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++
Server: the card roles (and a stop when every card is Off) come before anything starts
Niko12217c11a196d55d
diff --git a/docs/TERMINAL_LAUNCH.md b/docs/TERMINAL_LAUNCH.mdindex 23444e1d1..3061efbfc 100644--- a/docs/TERMINAL_LAUNCH.md+++ b/docs/TERMINAL_LAUNCH.md@@ -22,7 +22,7 @@ The release folder is the latest GitHub release of Niko1221/Strata (`https://git1. **Start**: saved presets (name, model, context, parallel requests, tuned or not, last used) or "Set up from the beginning" or "Use model files I already have".2. **Model**: the catalog of the app and `setup.py` (one source), with download size, what it needs (graphics memory + RAM) and "fits / tight / does not fit this computer".A warning, never a block. The n-gram table stays on your SSD and is not counted. `o` = your own files (a folder or a .gguf; checked with the fingerprint table).- *(Cards: only with two or more usable graphics cards.)*+ *(Cards: only with two or more usable graphics cards. Each card gets a role: M Main, H Helper, O Off, R back to the recommendation, D details. `--gpu-roles 1=main,0=helper` does the same from the command line; see [MULTI_GPU.md](MULTI_GPU.md).)*3. **Folder**: where model files are stored, free space shown. For a model that is already installed (also one the desktop app installed) it starts where the model is, says"Already downloaded" and lists only what is really missing with its size (the draft layer, about 5.2 GB; the picture reader, 0.9 GB). It never says "download the model" when onlya helper file is missing.diff --git a/serve/server.py b/serve/server.pyindex c212ae249..c76f4899a 100644--- a/serve/server.py+++ b/serve/server.py@@ -7995,6 +7995,8 @@ def main(argv=None, *, lifecycle_event=None, interrupt_exit_code: int = 0) -> inpretty = ", ".join(f"{k}={v}" for k, v in sampling_defaults.items())print(f"[strata] sampling defaults from the config: {pretty}", flush=True)lazy = a.lazy or cfg.get("lazy_load") is True+ for line in card_roles_lines(cfg): # the card roles: said first; every card Off stops here, before anything touches a card+ print(line, flush=True)if cfg.get("vision"):elastic = cfg["vision"].get("elastic") is True and cfg.get("vram_elastic") is Trueprint("the vision encoder starts with the first picture (elastic) ..." if elastic else@@ -8015,8 +8017,6 @@ def main(argv=None, *, lifecycle_event=None, interrupt_exit_code: int = 0) -> inprint("model unloaded; the first request loads it ..." if lazy else"loading the model (the first start takes a minute or two) ...", flush=True)warn_budget_over_ram(engine_args(cfg) if "args" in cfg else []) # #1080- for line in card_roles_lines(cfg):- print(line, flush=True)note = card_note(cfg, env)if note:print(note, flush=True)diff --git a/strata_cli/launch/params.py b/strata_cli/launch/params.pyindex fb87cd77d..7297d90c9 100644--- a/strata_cli/launch/params.py+++ b/strata_cli/launch/params.py@@ -40,6 +40,7 @@ PLAIN = {"--suffix-draft": "Guess the next words from text already in the conversation.","--port": "The network port the server listens on.","--gpu": "Which graphics card to use.",+ "--gpu-roles": "A role for each graphics card: main, helper or off.","--gpus": "Which graphics cards to use together.",}_CODE = re.compile(r"\b[RTS]\d{1,2}\b[:,)]?\s*")