Revert "[Spec] Split init_backends; account draft weights in --mem-fraction-static" (#28841)

This commit is contained in:
Lianmin Zheng
2026-06-21 07:52:35 -07:00
committed by GitHub
parent 320b231ea6
commit 7942d546d1
12 changed files with 99 additions and 148 deletions
+1 -2
View File
@@ -324,8 +324,7 @@ def load_model(server_args, port_args, gpu_id, tp_rank):
else:
model_runner = ModelRunner(**runner_kwargs)
model_runner.alloc_memory_pool()
model_runner.init_attention_backends()
model_runner.init_cuda_graphs()
model_runner.init_backends()
rank_print(f"max_total_num_tokens={model_runner.max_total_num_tokens}")
tokenizer = get_tokenizer(
server_args.tokenizer_path,