pipelines go to gravenhollow now.

cpu_training_pipeline.py

@@ -7,7 +7,7 @@ distributed across multiple cluster nodes via KubeRay RayJob.
 GPUs remain 100 % dedicated to inference serving.
 
 Architecture:
-  1. Fetch PDFs from Quobjects S3
+  1. Fetch PDFs from RustFS S3
   2. Prepare instruction-tuning dataset
   3. Upload prepared data to S3 (shared storage for Ray workers)
   4. Submit a KubeRay RayJob that runs Ray Train TorchTrainer
@@ -41,7 +41,7 @@ from typing import NamedTuple
 
 
 # ──────────────────────────────────────────────────────────────
-# 1. Fetch PDFs from Quobjects S3
+# 1. Fetch PDFs from RustFS S3
 # ──────────────────────────────────────────────────────────────
 @dsl.component(
     base_image="python:3.13-slim",
@@ -54,7 +54,7 @@ def fetch_pdfs_from_s3(
     aws_access_key_id: str,
     aws_secret_access_key: str,
 ) -> NamedTuple("PDFOutput", [("pdf_dir", str), ("num_files", int)]):
-    """Download all PDFs from a Quobjects S3 bucket."""
+    """Download all PDFs from an S3 bucket."""
     import os
 
     import boto3
@@ -994,7 +994,7 @@ def log_training_metrics(
                 "num_epochs": num_epochs,
                 "num_pdfs": num_pdfs,
                 "backend": "ray-train-gloo",
-                "data_source": "quobjects/training-data",
+                "data_source": "rustfs/training-data",
             }
         )
         mlflow.log_metrics(
@@ -1023,7 +1023,7 @@ def log_training_metrics(
     ),
 )
 def cpu_training_pipeline(
-    # ── S3 / Quobjects ──
+    # ── S3 / RustFS ──
     s3_endpoint: str = "https://gravenhollow.lab.daviestechlabs.io:30292",
     s3_bucket: str = "training-data",
     s3_prefix: str = "",