Upload 18 files

llm_local.py

@@ -25,6 +25,8 @@ from huggingface_hub import hf_hub_download
 
 # name -> (HF repo, gguf filename)
 MODEL_ZOO = {
+    "Chan-Tuned Qwen3-1.7B · my fine-tune (Well-Tuned)": (
+        "ranranrunforit/chan-compass-qwen3-1.7b-gguf", "qwen3-1.7b.Q8_0.gguf"),
     "Qwen3-1.7B · Tiny Titan (≤4B award class)": (
         "Qwen/Qwen3-1.7B-GGUF", "Qwen3-1.7B-Q8_0.gguf"),
     "Qwen3-4B  · default — fast + smart, still ≤4B": (
@@ -33,14 +35,11 @@ MODEL_ZOO = {
         "Qwen/Qwen3-8B-GGUF", "Qwen3-8B-Q4_K_M.gguf"),
     "Qwen3-14B · max quality (still far under 32B cap)": (
         "Qwen/Qwen3-14B-GGUF", "Qwen3-14B-Q4_K_M.gguf"),
-    # 🎯 Well-Tuned: after you publish your LoRA-merged GGUF, uncomment and edit
-    # this line (repo = your HF id, filename = the .gguf you uploaded). It will
-    # then appear in the Model tab as the fast Translator sub-agent.
-    "Chan-Tuned Qwen3-1.7B · my fine-tune": (
-        "ranranrunforit/chan-compass-qwen3-1.7b-gguf", "chan-compass-qwen3-1.7b-gguf.Q8_0.gguf"),
 }
-# FAST_MODEL = "Qwen3-1.7B · Tiny Titan (≤4B award class)"
-FAST_MODEL = "Chan-Tuned Qwen3-1.7B · my fine-tune"
+# Only the Translator sub-agent (Signals · Explain) uses the published
+# fine-tune; every other sub-agent stays on the stock models.
+FAST_MODEL = "Qwen3-1.7B · Tiny Titan (≤4B award class)"
+TRANSLATOR_MODEL = "Chan-Tuned Qwen3-1.7B · my fine-tune (Well-Tuned)"
 DEFAULT_MODEL = "Qwen3-4B  · default — fast + smart, still ≤4B"
 
 _THINK_RE = re.compile(r"<think>.*?</think>", re.S)
@@ -64,7 +63,7 @@ def _mk(model):
 
 
 WORKERS = {
-    "translator": _mk(FAST_MODEL),
+    "translator": _mk(TRANSLATOR_MODEL),
     "narrator":   _mk(FAST_MODEL),
     "reporter":   _mk(FAST_MODEL),
     "analyst":    _mk(DEFAULT_MODEL),