docs(quickstart): load mirrored local weights (no upstream fetch)

quickstart.py

@@ -1,64 +1,53 @@
-"""
-LumynaX Reasoning QwQ-32B GGUF — LumynaX quickstart.
-
-This script fetches the upstream model from Hugging Face and runs a short
-LumynaX-flavoured prompt. Run it on a host that satisfies the resource budget
-documented in the README (LumynaX Reasoning QwQ-32B GGUF).
-
-Usage:
-    python quickstart.py                # one-shot demo prompt
-    python quickstart.py --interactive  # REPL
-    python quickstart.py --gguf         # use the GGUF mirror via llama-cpp
-
-LumynaX package repo: https://huggingface.co/AbteeXAILab/lumynax-reasoning-qwq-32b-gguf
-Upstream weights:     https://huggingface.co/Qwen/QwQ-32B
-"""
-from __future__ import annotations
-import argparse, os, sys
-
-LUMYNAX_SYSTEM = (
-    "You are LumynaX, the AbteeX AI Labs assistant from Aotearoa New Zealand. "
-    "Ko te marama te tuapapa - the light is the foundation. "
-    "Answer with care, cite uncertainty, and prefer local-first reasoning. "
-    "Refuse unsafe, unlawful, or sovereignty-violating requests."
-)
-DEMO_PROMPT = "Explain in 3 bullets why local-first AI matters for Aotearoa New Zealand."
-
-def _run_gguf(prompt: str, interactive: bool):
-    from llama_cpp import Llama
-    print("[lumynax] Loading GGUF from Qwen/QwQ-32B-GGUF (this can be large)...")
-    llm = Llama.from_pretrained(
-        repo_id="Qwen/QwQ-32B-GGUF",
-        filename="qwq-32b-q4_k_m.gguf",
-        n_ctx=16384,
-        n_gpu_layers=int(os.environ.get("N_GPU_LAYERS", "-1")),
-        verbose=False,
-    )
-    def chat(user):
-        out = llm.create_chat_completion(messages=[
-            {"role": "system", "content": LUMYNAX_SYSTEM},
-            {"role": "user",   "content": user},
-        ], max_tokens=512, temperature=0.4)
-        return out["choices"][0]["message"]["content"]
-    if interactive:
-        print("[lumynax] interactive mode — empty line exits.")
-        while True:
-            try: q = input("you> ").strip()
-            except EOFError: break
-            if not q: break
-            print("lumynax> " + chat(q))
-    else:
-        print(chat(prompt))
-
-
-def main():
-    p = argparse.ArgumentParser()
-    p.add_argument("--interactive", action="store_true")
-    p.add_argument("--prompt", default=DEMO_PROMPT)
-    p.add_argument("--gguf", action="store_true", help="kept for compatibility — this build is GGUF-only")
-    args = p.parse_args()
-    _run_gguf(args.prompt, args.interactive)
-
-
-if __name__ == "__main__":
-    main()
+"""
+Lumynax Reasoning Qwq 32B Gguf — LumynaX quickstart (clone & run).
+
+This loads the GGUF that ships with this repo. No upstream HF call needed
+once you've done `hf download AbteeXAILab/lumynax-reasoning-qwq-32b-gguf`.
+
+Usage:
+  python quickstart.py                   # one-shot demo prompt
+  python quickstart.py --interactive     # REPL
+"""
+from __future__ import annotations
+import argparse, glob, os, sys
+from pathlib import Path
+
+LUMYNAX_SYSTEM = "You are LumynaX, the AbteeX AI Labs assistant from Aotearoa New Zealand. Ko te marama te tuapapa. Answer with care; cite uncertainty; refuse unsafe asks."
+DEMO_PROMPT = "Explain in 3 bullets why local-first AI matters for Aotearoa New Zealand."
+
+# Locate the primary GGUF that was downloaded alongside this script.
+HERE = Path(__file__).resolve().parent
+PRIMARY = HERE / r"qwq-32b-q4_k_m.gguf"
+
+
+def main():
+    from llama_cpp import Llama
+    p = argparse.ArgumentParser()
+    p.add_argument("--interactive", action="store_true")
+    p.add_argument("--prompt", default=DEMO_PROMPT)
+    args = p.parse_args()
+    if not PRIMARY.exists():
+        print(f"[lumynax] primary weight file missing: {PRIMARY}", file=sys.stderr)
+        print(f"[lumynax] run: hf download AbteeXAILab/lumynax-reasoning-qwq-32b-gguf --local-dir <dir> first.", file=sys.stderr)
+        sys.exit(2)
+    print(f"[lumynax] loading {PRIMARY.name}{shard_log_suffix}")
+    llm = Llama(model_path=str(PRIMARY), n_ctx=16384,
+                n_gpu_layers=int(os.environ.get("N_GPU_LAYERS","-1")), verbose=False)
+    def chat(user):
+        out = llm.create_chat_completion(messages=[
+            {"role":"system","content":LUMYNAX_SYSTEM},
+            {"role":"user","content":user},
+        ], max_tokens=512, temperature=0.4)
+        return out["choices"][0]["message"]["content"]
+    if args.interactive:
+        print("[lumynax] interactive mode — empty line exits.")
+        while True:
+            try: q = input("you> ").strip()
+            except EOFError: break
+            if not q: break
+            print("lumynax> " + chat(q))
+    else:
+        print(chat(args.prompt))
+
+if __name__ == "__main__":
+    main()