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++

merge f-145-int 194c26881 (model-card sampling defaults, web sampling, answer room, thinking budget) into f-split-int

Niko122107fe758ac916
diff --cc serve/server.py
index 606b70260,252d93681..dabd9a911
--- a/serve/server.py
+++ b/serve/server.py
@@@ -7584,22 -6853,9 +7617,24 @@@ def main(argv=None, *, lifecycle_event=
model_name=default_model_name(cfg), vision=vision,
sampling_defaults=sampling_defaults,
fit_max_tokens=a.fit_max_tokens or cfg.get("fit_max_tokens") is True)
+ svc.launch_attestation = launch_attestation
+ svc.offline = a.offline
+ svc.max_queue = max_queue
+ svc.queue_timeout_s = float(queue_timeout) if queue_timeout is not None else None
- if a.engine == "strata":
+ if a.engine == "strata": # the model card's sampling when nothing else is said (see MODEL_CARD_SAMPLING)
+ svc.model_arch = cfg.get("arch") or "qwen4exp"
+ requested = cfg.get("parallel")
+ if requested is None:
+ args = cfg.get("args", [])
+ for flag in ("--batch", "--slots"):
+ if flag in args and args.index(flag) + 1 < len(args):
+ try:
+ requested = int(args[args.index(flag) + 1])
+ except ValueError:
+ pass
+ per_replica = requested if isinstance(requested, int) and not isinstance(requested, bool) and requested > 0 else 1
+ svc.requested_slots = per_replica * len(groups) if groups else per_replica
+ svc.answer_room = cfg.get("answer_room") is not False
svc.reasoning_close_retry = cfg.get("reasoning_close_retry") is True # #1053: opt-in, off by default
if cfg.get("images"): # image generation (serve/images.py), set up by setup --images
icfg = {k: (os.path.abspath(os.path.join(cfg.get("cwd") or ".", v))