Patch: .dockerignore, disable Dockerfile, add remote dataset loader

.dockerignore

@@ -1,4 +1,4 @@
-# datasets / binaries / artifacts
+# Keep image tiny: exclude datasets, models, artifacts, logs, media
 data/**
 savant_rrf1/**
 **/*.jsonl
@@ -23,7 +23,6 @@ savant_rrf1/**
 # caches & venvs
 __pycache__/
 *.pyc
-*.pyo
 .venv/
 .env
 .cache/

app.py

@@ -139,3 +139,25 @@ with gr.Blocks() as demo:
 
 # En Spaces no se usa share=True
 demo.launch()
+
+
+# ---- Remote dataset fallback (auto-injected) ----
+import os
+def _try_remote_sample(repo_id: str, split: str = "train", fields=("prompt","completion"), limit=2000):
+    try:
+        from datasets import load_dataset
+        ds = load_dataset(repo_id, split=split)
+        texts=[]
+        for i,row in enumerate(ds):
+            if all(k in row for k in fields):
+                texts.append(f"prompt: {row['prompt']}\ncompletion: {row['completion']}")
+            else:
+                for k in ("text","content","body"):
+                    if k in row and isinstance(row[k], str):
+                        texts.append(row[k]); break
+            if len(texts) >= limit: break
+        return texts
+    except Exception:
+        return []
+USE_REMOTE = os.getenv("SAVANT_REMOTE_DATASET","").strip()
+