Upload h3_quantization.py

h3_quantization.py

@@ -12,7 +12,7 @@ def load_int8_transformer(repo):
         from torchao.quantization import Int8WeightOnlyConfig
     except ImportError as error:
         raise RuntimeError(
-            "INT8 needs the optional requirements-int8.txt dependencies. "
+            "INT8 needs torchao==0.17.0 in the main requirements.txt. "
             "Use H3_QUANTIZATION=bf16 for the existing supported path."
         ) from error
     model = MiniMaxH3Transformer3DModel.from_pretrained(
@@ -24,7 +24,9 @@ def load_int8_transformer(repo):
                 "token_refiner", "norm_out", "proj_out", "audio_proj_out",
             ],
         ),
-        low_cpu_mem_usage=False,
+        # The pinned Diffusers ModelMixin rejects False with a quantizer.
+        # Accelerate is already pinned in requirements.txt for this load path.
+        low_cpu_mem_usage=True,
     )
     model.requires_grad_(False)
     return model