pass model explicitly to all pipeline stages

main.py

@@ -59,14 +59,16 @@ async def generate(
         else:
             source_args = ["--url", url.strip()]
 
-        provider  = os.getenv("PROVIDER", "gemini")
-        model     = os.getenv("MODEL", "")
-        extra_env = {"MODEL": model} if model else {}
+        provider = os.getenv("PROVIDER", "gemma")
+        # Default model for NVIDIA/Llama — overridable via MODEL env var
+        default_model = "meta/llama-3.3-70b-instruct"
+        model = os.getenv("MODEL", default_model)
+        extra_env = {"MODEL": model}
 
         blog_json = data_dir / "blog_data.json"
 
         def run_pipeline():
-            # Stage 0 – parse blog
+            # Stage 0 – parse blog (no LLM, no --model needed)
             _run("0_blog_process.py",
                  source_args + ["--output_json_path", str(blog_json)],
                  extra_env)
@@ -85,9 +87,10 @@ async def generate(
                 "--output_dir",     str(data_dir),
                 "--provider",       provider,
                 "--content_type",   "blog",
+                "--model",          model,
             ], extra_env)
 
-            # Stage 1.1 – extract config
+            # Stage 1.1 – extract config (no LLM, no --model needed)
             _run("1_1_extract_config.py", [
                 "--output_dir", str(data_dir),
             ], extra_env)
@@ -101,6 +104,7 @@ async def generate(
                 "--pdf_json_path", str(blog_json_path),
                 "--output_dir",    str(data_dir),
                 "--provider",      provider,
+                "--model",         model,
             ], extra_env)
 
             # Stage 3 – code generation
@@ -109,6 +113,7 @@ async def generate(
                 "--output_dir",      str(data_dir),
                 "--output_repo_dir", str(output_dir),
                 "--provider",        provider,
+                "--model",           model,
             ], extra_env)
 
         await asyncio.get_event_loop().run_in_executor(None, run_pipeline)