification got no answer
@@ -377,7 +377,7 @@ class Install(FakeRoot):
res, err = self.call("strata_install", {})
self.assertFalse(err)
self.assertIn("Nothing was done yet", res["summary"])
- self.assertEqual((res["plan"]["family"], res["plan"]["model"], res["plan"]["context"]), ("qwen", "IQ3_XXS", 32768))
+ self.assertEqual((res["plan"]["family"], res["plan"]["model"], res["plan"]["context"]), ("qwen", "IQ3_S", 32768))
self.assertFalse((self.root / "setup-args.json").exists())
self.assertIsNone(self.strata.install_job())
@@ -641,7 +641,7 @@ class Helpers(unittest.TestCase):
def test_recommendation_follows_ram(self):
s = M.Strata(HERE.parent)
gpu = [{"index": 0, "name": "x", "vram_gb": 24.0, "usable": True, "vendor": "nvidia"}]
- cases = {96: ("qwen", "IQ3_XXS", 131072), 48: ("qwen", "Q2_0", 131072), 32: ("coder", "IQ1_M", 131072)}
+ cases = {96: ("qwen", "IQ3_S", 131072), 48: ("qwen", "IQ3_S", 131072), 32: ("qwen", "IQ3_XXS", 131072), 12: ("coder", "IQ1_M", 131072)} # the owner's tiers count RAM + VRAM (24 GB card): 120, 72, 56 GB; the Coder where the older rules stand
for ram, want in cases.items():
r = s.recommend({"ram_gb": ram, "gpus": gpu, "cpu": {"avx2": True}})
self.assertEqual((r["family"], r["model"], r["context"]), want, ram)
@@ -732,7 +732,7 @@ class MacCatalog(unittest.TestCase):
s = M.Strata(HERE.parent)
r = s.recommend({"ram_gb": 96, "gpus": [{"index": 0, "name": "x", "vram_gb": 24.0, "usable": True, "vendor": "nvidia"}],
"cpu": {"avx2": True}})
- self.assertEqual((r["family"], r["model"]), ("qwen", "IQ3_XXS"))
+ self.assertEqual((r["family"], r["model"]), ("qwen", "IQ3_S"))
self.assertIn("unsloth", [f["family"] for f in s.model_table({"ram_gb": 96, "gpus": []})])