HF's workaround breaks stuff

2025-06-05 21:59:24 +02:00 · 2023-07-22 16:29:55 +02:00
parent fa9d17b3d3
commit 8dd7b93a6c
1 changed files with 0 additions and 1 deletions
--- a/modeling/inference_models/generic_hf_torch/class.py
+++ b/modeling/inference_models/generic_hf_torch/class.py
@@ -81,7 +81,6 @@ class model_backend(HFTorchInferenceModel):

        tf_kwargs = {
            "low_cpu_mem_usage": True,
-            "pretraining_tp": 1,
        }
        
        if self.quantization == "8bit":