Pin Flux2 GGUF config to klein base repo and pass HF token

app.py

@@ -41,16 +41,20 @@ def get_pipe() -> Flux2KleinPipeline:
         with _lock:
             if _pipe is None:
                 try:
+                    hf_token = os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACEHUB_API_TOKEN")
                     gguf_path = hf_hub_download(repo_id=GGUF_REPO, filename=GGUF_FILE)
                     qconfig = GGUFQuantizationConfig(compute_dtype=torch.float32)
                     transformer = Flux2Transformer2DModel.from_single_file(
                         gguf_path,
+                        config=BASE_REPO,
+                        token=hf_token,
                         quantization_config=qconfig,
                         torch_dtype=torch.float32,
                     )
                     pipe = Flux2KleinPipeline.from_pretrained(
                         BASE_REPO,
                         transformer=transformer,
+                        token=hf_token,
                         torch_dtype=torch.float32,
                     )
                     pipe = pipe.to("cpu")