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: DeepSeek official sets start with --prefill-chunk 1024 (layer-major prompt path, measured 231 tok/s at 8K on a 12 GB PC)

Niko12219469412db20a
diff --git a/setup.py b/setup.py
index 678b5c676..d0e39efeb 100644
--- a/setup.py
+++ b/setup.py
@@ -7529,8 +7529,11 @@ def install_dsv41(a, family, gpu, hip, amd, cuda_tk, multi, port, ram, adopted,
step(7, "writing the start script")
# --dspark auto, never off: "off" still LOADS the drafter, and its 7 GB of experts take the graphics memory (found 2026-10-23);
# auto keeps it unloaded when the experts are far larger than the card. The RAM store sizes itself (--ssd-cache auto is the
- # engine's default), and <pack>/expert-profile.bin is read from the pack folder without a flag.
- args = ["--pack", str(pack), "--max-context", str(ctx), "--eos", "1", "--dspark", "auto", *plan["args"]]
+ # engine's default), and <pack>/expert-profile.bin is read from the pack folder without a flag. --prefill-chunk 1024 is the layer-major
+ # prompt path (measured: 8K prompt 231 tokens/s on a 12 GB PC against 3.7 windowed, 330-373 on 4 x R9700; text identical); the engine's
+ # own default is still windowed, so it is written here as a recommended default (the person can change it).
+ args = ["--pack", str(pack), "--max-context", str(ctx), "--eos", "1", "--dspark", "auto",
+ *(["--prefill-chunk", "1024"] if md.get("recipe") == "ista" else []), *plan["args"]]
cfg = {"exe": str(exe), "args": args, "cwd": str(ROOT), "tokenizer": str(pack / "tokenizer"),
"model_name": fam["name"], "arch": fam["arch"], "log": str(home() / f"strata-{tag.lower()}.log"),
"lib_dirs": lib_dirs, "port": port, "engine_ready_s": 3600} # 300+ GB read from the files: a slow disk needs its time
diff --git a/tools/test_setup_dsv41.py b/tools/test_setup_dsv41.py
index 70939b5a2..b6251e9b7 100644
--- a/tools/test_setup_dsv41.py
+++ b/tools/test_setup_dsv41.py
@@ -430,6 +430,7 @@ class InstallOfficial(Install):
a = self.cfg["args"]
self.assertEqual(a[a.index("--dspark") + 1], "auto") # never "off": that still loads the drafter and takes the VRAM
self.assertNotIn("off", a)
+ self.assertEqual(a[a.index("--prefill-chunk") + 1], "1024") # the layer-major prompt path
self.assertNotIn("--ssd-cache", a) # the store sizes itself
self.assertEqual(a[a.index("--max-context") + 1], "8192")
self.assertIn("checked: every file the engine needs is there", out)