Upload 18 files

app.py

@@ -82,12 +82,12 @@ def ui_explain_detail(ticker):
               f"FACT LINE:\n{raw}")
     if chain_core:
         prompt += f"\n\nRULING CHAIN (translate & synthesize, don't quote):\n{chain_core}"
-    yield "🤖 _Translator sub-agent (Qwen3-1.7B · llama.cpp) is summarizing…_"
+    yield "🤖 _Summary Sub-Agent (Chan-Tuned Qwen3-1.7B · llama.cpp) is summarizing…_"
     final = ""
     for acc in llm_local.chat_stream(prompt, max_tokens=260, temperature=0.2,
                                      worker="translator"):
         final = acc
-        yield "🤖 **AI narrative (Translator sub-agent · Qwen3-1.7B):**\n\n" + acc
+        yield "🤖 **AI Summary (Summary Sub-Agent · Chan-Tuned Qwen3-1.7B):**\n\n" + acc
     # capture (raw read → narrative) as a fine-tuning pair (🎯 Well-Tuned)
     try:
         import finetune_data
@@ -162,7 +162,7 @@ def _export_dataset():
     if n == 0:
         return ("⚠️ **0 training pairs captured yet.** Pairs are saved only when "
                 "the **Signals → AI summary** finishes with a model loaded. Steps: "
-                "1) Model tab — wait for the Translator sub-agent to show ✅; "
+                "1) Model tab — wait for the Summary sub-agent to show ✅; "
                 "2) Signals — Run analysis, pick a ticker, click **AI summary**, "
                 "let it finish; repeat a few times; 3) come back and Export.",
                 gr.update(visible=False))
@@ -243,7 +243,7 @@ with gr.Blocks(title="Chan Compass · US", **_style_kw) as demo:
         sig_table = gr.Dataframe(label="Tomorrow's plan — long-hold mode (sorted: BUY → SELL → HOLD → WAIT)",
                                  interactive=False, wrap=True)
         gr.Markdown("**Stock summary** — pick a ticker for a plain-English raw read, "
-                    "then let the Translator sub-agent write an AI narrative.",
+                    "then let the Summary sub-agent write an AI Summary.",
                     elem_classes=["s2-footnote"])
         with gr.Row():
             detail_pick = gr.Dropdown(choices=[], label="Ticker", scale=2)
@@ -341,7 +341,7 @@ with gr.Blocks(title="Chan Compass · US", **_style_kw) as demo:
                     "and nothing leaves the machine. **First load installs the llama.cpp "
                     "runtime + downloads the GGUF (one-time, usually 1–3 min; worst case "
                     "~15 min if it has to compile).** Signals/rotation/news never depend on it.")
-        gr.Markdown("**Sub-agent pool:** `fast` Translator/Narrator (Qwen3-1.7B, "
+        gr.Markdown("**Sub-agent pool:** Summary Sub-Agent (Chan-Tuned Qwen3-1.7B, "
                     "fixed) handles Explain / rotation narrative / news briefs; "
                     "`deep` Analyst writes research reports. Each has its own lock — "
                     "they run in parallel. Pick the Analyst model below:")

llm_local.py

@@ -4,7 +4,7 @@ llm_local.py — local sub-agent pool (llama.cpp runtime, no cloud APIs).
 Two independent model instances ("sub-agents"), each with its own lock, so
 features never block each other with "model busy":
 
-    fast  · Translator/Narrator — Qwen3-1.7B Q8 (≈2 GB RAM)
+    fast  · Summary Sub-Agent (Chan-Tuned Qwen3-1.7B)
             → Explain-in-English, sector-rotation narrative, news briefs.
             Small = quick CPU prefill, answers start streaming in seconds.
     deep  · Analyst — Qwen3-4B Q4_K_M by default (swappable in the Model tab)
@@ -25,7 +25,7 @@ from huggingface_hub import hf_hub_download
 
 # name -> (HF repo, gguf filename)
 MODEL_ZOO = {
-    "Chan-Tuned Qwen3-1.7B · my fine-tune (Well-Tuned)": (
+    "Chan-Tuned Qwen3-1.7B · my fine-tune": (
         "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"),
@@ -36,10 +36,10 @@ MODEL_ZOO = {
     "Qwen3-14B · max quality (still far under 32B cap)": (
         "Qwen/Qwen3-14B-GGUF", "Qwen3-14B-Q4_K_M.gguf"),
 }
-# Only the Translator sub-agent (Signals · Explain) uses the published
+# Only the Summary 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)"
+TRANSLATOR_MODEL = "Chan-Tuned Qwen3-1.7B · my fine-tune"
 DEFAULT_MODEL = "Qwen3-4B  · default — fast + smart, still ≤4B"
 
 _THINK_RE = re.compile(r"<think>.*?</think>", re.S)
@@ -50,7 +50,7 @@ _NCPU = max(2, (os.cpu_count() or 4))
 # Three tiny 1.7B instances share ONE GGUF file on disk (~2 GB RAM each) and
 # the 4B Analyst writes reports. Total ≈ 9 GB on a 32 GB Space.
 WORKER_LABEL = {
-    "translator": "Translator sub-agent (Signals · Explain)",
+    "translator": "Summary Sub-Agent (Chan-Tuned Qwen3-1.7B)",
     "narrator":   "Narrator sub-agent (Sector Rotation)",
     "reporter":   "Reporter sub-agent (News · Research support)",
     "analyst":    "Analyst sub-agent (Auto Research)",