diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index 44c700bf3d..d752500bf4 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -52,8 +52,14 @@ _REPROMPT_MAX_CHARS = 2000
 _SHARD_FULL_RE = re.compile(r"^(.*)-(\d{5})-of-(\d{5})\.gguf$")
 _SHARD_RE = re.compile(r"^(.*)-\d{5}-of-\d{5}\.gguf$")
 
-# Model size extraction (shared with routes/inference.py)
-from utils.models import extract_model_size_b as _extract_model_size_b
+
+# Model size extraction — lazy import to avoid pulling in transformers
+# at module level.  See PR description for the full explanation.
+def _extract_model_size_b(model_id: str):
+    from utils.models import extract_model_size_b
+
+    return extract_model_size_b(model_id)
+
 
 # ── Pre-compiled patterns for tool XML stripping ─────────────
 _TOOL_CLOSED_PATS = [
diff --git a/studio/backend/utils/models/model_config.py b/studio/backend/utils/models/model_config.py
index df1058abf6..6cffc534aa 100644
--- a/studio/backend/utils/models/model_config.py
+++ b/studio/backend/utils/models/model_config.py
@@ -5,7 +5,6 @@
 Model and LoRA configuration handling
 """
 
-from transformers import AutoConfig
 from dataclasses import dataclass
 from typing import Optional, Dict, Any
 from utils.paths import (
@@ -422,6 +421,7 @@ def load_model_config(
     """
     Load model config with optional authentication control.
     """
+    from transformers import AutoConfig
 
     if token:
         # Explicit token provided - use it