Fix --no-template: gate TEMPLATE emission, not just preflight

src/dlm/export/ollama/modelfile.py

@@ -98,7 +98,7 @@ def render_modelfile(ctx: ModelfileContext) -> str:
     )
     from_line = f"FROM ./{ctx.base_gguf_name}"
     adapter_line = f"ADAPTER ./{ctx.adapter_gguf_name}" if ctx.adapter_gguf_name else None
-    template_block = _build_template_block(template_row)
+    template_block = _build_template_block(template_row) if ctx.plan.include_template else None
     num_ctx = resolve_num_ctx(ctx.training_sequence_len, ctx.spec.context_length)
     temperature = (
         ctx.override_temperature
@@ -121,7 +121,10 @@ def render_modelfile(ctx: ModelfileContext) -> str:
     parts: list[str] = [header, "", from_line]
     if adapter_line is not None:
         parts.append(adapter_line)
-    parts.extend(["", template_block, ""])
+    if template_block is not None:
+        parts.extend(["", template_block, ""])
+    else:
+        parts.append("")
     parts.extend(param_lines)
     if system_line is not None:
         parts.extend(["", system_line])

src/dlm/export/ollama/vl_modelfile.py

@@ -131,7 +131,7 @@ def render_vl_modelfile(ctx: VlModelfileContext) -> str:
     )
     from_line = f"FROM ./{ctx.base_gguf_name}"
     adapter_line = f"ADAPTER ./{ctx.adapter_gguf_name}" if ctx.adapter_gguf_name else None
-    template_block = f'TEMPLATE """{_VL_TEMPLATE_BODY}"""'
+    template_block = f'TEMPLATE """{_VL_TEMPLATE_BODY}"""' if ctx.plan.include_template else None
     num_ctx = resolve_num_ctx(ctx.training_sequence_len, ctx.spec.context_length)
     temperature = (
         ctx.override_temperature
@@ -152,7 +152,10 @@ def render_vl_modelfile(ctx: VlModelfileContext) -> str:
     parts: list[str] = [header, "", from_line]
     if adapter_line is not None:
         parts.append(adapter_line)
-    parts.extend(["", template_block, ""])
+    if template_block is not None:
+        parts.extend(["", template_block, ""])
+    else:
+        parts.append("")
     parts.extend(param_lines)
     if system_line is not None:
         parts.extend(["", system_line])

tests/unit/export/ollama/test_modelfile.py

@@ -82,6 +82,33 @@ def test_template_block_present(self, tmp_path: Path) -> None:
         assert 'TEMPLATE """' in text
         assert "<|im_start|>" in text  # chatml dialect
 
+    def test_no_template_when_include_template_false(self, tmp_path: Path) -> None:
+        """Audit 13 M13.1: ``--no-template`` must actually suppress the
+        TEMPLATE block in the emitted Modelfile, not just the preflight."""
+        ctx = _ctx(tmp_path)
+        # Bypass the frozen-dataclass replace dance: rebuild with a plan
+        # that has include_template=False.
+        plan = ExportPlan(quant="Q4_K_M", merged=False, include_template=False)
+        ctx_no_tmpl = ModelfileContext(
+            spec=ctx.spec,
+            plan=plan,
+            adapter_dir=ctx.adapter_dir,
+            base_gguf_name=ctx.base_gguf_name,
+            adapter_gguf_name=ctx.adapter_gguf_name,
+            dlm_id=ctx.dlm_id,
+            adapter_version=ctx.adapter_version,
+            system_prompt=ctx.system_prompt,
+            training_sequence_len=ctx.training_sequence_len,
+            override_temperature=ctx.override_temperature,
+            override_top_p=ctx.override_top_p,
+            draft_model_ollama_name=ctx.draft_model_ollama_name,
+        )
+        text = render_modelfile(ctx_no_tmpl)
+        assert "TEMPLATE" not in text
+        # Other directives still emit normally.
+        assert "FROM ./base.Q4_K_M.gguf" in text
+        assert "PARAMETER temperature" in text
+
     def test_params_emitted(self, tmp_path: Path) -> None:
         text = render_modelfile(_ctx(tmp_path))
         assert "PARAMETER temperature" in text

tests/unit/export/ollama/test_vl_modelfile.py

@@ -150,6 +150,21 @@ def test_blank_system_prompt_is_omitted(self, tmp_path: Path) -> None:
 
         assert "SYSTEM " not in text
 
+    def test_no_template_when_include_template_false(self, tmp_path: Path) -> None:
+        """`--no-template` skips TEMPLATE emission on the VL renderer."""
+        ctx = VlModelfileContext(
+            spec=_QWEN_VL_SPEC,
+            plan=ExportPlan(quant="Q4_K_M", merged=False, include_template=False),
+            adapter_dir=_adapter_dir(tmp_path),
+            base_gguf_name="base.Q4_K_M.gguf",
+            adapter_gguf_name="adapter.gguf",
+            dlm_id="01VLTEST",
+            adapter_version=4,
+        )
+        text = render_vl_modelfile(ctx)
+        assert "TEMPLATE" not in text
+        assert "FROM ./base.Q4_K_M.gguf" in text
+
 
 class TestVlStopsFromAdapter:
     def test_family_defaults_and_adapter_special_tokens_are_merged(self, tmp_path: Path) -> None:

tests/unit/export/test_vl_modelfile.py

@@ -48,10 +48,10 @@ def _fake_vl_spec() -> Any:
     )
 
 
-def _fake_plan() -> Any:
+def _fake_plan(*, include_template: bool = True) -> Any:
     from types import SimpleNamespace
 
-    return SimpleNamespace(quant="Q4_K_M", merged=False)
+    return SimpleNamespace(quant="Q4_K_M", merged=False, include_template=include_template)
 
 
 @pytest.fixture
@@ -166,6 +166,21 @@ def test_ends_with_newline(self, vl_ctx: VlModelfileContext) -> None:
         out = render_vl_modelfile(vl_ctx)
         assert out.endswith("\n")
 
+    def test_no_template_when_include_template_false(self, adapter_dir: Path) -> None:
+        """`--no-template` skips TEMPLATE emission on the VL path too."""
+        ctx = VlModelfileContext(
+            spec=_fake_vl_spec(),
+            plan=_fake_plan(include_template=False),
+            adapter_dir=adapter_dir,
+            base_gguf_name="base.Q4_K_M.gguf",
+            adapter_gguf_name="adapter.Q4_K_M.gguf",
+            dlm_id="01JZZZZZZZZZZZZZZZZZZZZZZZ",
+            adapter_version=1,
+        )
+        out = render_vl_modelfile(ctx)
+        assert "TEMPLATE" not in out
+        assert "FROM ./base.Q4_K_M.gguf" in out
+
 
 class TestStopsFallback:
     def test_paligemma_uses_gemma_style_stop(self, tmp_path: Path) -> None: