Drop torchair (#4814)

aclgraph is stable and fast now. Let's drop torchair graph mode now.

TODO: some logic to adapt torchair should be cleaned up as well. We'll
do it in the following PR.

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>

tests/e2e/nightly/models/test_deepseek_r1_0528_w8a8.py

@@ -29,7 +29,6 @@ MODELS = [
 ]
 
 MODES = [
-    "torchair",
     "single",
     "aclgraph",
     "aclgraph_mlapo",
@@ -78,13 +77,6 @@ async def test_models(model: str, mode: str) -> None:
     }
     speculative_config = {"num_speculative_tokens": 1, "method": "mtp"}
     additional_config = {
-        "torchair_graph_config": {
-            "enabled": True,
-            "enable_multistream_moe": False,
-            "enable_multistream_mla": True,
-            "graph_batch_sizes": [16],
-            "use_cached_graph": True
-        },
         "chunked_prefill_for_mla": True,
         "enable_weight_nz_layout": True
     }
@@ -99,12 +91,8 @@ async def test_models(model: str, mode: str) -> None:
     ]
     if mode == "single":
         server_args.append("--enforce-eager")
-        additional_config["torchair_graph_config"] = {"enabled": False}
-    if mode == "aclgraph":
-        additional_config["torchair_graph_config"] = {"enabled": False}
     if mode == "aclgraph_mlapo":
         env_dict["VLLM_ASCEND_ENABLE_MLAPO"] = "1"
-        additional_config["torchair_graph_config"] = {"enabled": False}
     server_args.extend(["--additional-config", json.dumps(additional_config)])
     request_keyword_args: dict[str, Any] = {
         **api_keyword_args,

tests/e2e/nightly/models/test_deepseek_r1_w8a8_eplb.py

@@ -68,9 +68,6 @@ async def test_models(model: str) -> None:
         "cudagraph_mode": "FULL_DECODE_ONLY"
     }
     additional_config: dict[str, Any] = {
-        "torchair_graph_config": {
-            "enabled": True
-        },
         "enable_shared_expert_dp": False,
         "multistream_overlap_shared_expert": False,
         "dynamic_eplb": True,

tests/e2e/nightly/models/test_deepseek_v3_2_exp_w8a8.py

@@ -72,27 +72,13 @@ async def test_models(model: str, tp_size: int, dp_size: int,
     port = get_open_port()
     env_dict = {"HCCL_BUFFSIZE": "1024", "VLLM_ASCEND_ENABLE_MLAPO": "0"}
     server_args = [
-        "--no-enable-prefix-caching",
-        "--enable-expert-parallel",
+        "--no-enable-prefix-caching", "--enable-expert-parallel",
         "--tensor-parallel-size",
-        str(tp_size),
-        "--data-parallel-size",
-        str(dp_size),
-        "--port",
-        str(port),
-        "--max-model-len",
-        "16384",
-        "--max-num-batched-tokens",
-        "16384",
-        "--block-size",
-        "16",
-        "--trust-remote-code",
-        "--quantization",
-        "ascend",
-        "--gpu-memory-utilization",
-        "0.9",
-        "--additional-config",
-        '{"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]}}',
+        str(tp_size), "--data-parallel-size",
+        str(dp_size), "--port",
+        str(port), "--max-model-len", "16384", "--max-num-batched-tokens",
+        "16384", "--block-size", "16", "--trust-remote-code", "--quantization",
+        "ascend", "--gpu-memory-utilization", "0.9"
     ]
     if full_graph:
         server_args += [