From ab58121cd89dc5e21b5e1aa7dfc8f710cd75e531 Mon Sep 17 00:00:00 2001
From: Roland Tannous <rolandtannous@gravityq.ai>
Date: Tue, 10 Mar 2026 15:43:35 +0000
Subject: [PATCH] fix: harden template mapping for complex column types and
 curly braces

- Handle dict columns (e.g. squad answers) by extracting text instead
  of raw repr()
- Handle list columns by joining or extracting single value
- Catch ValueError in .format() calls (stray { } in column data)
- Add missing json import to dataset_utils.py
---
 .../backend/utils/datasets/dataset_utils.py   | 22 ++++++++++++++++---
 studio/backend/utils/datasets/llm_assist.py   |  4 ++--
 2 files changed, 21 insertions(+), 5 deletions(-)

diff --git a/studio/backend/utils/datasets/dataset_utils.py b/studio/backend/utils/datasets/dataset_utils.py
index 1d1e2e78c0..95095d2ab0 100644
--- a/studio/backend/utils/datasets/dataset_utils.py
+++ b/studio/backend/utils/datasets/dataset_utils.py
@@ -18,6 +18,8 @@ All internal utilities have been moved to separate modules:
 - model_mappings: TEMPLATE_TO_MODEL_MAPPER
 """
 
+import json
+
 # Import from modular files
 from .format_detection import (
     detect_dataset_format,
@@ -269,7 +271,20 @@ def _apply_template_mapping(
             row_values = {}
             for col in all_columns:
                 val = examples[col][i]
-                str_val = str(val) if val is not None else ""
+
+                # Handle complex types (dicts, lists) — extract
+                # useful text instead of raw repr
+                if isinstance(val, dict):
+                    # Common pattern: {"text": [...]} in QA datasets
+                    if "text" in val:
+                        inner = val["text"]
+                        str_val = inner[0] if isinstance(inner, list) and inner else str(inner)
+                    else:
+                        str_val = json.dumps(val, ensure_ascii=False)
+                elif isinstance(val, list):
+                    str_val = val[0] if len(val) == 1 else ", ".join(str(v) for v in val)
+                else:
+                    str_val = str(val) if val is not None else ""
 
                 # Apply label mapping if this column has one
                 if col in label_mapping and isinstance(label_mapping[col], dict):
@@ -288,7 +303,8 @@ def _apply_template_mapping(
             if user_template:
                 try:
                     user_content = user_template.format(**row_values)
-                except (KeyError, IndexError):
+                except (KeyError, IndexError, ValueError):
+                    # ValueError: stray { } in column values
                     user_content = user_template
                 convo.append({"role": "user", "content": user_content})
 
@@ -296,7 +312,7 @@ def _apply_template_mapping(
             if assistant_template:
                 try:
                     asst_content = assistant_template.format(**row_values)
-                except (KeyError, IndexError):
+                except (KeyError, IndexError, ValueError):
                     asst_content = assistant_template
                 convo.append({"role": "assistant", "content": asst_content})
 
diff --git a/studio/backend/utils/datasets/llm_assist.py b/studio/backend/utils/datasets/llm_assist.py
index 65df51ff15..09791623c4 100644
--- a/studio/backend/utils/datasets/llm_assist.py
+++ b/studio/backend/utils/datasets/llm_assist.py
@@ -613,11 +613,11 @@ def _run_multi_pass_advisor(
 
             try:
                 user_msg = user_tpl.format(**row_vals)
-            except (KeyError, IndexError):
+            except (KeyError, IndexError, ValueError):
                 user_msg = user_tpl
             try:
                 asst_msg = asst_tpl.format(**row_vals)
-            except (KeyError, IndexError):
+            except (KeyError, IndexError, ValueError):
                 asst_msg = asst_tpl
 
             examples_text += f"Example {i}:\n  System: {sys_prompt}\n  User: {user_msg}\n  Assistant: {asst_msg}\n\n"