Fix compile bug for stella (#757)

predibase · Feb 7, 2025 · d6072b9 · d6072b9
1 parent acaa217
commit d6072b9
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/server/lorax_server/models/flash_causal_lm.py b/server/lorax_server/models/flash_causal_lm.py
@@ -1352,7 +1352,7 @@ def warmup(self, batch: FlashCausalLMBatch, max_new_tokens: int, embedding_model
             torch.cuda.synchronize(self.device)
 
         logger.info(f'Post warmup cuda memory: {get_cuda_free_memory(self.device, 1) / (1024 ** 3):.2f} GB')
-        del self.model_graph_wrapper
+        self.model_graph_wrapper = None
         self.kv_cache = []
         torch.cuda.synchronize(self.device)
         torch.cuda.empty_cache()