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: drop --prefill-chunk (the chunked prompt is the engine's default now); doc line

Niko122110af5bb1e43a
diff --git a/setup.py b/setup.py
index f9795ee99..2f11e46a0 100644
--- a/setup.py
+++ b/setup.py
@@ -7529,13 +7529,12 @@ 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. --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). --dspark-conf 0.65: the
+ # engine's default), and <pack>/expert-profile.bin is read from the pack folder without a flag. The layer-major chunked prompt is the engine's own default now (f-dsv41-rel d6f18babe: 1024 on a single card,
+ # its own choice for several), so no --prefill-chunk is written. --dspark-conf 0.65: the
# drafts' confidence floor (H100 replay, DSPARK-DEFAULTS.md: within 0.01-0.02 of the per-class best, never below 0.3). The engine has
# no flag for the draft block (3 in the same table): it reads the block size from the pack, so only the floor is written.
args = ["--pack", str(pack), "--max-context", str(ctx), "--eos", "1", "--dspark", "auto",
- *(["--prefill-chunk", "1024", "--dspark-conf", "0.65"] if md.get("recipe") == "ista" else []), *plan["args"]]
+ *(["--dspark-conf", "0.65"] 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