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++

setup: pack recipe states --dspark-noise-token 128799 and --dspark-markov-rank 256; plain wording (nothing to download, one unit: tokens a second, the one-time SHA-256 check and its time, the remembered-folder line)

Niko12211a7b4ae236b3
diff --git a/setup.py b/setup.py
index 67083fd87..cc58ea269 100644
--- a/setup.py
+++ b/setup.py
@@ -706,7 +706,7 @@ DSV41_CONTEXT = 8192 # the context the engine was checked at (4096 and
MULTI_FILE_FAMILIES = {
"deepseek41": {"title": "DeepSeek-V4.1-Flash", "by": "DeepSeek; the 2-bit GGUFs by IST-DAS, UD-Q8_K_XL by Unsloth", "arch": "deepseek41",
"about": "a big model that runs on a PC with a 12 GB card and 64 GB of RAM (the two official 2-bit versions: 257 or "
- "274 GB of disk, about 5-7 words a second) or, in Unsloth's 508 GB version, on several graphics cards",
+ "274 GB of disk, about 5-7 tokens a second) or, in Unsloth's 508 GB version, on several graphics cards",
"hf": hf(DSV41_REPO) + "UD-Q8_K_XL/", "tag": "deepseek-v4.1-flash-", "name": "deepseek-v4.1-flash",
"license": "DeepSeek's model license: https://huggingface.co/unsloth/DeepSeek-V4.1-Flash-1017-GGUF",
"models": {"UD-Q8_K_XL": {"about": "Unsloth Dynamic 8-bit: the experts in MXFP4 as DeepSeek made them, the rest "
@@ -720,7 +720,7 @@ MULTI_FILE_FAMILIES["deepseek41"]["models"] = {
"IST-DAS-2.00bpw": {
"label": "IST-DAS 2.00 bpw",
"about": "the smaller official version: 138 GB of model + 111 GB n-gram table + 8 GB drafter (257 GB of disk). Runs on a PC with "
- "a 12 GB card and 40-64 GB of RAM at about 5-7 words a second; quality about 16% above the original model's "
+ "a 12 GB card and 40-64 GB of RAM at about 5-7 tokens a second; quality about 16% above the original model's "
"perplexity",
"ram_gb": 40, "recipe": "ista", "subdir": "2.00bpw", "profile": "profile-2.00.bin", "hf": hf(DSV41_ISTA_REPO) + "2.00bpw/",
"shards": [DSV41_ISTA_SHARD.format(bpw="2.00bpw", i=i) for i in range(1, 5)],
@@ -5638,7 +5638,7 @@ def data_folder(requested: str | None, move: bool = False, yes: bool = True, set
["keep", "move"], "keep", False) == "move"
if held and not move:
for folder in held:
- say(f" your model files stay in {folder}; {dest} is used for new downloads")
+ say(f" Models you already have in {folder} stay there; new downloads go to {dest}.")
if set_default:
save_settings({**settings, "data_dir": str(dest)})
return dest, held
@@ -7338,10 +7338,12 @@ def dsv41_find_shared(a, name: str, folder: Path, shared: Path) -> Path:
def dsv41_pack_cmd(folder: Path, shards: list, engram: Path, meta: Path, dspark: Path, pack: Path) -> list:
"""The pack recipe for an official IST-DAS set - this and nothing else: another one (--skip-tensors / --dense-override + engram_fold)
- packs without an error and gives garbage text (P2003, 2026-10-23). --shards names the four files, so an Engram or another
+ packs without an error and gives garbage text (P2003, 2026-10-23). The drafter's noise token (128799) and markov rank (256) are stated:
+ the IST-DAS GGUFs do not carry them, and a pack with noise id 0 drafts deep less often (decode ~10% slower with DSpark). --shards names the four files, so an Engram or another
set in the same folder is not read as part of the model."""
return [sys.executable, str(ROOT / "tools" / "dsv41" / "pack_gguf.py"), "--model", str(folder), "--shards", *[Path(n).name for n in shards],
- "--engram", str(engram), "--engram-meta", str(meta), "--dspark-gguf", str(dspark), "--out", str(pack)]
+ "--engram", str(engram), "--engram-meta", str(meta), "--dspark-gguf", str(dspark),
+ "--dspark-noise-token", "128799", "--dspark-markov-rank", "256", "--out", str(pack)]
def dsv41_engine_env(lib_dirs) -> dict:
@@ -7437,6 +7439,14 @@ def install_dsv41(a, family, gpu, hip, amd, cuda_tk, multi, port, ram, adopted,
if left:
say(f" Download: {left:.0f} GB to fetch (checked against pinned sizes and SHA-256; it resumes if it stops). At 50 MB/s that "
f"is {left * 1e9 / 50e6 / 3600:.1f} hours. Files you already have: --gguf-dir DIR.")
+ else:
+ say(" Nothing to download: the files are already here.")
+ unchecked = [f for f in [*shards, *shared] if f.exists() and f"sha256 {fam['sha256'][f.name][1]}" not in
+ (f.with_name(f.name + ".done").read_text(encoding="utf-8", errors="replace") if done(f) else "")]
+ if unchecked:
+ gb = sum(fam["sha256"][f.name][0] for f in unchecked) / 1e9
+ say(f" The files already here ({gb:.0f} GB) are checked against their SHA-256 once, now: about {max(1, round(gb / 55))} minutes on a fast "
+ "drive (4-5 minutes for the 257 GB of 2.00), longer on a slow one. The next runs skip it.")
for w in plan["warnings"]:
if "free disk space" in w:
continue # the disk is a hard limit, said below
@@ -8260,7 +8270,7 @@ def main() -> int:
for mname in ("IST-DAS-2.00bpw", "IST-DAS-2.25bpw", "UD-Q8_K_XL"):
ds = dsv41_fit.plan(cards_ok, ram, None, model=mname)
tail = {"fits": f"fits ({ds['experts_in_vram']:,} of {ds['experts_total']:,} experts in graphics memory, the rest on the CPU)",
- "tight": ("runs, with the experts RAM cannot hold read from the SSD (about 5-7 words a second measured)"
+ "tight": ("runs, with the experts RAM cannot hold read from the SSD (about 5-7 tokens a second measured)"
if ds.get("variant") else "fits, but RAM is tight"),
"spills": (f"works but slow: less RAM than it was measured with (46-64 GB)" if ds.get("variant")
else f"about {ds['spill_gb']:.0f} GB of its experts would be read from the SSD (slow)"),
diff --git a/strata_control/dsv41_fit.py b/strata_control/dsv41_fit.py
index e8ed199a4..fea934cdf 100644
--- a/strata_control/dsv41_fit.py
+++ b/strata_control/dsv41_fit.py
@@ -227,7 +227,7 @@ def _plan_ista(model: str, cards: list[dict] | None, ram_gb: float | None, disk_
elif float(ram_gb) >= v["ram_ok_gb"]:
out["verdict"] = "tight"
lines.append(f"This computer has {ram_gb:.0f} GB of RAM: Strata keeps the experts you use most in RAM (it sizes this itself) and "
- "reads the rest from the SSD. Measured: about 4.7 words a second on 64 GB of RAM and a 12 GB card, about 7 on "
+ "reads the rest from the SSD. Measured: about 4.7 tokens a second on 64 GB of RAM and a 12 GB card, about 7 on "
"46 GB with a faster drive. More RAM or a faster NVMe drive makes it quicker.")
else:
out["verdict"] = "spills"
diff --git a/tools/strata_mcp.py b/tools/strata_mcp.py
index 5d0171e41..07e9d18ee 100644
--- a/tools/strata_mcp.py
+++ b/tools/strata_mcp.py
@@ -1247,7 +1247,7 @@ class Strata:
return (f"slow here: this computer has {pl['ram_gb']:.0f} GB of RAM and the set was measured with 46-64 GB, so most of its "
"experts are read from the SSD; it can still be installed")
if v == "tight":
- return ("tight: runs, with the experts RAM cannot hold read from the SSD (about 5-7 words a second measured on 46-64 GB "
+ return ("tight: runs, with the experts RAM cannot hold read from the SSD (about 5-7 tokens a second measured on 46-64 GB "
"of RAM and a 12 GB card)")
if v == "fits":
return (f"fits: {pl['experts_in_vram']:,} of {pl['experts_total']:,} experts in graphics memory, the rest cached in RAM "
diff --git a/tools/test_setup_dsv41.py b/tools/test_setup_dsv41.py
index c02bb4e6e..da81c54c8 100644
--- a/tools/test_setup_dsv41.py
+++ b/tools/test_setup_dsv41.py
@@ -360,7 +360,8 @@ class OfficialSets(unittest.TestCase):
cmd = setup.dsv41_pack_cmd(folder, names, Path("/s/engram-IQ4_NL.gguf"), Path("/s/meta.gguf"), Path("/s/dspark.gguf"), pack)
self.assertTrue(cmd[1].endswith("pack_gguf.py"))
self.assertEqual(cmd[2:], ["--model", str(folder), "--shards", *names, "--engram", str(Path("/s/engram-IQ4_NL.gguf")),
- "--engram-meta", str(Path("/s/meta.gguf")), "--dspark-gguf", str(Path("/s/dspark.gguf")), "--out", str(pack)])
+ "--engram-meta", str(Path("/s/meta.gguf")), "--dspark-gguf", str(Path("/s/dspark.gguf")),
+ "--dspark-noise-token", "128799", "--dspark-markov-rank", "256", "--out", str(pack)])
for bad in ("--skip-tensors", "--dense-override", "--override-experts", "--tq-experts"): # these give garbage text on the official sets
self.assertNotIn(bad, cmd)