[Bug] LMDeploy v0.6.4-cu12使用2张4090无法启动和推理, v0.6.0-cu12 可以正常启动和推理

### Checklist

- [x] 1. I have searched related issues but cannot get the expected help.
- [x] 2. The bug has not been fixed in the latest version.
- [x] 3. Please note that if the bug-related issue you submitted lacks corresponding environment info and a minimal reproducible demo, it will be challenging for us to reproduce and resolve the issue, reducing the likelihood of receiving feedback.

### Describe the bug

使用 `v0.6.0` 和 `v0.6.4` 版本在 **RTX4090** 机器上推理 `internlm2_5-20b-chat`

- `openmmlab/lmdeploy:v0.6.0-cu12`
- 2卡4090启动和推理成功
- 4卡4090启动和推理成功
- `openmmlab/lmdeploy:v0.6.4-cu12`
- 2卡4090启动失败，runtime_error 空指针map报错
```bash
terminate called after throwing an instance of 'std::runtime_error'
what(): [TM][ERROR] pointer_mapping_ does not have information of ptr at 0x150ffa3000. Assertion fail: /opt/lmdeploy/src/turbomind/utils/allocator.h:284
```
- 4卡4090启动和推理成功

### Reproduction

## 2卡4090启动命令

- `--max-batch-size 64`
- `--tp 2`

```bash
lmdeploy serve api_server internlm2_5-20b-chat --cache-max-entry-count 0.8 --max-batch-size 64 --tp 2 --log-level INFO --server-port 80
```

## 4卡4090启动命令

- `--max-batch-size 128`
- `--tp 4`

```bash
lmdeploy serve api_server internlm2_5-20b-chat --cache-max-entry-count 0.8 --max-batch-size 128 --tp 4 --log-level INFO --server-port 80
```

### Environment

```Shell
sys.platform: linux
Python: 3.10.12 (main, Nov 6 2024, 20:22:13) [GCC 11.4.0]
CUDA available: True
MUSA available: False
numpy_random_seed: 2147483648
GPU 0,1,2,3: NVIDIA GeForce RTX 4090 D
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 12.4, V12.4.131
GCC: x86_64-linux-gnu-gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
PyTorch: 2.3.0+cu121
PyTorch compiling details: PyTorch built with:
- GCC 9.3
- C++ Version: 201703
- Intel(R) oneAPI Math Kernel Library Version 2022.2-Product Build 20220804 for Intel(R) 64 architecture applications
- Intel(R) MKL-DNN v3.3.6 (Git Hash 86e6af5974177e513fd3fee58425e1063e7f1361)
- OpenMP 201511 (a.k.a. OpenMP 4.5)
- LAPACK is enabled (usually provided by MKL)
- NNPACK is enabled
- CPU capability usage: AVX512
- CUDA Runtime 12.1
- NVCC architecture flags: -gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90
- CuDNN 8.9.2
- Magma 2.6.1
- Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=12.1, CUDNN_VERSION=8.9.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -D_GLIBCXX_USE_CXX11_ABI=0 -fabi-version=11 -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=pedantic -Wno-error=old-style-cast -Wno-missing-braces -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=2.3.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF,

TorchVision: 0.18.0+cu121
LMDeploy: 0.6.4+
transformers: 4.47.0
gradio: 5.8.0
fastapi: 0.115.6
pydantic: 2.10.3
triton: 2.3.0
NVIDIA Topology:
GPU0 GPU1 GPU2 GPU3 NIC0 NIC1 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X NODE NODE NODE NODE NODE 32-63,96-127 1 N/A
GPU1 NODE X NODE NODE NODE NODE 32-63,96-127 1 N/A
GPU2 NODE NODE X NODE NODE NODE 32-63,96-127 1 N/A
GPU3 NODE NODE NODE X NODE NODE 32-63,96-127 1 N/A
NIC0 NODE NODE NODE NODE X PIX
NIC1 NODE NODE NODE NODE PIX X

Legend:

X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks

NIC Legend:

NIC0: rocep171s0f0
NIC1: rocep171s0f1
```

### Error traceback

```Shell
2025-01-21 08:07:50,565 - lmdeploy - INFO - async_engine.py:143 - input backend=turbomind, backend_config=TurbomindEngineConfig(dtype='auto', model_format=None, tp=2, session_len=None, max_batch_size=64, cache_max_entry_count=0.8, cache_chunk_size=-1, cache_block_seq_len=64, enable_prefix_caching=False, quant_policy=0, rope_scaling_factor=0.0, use_logn_attn=False, download_dir=None, revision=None, max_prefill_token_num=8192, num_tokens_per_iter=0, max_prefill_iters=1)
2025-01-21 08:07:50,566 - lmdeploy - INFO - async_engine.py:145 - input chat_template_config=None
2025-01-21 08:07:50,635 - lmdeploy - INFO - async_engine.py:155 - updated chat_template_onfig=ChatTemplateConfig(model_name='internlm2', system=None, meta_instruction=None, eosys=None, user=None, eoh=None, assistant=None, eoa=None, separator=None, capability=None, stop_words=None)
2025-01-21 08:07:50,635 - lmdeploy - INFO - turbomind.py:301 - model_source: hf_model
2025-01-21 08:07:51,144 - lmdeploy - INFO - turbomind.py:200 - turbomind model config:

{
"model_config": {
"model_name": "",
"chat_template": "",
"model_arch": "InternLM2ForCausalLM",
"head_num": 48,
"kv_head_num": 8,
"hidden_units": 6144,
"vocab_size": 92544,
"embedding_size": 92544,
"num_layer": 48,
"inter_size": [
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384,
16384
],
"norm_eps": 1e-05,
"attn_bias": 0,
"start_id": 1,
"end_id": 2,
"size_per_head": 128,
"group_size": 64,
"weight_type": "bfloat16",
"session_len": 32768,
"tp": 2,
"model_format": "hf",
"expert_num": [],
"expert_inter_size": 0,
"experts_per_token": 0,
"moe_shared_gate": false,
"norm_topk_prob": false,
"routed_scale": 1.0,
"topk_group": 1,
"topk_method": "greedy",
"moe_group_num": 1,
"q_lora_rank": 0,
"kv_lora_rank": 0,
"qk_rope_dim": 0,
"v_head_dim": 0,
"tune_layer_num": 1
},
"attention_config": {
"rotary_embedding": 128,
"rope_theta": 50000000.0,
"softmax_scale": 0.0,
"attention_factor": -1.0,
"max_position_embeddings": 32768,
"original_max_position_embeddings": 0,
"rope_scaling_type": "dynamic",
"rope_scaling_factor": 2.5,
"use_dynamic_ntk": 1,
"low_freq_factor": 1.0,
"high_freq_factor": 1.0,
"beta_fast": 32.0,
"beta_slow": 1.0,
"use_logn_attn": 0,
"cache_block_seq_len": 64
},
"lora_config": {
"lora_policy": "",
"lora_r": 0,
"lora_scale": 0.0,
"lora_max_wo_r": 0,
"lora_rank_pattern": "",
"lora_scale_pattern": ""
},
"engine_config": {
"dtype": "auto",
"model_format": null,
"tp": 2,
"session_len": null,
"max_batch_size": 64,
"cache_max_entry_count": 0.8,
"cache_chunk_size": -1,
"cache_block_seq_len": 64,
"enable_prefix_caching": false,
"quant_policy": 0,
"rope_scaling_factor": 0.0,
"use_logn_attn": false,
"download_dir": null,
"revision": null,
"max_prefill_token_num": 8192,
"num_tokens_per_iter": 8192,
"max_prefill_iters": 4
}
}
[TM][WARNING] [LlamaTritonModel] `max_context_token_num` is not set, default to 32768.
[TM][INFO] Model:
head_num: 48
kv_head_num: 8
size_per_head: 128
num_layer: 48
vocab_size: 92544
attn_bias: 0
max_batch_size: 64
max_prefill_token_num: 8192
max_context_token_num: 32768
num_tokens_per_iter: 8192
max_prefill_iters: 4
session_len: 32768
cache_max_entry_count: 0.8
cache_block_seq_len: 64
cache_chunk_size: -1
enable_prefix_caching: 0
start_id: 1
tensor_para_size: 2
pipeline_para_size: 1
enable_custom_all_reduce: 0
model_name:
model_dir:
quant_policy: 0
group_size: 64
expert_per_token: 0
moe_method: 1

[TM][INFO] TM_FUSE_SILU_ACT=1
2025-01-21 08:07:51,613 - lmdeploy - WARNING - turbomind.py:231 - get 581 model params
[TM][INFO] [LlamaWeight::prepare] workspace size: 201326592

[TM][INFO] [LlamaWeight::prepare] workspace size: 201326592

[TM][WARNING] Devicle 0 peer access Device 1 is not available.
[TM][WARNING] Devicle 1 peer access Device 0 is not available.
[WARNING] gemm_config.in is not found; using default GEMM algo
[WARNING] gemm_config.in is not found; using default GEMM algo
[TM][INFO] [BlockManager] block_size = 6 MB
[TM][INFO] [BlockManager] max_block_count = 339
[TM][INFO] [BlockManager] block_size = 6 MB
[TM][INFO] [BlockManager] max_block_count = 339
[TM][INFO] [BlockManager] chunk_size = 339
[TM][INFO] [BlockManager] chunk_size = 339
[TM][WARNING] No enough blocks for `session_len` (32768), `session_len` truncated to 21696.
[TM][INFO] LlamaBatch::Start()
[TM][INFO] LlamaBatch::Start()
[TM][INFO] [Gemm2] Tuning sequence: 8, 16, 32, 48, 64, 96, 128, 192, 256, 384, 512, 768, 1024, 1536, 2048, 3072, 4096, 6144, 8192
[TM][INFO] [Gemm2] 8
[TM][INFO] [Gemm2] 16
[TM][INFO] [Gemm2] 32
[TM][INFO] [Gemm2] 48
[TM][INFO] [Gemm2] 64
[TM][INFO] [Gemm2] 96
[TM][INFO] [Gemm2] 128
[TM][INFO] [Gemm2] 192
[TM][INFO] [Gemm2] 256
[TM][INFO] [Gemm2] 384
[TM][INFO] [Gemm2] 512
[TM][INFO] [Gemm2] 768
[TM][INFO] [Gemm2] 1024
[TM][INFO] [Gemm2] 1536
[TM][INFO] [Gemm2] 2048
[TM][INFO] [Gemm2] 3072
[TM][INFO] [InternalThreadEntry] stop requested.
[TM][INFO] [InternalThreadEntry] stop requested.
terminate called after throwing an instance of 'std::runtime_error'
what(): [TM][ERROR] pointer_mapping_ does not have information of ptr at 0x150ffa3000. Assertion fail: /opt/lmdeploy/src/turbomind/utils/allocator.h:284
```