diff --git a/src/autora/doc/pipelines/train.py b/src/autora/doc/pipelines/train.py
index 870a016..ad81518 100644
--- a/src/autora/doc/pipelines/train.py
+++ b/src/autora/doc/pipelines/train.py
@@ -33,7 +33,7 @@ def fine_tune(base_model: str, new_model_name: str, dataset: Dataset) -> None:
 
     model = AutoModelForCausalLM.from_pretrained(
         base_model,
-        *kwargs,
+        **kwargs,
     )
     model.config.use_cache = False
     model.config.pretraining_tp = 1
@@ -61,8 +61,8 @@ def fine_tune(base_model: str, new_model_name: str, dataset: Dataset) -> None:
         logging_steps=1,  # TODO: Increase once there's more data
         learning_rate=2e-4,
         weight_decay=0.001,
-        fp16=False,
-        bf16=cuda_available,
+        fp16=cuda_available,
+        bf16=False,
         max_grad_norm=0.3,
         max_steps=-1,
         warmup_ratio=0.03,
diff --git a/src/autora/doc/runtime/predict_hf.py b/src/autora/doc/runtime/predict_hf.py
index 97a7f4f..d8bf424 100644
--- a/src/autora/doc/runtime/predict_hf.py
+++ b/src/autora/doc/runtime/predict_hf.py
@@ -107,5 +107,5 @@ def get_quantization_config() -> Any:
         load_in_4bit=True,
         bnb_4bit_use_double_quant=True,
         bnb_4bit_quant_type="nf4",
-        bnb_4bit_compute_dtype=torch.bfloat16,
+        bnb_4bit_compute_dtype=torch.float16,
     )