[Bug][Regression]: Dimension out of range when using MooncakeStoreConnector

[gronsti-amd]

Your current environment

The output of python collect_env.py

==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version                : 18.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.3.1 24491 1e0fda770a2079fbd71e4b70974d74f62fd3af10)
CMake version                : version 3.31.6
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.7.0+gitf717b2a
Is debug build               : False
CUDA used to build PyTorch   : N/A
ROCM used to build PyTorch   : 6.3.42133-1b9c17779

==============================
      Python Environment
==============================
Python version               : 3.12.10 (main, Apr  9 2025, 08:55:05) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-5.15.0-140-generic-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : Could not collect
CUDA_MODULE_LOADING set to   : LAZY
GPU models and configuration : AMD Instinct MI300X (gfx942:sramecc+:xnack-)
Nvidia driver version        : Could not collect
cuDNN version                : Could not collect
HIP runtime version          : 6.3.42133
MIOpen runtime version       : 3.3.0
Is XNNPACK available         : True

==============================
          CPU Info
==============================
Architecture:                         x86_64
CPU op-mode(s):                       32-bit, 64-bit
Address sizes:                        52 bits physical, 57 bits virtual
Byte Order:                           Little Endian
CPU(s):                               128
On-line CPU(s) list:                  0-127
Vendor ID:                            AuthenticAMD
Model name:                           AMD EPYC 9354 32-Core Processor
CPU family:                           25
Model:                                17
Thread(s) per core:                   2
Core(s) per socket:                   32
Socket(s):                            2
Stepping:                             1
Frequency boost:                      enabled
CPU max MHz:                          3250.0000
CPU min MHz:                          1500.0000
BogoMIPS:                             6490.32
Flags:                                fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 invpcid_single hw_pstate ssbd mba ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid overflow_recov succor smca fsrm flush_l1d
Virtualization:                       AMD-V
L1d cache:                            2 MiB (64 instances)
L1i cache:                            2 MiB (64 instances)
L2 cache:                             64 MiB (64 instances)
L3 cache:                             512 MiB (16 instances)
NUMA node(s):                         2
NUMA node0 CPU(s):                    0-31,64-95
NUMA node1 CPU(s):                    32-63,96-127
Vulnerability Gather data sampling:   Not affected
Vulnerability Itlb multihit:          Not affected
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Not affected
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed:               Not affected
Vulnerability Spec rstack overflow:   Mitigation; safe RET
Vulnerability Spec store bypass:      Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1:             Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:             Mitigation; Enhanced / Automatic IBRS; IBPB conditional; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

==============================
Versions of relevant libraries
==============================
[pip3] numpy==1.26.4
[pip3] pyzmq==26.4.0
[pip3] torch==2.7.0+gitf717b2a
[pip3] torchvision==0.21.0+7af6987
[pip3] transformers==4.52.3
[pip3] triton==3.2.0+gite5be006a
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : 6.3.42133-1b9c17779
Neuron SDK Version           : N/A
vLLM Version                 : 0.9.1.dev18+gce75efeec (git sha: ce75efeec)
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
  ============================ ROCm System Management Interface ============================
================================ Weight between two GPUs =================================
       GPU0         GPU1         GPU2         GPU3         GPU4         GPU5         GPU6         GPU7         
GPU0   0            15           15           15           15           15           15           15           
GPU1   15           0            15           15           15           15           15           15           
GPU2   15           15           0            15           15           15           15           15           
GPU3   15           15           15           0            15           15           15           15           
GPU4   15           15           15           15           0            15           15           15           
GPU5   15           15           15           15           15           0            15           15           
GPU6   15           15           15           15           15           15           0            15           
GPU7   15           15           15           15           15           15           15           0            

================================= Hops between two GPUs ==================================
       GPU0         GPU1         GPU2         GPU3         GPU4         GPU5         GPU6         GPU7         
GPU0   0            1            1            1            1            1            1            1            
GPU1   1            0            1            1            1            1            1            1            
GPU2   1            1            0            1            1            1            1            1            
GPU3   1            1            1            0            1            1            1            1            
GPU4   1            1            1            1            0            1            1            1            
GPU5   1            1            1            1            1            0            1            1            
GPU6   1            1            1            1            1            1            0            1            
GPU7   1            1            1            1            1            1            1            0            

=============================== Link Type between two GPUs ===============================
       GPU0         GPU1         GPU2         GPU3         GPU4         GPU5         GPU6         GPU7         
GPU0   0            XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         
GPU1   XGMI         0            XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         
GPU2   XGMI         XGMI         0            XGMI         XGMI         XGMI         XGMI         XGMI         
GPU3   XGMI         XGMI         XGMI         0            XGMI         XGMI         XGMI         XGMI         
GPU4   XGMI         XGMI         XGMI         XGMI         0            XGMI         XGMI         XGMI         
GPU5   XGMI         XGMI         XGMI         XGMI         XGMI         0            XGMI         XGMI         
GPU6   XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         0            XGMI         
GPU7   XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         XGMI         0            

======================================= Numa Nodes =======================================
GPU[0]          : (Topology) Numa Node: 0
GPU[0]          : (Topology) Numa Affinity: 0
GPU[1]          : (Topology) Numa Node: 0
GPU[1]          : (Topology) Numa Affinity: 0
GPU[2]          : (Topology) Numa Node: 0
GPU[2]          : (Topology) Numa Affinity: 0
GPU[3]          : (Topology) Numa Node: 0
GPU[3]          : (Topology) Numa Affinity: 0
GPU[4]          : (Topology) Numa Node: 1
GPU[4]          : (Topology) Numa Affinity: 1
GPU[5]          : (Topology) Numa Node: 1
GPU[5]          : (Topology) Numa Affinity: 1
GPU[6]          : (Topology) Numa Node: 1
GPU[6]          : (Topology) Numa Affinity: 1
GPU[7]          : (Topology) Numa Node: 1
GPU[7]          : (Topology) Numa Affinity: 1
================================== End of ROCm SMI Log ===================================

==============================
     Environment Variables
==============================
PYTORCH_TUNABLEOP_TUNING=0
PYTORCH_TUNABLEOP_ENABLED=1
VLLM_WORKER_MULTIPROC_METHOD=spawn
PYTORCH_ROCM_ARCH=gfx90a;gfx942;gfx1100;gfx1101;gfx1200;gfx1201
LD_LIBRARY_PATH=/opt/rocm/lib:/usr/local/lib:
PYTORCH_TUNABLEOP_FILENAME=/app/afo_tune_device_%d_full.csv
NCCL_CUMEM_ENABLE=0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

🐛 Describe the bug

When using disaggregated prefill with the mooncake store connector, the decoding instance crashes with an IndexError: Dimension out of range from the reshape_and_cache_flash kernel.

The bug is not triggered when using normal non-disaggregated serving with the same model. It seems that it is the vLLM Mooncake integration not conforming to the new expected shape of the KV cache.

The bug appears to be a regression, triggered by the following change: (#16605)

commit 9e96f56efb5b44a12d9d516276daa5538700a211 2025-04-26 Shu Wang <shuw@nvidia.com> Allocate kv_cache with stride order (#16605)                                                
9e96f56efb csrc/cache_kernels.cu (Shu Wang           2025-04-26 00:03:31 -0500 439)   int64_t page_stride = key_cache.stride(1);                                                  
9e96f56efb csrc/cache_kernels.cu (Shu Wang           2025-04-26 00:03:31 -0500 440)   int64_t head_stride = key_cache.stride(2);    

Sample script to reproduce:

#!/bin/bash
set -eux

apt-get install --yes iproute2

LOCAL_IP=$(ip route get 1.2.3.4 | cut -d' ' -f7)
MODEL="meta-llama/Meta-Llama-3.1-8B-Instruct"

start_vllm() {
    instance_name=$1
    port=$2
    visible_devices=$3
    kv_role=$4
    kv_rank=$5
    tp_flag=$6

    CUDA_VISIBLE_DEVICES=${visible_devices} LD_LIBRARY_PATH=/usr/local/lib/:$LD_LIBRARY_PATH MOONCAKE_CONFIG_PATH=./mooncake.json vllm serve ${MODEL} \
        --port ${port} \
        --max-model-len 12000 \
        --max-num-batched-tokens 12000 \
        --gpu-memory-utilization 0.95 \
        ${tp_flag} \
        --kv-transfer-config '{"kv_connector":"MooncakeStoreConnector", "kv_role":"'$kv_role'", "kv_rank":'$kv_rank', "kv_parallel_size":2}' &
}

etcd --listen-client-urls http://${LOCAL_IP}:2379 --advertise-client-urls http://${LOCAL_IP}:2379 &
mooncake_master --port 50001 &

start_vllm "instance_1" 8100 "0" kv_producer 0 ""
start_vllm "instance_2" 8200 "4" kv_consumer 1 ""

# Confirm that the server is running
sleep 40
until curl http://localhost:8200/health; do sleep 5; done

python3 disagg_proxy_demo.py --model $MODEL --prefill localhost:8100 --decode localhost:8200 --port 8000 &

sleep 10

curl -X POST -s http://localhost:8000/v1/completions -H "Content-Type: application/json" -d '{"model": "'$MODEL'","prompt": "San Francisco is a","max_
tokens": 50,"temperature": 0, "top_k":1}' | awk -F'"' '{print $22}' &

wait

Error message:

ERROR 05-28 10:18:40 [engine.py:164] IndexError('Dimension out of range (expected to be in range of [-2, 1], but got 2)')^M
ERROR 05-28 10:18:40 [engine.py:164] Traceback (most recent call last):^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/engine/multiprocessing/engine.py", line 162, in start^M
ERROR 05-28 10:18:40 [engine.py:164]     self.run_engine_loop()^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/engine/multiprocessing/engine.py", line 225, in run_engine_loop^M
ERROR 05-28 10:18:40 [engine.py:164]     request_outputs = self.engine_step()^M
ERROR 05-28 10:18:40 [engine.py:164]                       ^^^^^^^^^^^^^^^^^^^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/engine/multiprocessing/engine.py", line 251, in engine_step^M
ERROR 05-28 10:18:40 [engine.py:164]     raise e^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/engine/multiprocessing/engine.py", line 234, in engine_step^M
ERROR 05-28 10:18:40 [engine.py:164]     return self.engine.step()^M
ERROR 05-28 10:18:40 [engine.py:164]            ^^^^^^^^^^^^^^^^^^^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/engine/llm_engine.py", line 1393, in step^M
ERROR 05-28 10:18:40 [engine.py:164]     outputs = self.model_executor.execute_model(^M
ERROR 05-28 10:18:40 [engine.py:164]               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/executor/executor_base.py", line 140, in execute_model^M
ERROR 05-28 10:18:40 [engine.py:164]     output = self.collective_rpc("execute_model",^M
ERROR 05-28 10:18:40 [engine.py:164]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/executor/uniproc_executor.py", line 56, in collective_rpc^M
ERROR 05-28 10:18:40 [engine.py:164]     answer = run_method(self.driver_worker, method, args, kwargs)^M
ERROR 05-28 10:18:40 [engine.py:164]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/utils.py", line 2605, in run_method^M
ERROR 05-28 10:18:40 [engine.py:164]     return func(*args, **kwargs)^M
ERROR 05-28 10:18:40 [engine.py:164]            ^^^^^^^^^^^^^^^^^^^^^^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/worker/worker_base.py", line 420, in execute_model^M
ERROR 05-28 10:18:40 [engine.py:164]     output = self.model_runner.execute_model(^M
ERROR 05-28 10:18:40 [engine.py:164]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/usr/local/lib/python3.12/dist-packages/torch/utils/_contextlib.py", line 116, in decorate_context^M
ERROR 05-28 10:18:40 [engine.py:164]     return func(*args, **kwargs)^M
ERROR 05-28 10:18:40 [engine.py:164]            ^^^^^^^^^^^^^^^^^^^^^^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/worker/model_runner.py", line 1817, in execute_model^M
ERROR 05-28 10:18:40 [engine.py:164]     get_kv_transfer_group().recv_kv_caches_and_hidden_states(^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/distributed/kv_transfer/kv_connector/mooncake_store_connector.py", line 174, in recv_kv_caches_  and_hidden_states^M
ERROR 05-28 10:18:40 [engine.py:164]     self.kv_helper.put_kv_to_cache(model_executable, remote_k,^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/distributed/kv_transfer/kv_connector/utils.py", line 82, in put_kv_to_cache^M
ERROR 05-28 10:18:40 [engine.py:164]     ops.reshape_and_cache_flash(^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/opt/vllm/vllm/_custom_ops.py", line 1564, in reshape_and_cache_flash^M
ERROR 05-28 10:18:40 [engine.py:164]     torch.ops._C_cache_ops.reshape_and_cache_flash(key, value, key_cache,^M
ERROR 05-28 10:18:40 [engine.py:164]   File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 1158, in __call__^M
ERROR 05-28 10:18:40 [engine.py:164]     return self._op(*args, **(kwargs or {}))^M
ERROR 05-28 10:18:40 [engine.py:164]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^M
ERROR 05-28 10:18:40 [engine.py:164] IndexError: Dimension out of range (expected to be in range of [-2, 1], but got 2)

### Before submitting a new issue...

- [x] Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the [documentation page](https://docs.vllm.ai/en/latest/), which can answer lots of frequently asked questions.

[ziruiliu]

Hit the same issue while running examples/offline_inference/disaggregated_prefill.py

The output of python collect_env.py

==============================
System Info

OS : Ubuntu 24.04.2 LTS (x86_64)
GCC version : (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version : Could not collect
CMake version : version 3.28.3
Libc version : glibc-2.39

==============================
PyTorch Info

PyTorch version : 2.7.0+cu128
Is debug build : False
CUDA used to build PyTorch : 12.8
ROCM used to build PyTorch : N/A

==============================
Python Environment

Python version : 3.12.3 (main, Feb 4 2025, 14:48:35) [GCC 13.3.0] (64-bit runtime)
Python platform : Linux-6.11.0-1014-gcp-x86_64-with-glibc2.39

==============================
CUDA / GPU Info

Is CUDA available : True
CUDA runtime version : Could not collect
CUDA_MODULE_LOADING set to : LAZY
GPU models and configuration :
GPU 0: Tesla T4
GPU 1: Tesla T4

Nvidia driver version : 570.124.06
cuDNN version : Could not collect
HIP runtime version : N/A
MIOpen runtime version : N/A
Is XNNPACK available : True

...some cpu info...

==============================
Versions of relevant libraries

[pip3] numpy==2.2.6
[pip3] nvidia-cublas-cu12==12.8.3.14
[pip3] nvidia-cuda-cupti-cu12==12.8.57
[pip3] nvidia-cuda-nvrtc-cu12==12.8.61
[pip3] nvidia-cuda-runtime-cu12==12.8.57
[pip3] nvidia-cudnn-cu12==9.7.1.26
[pip3] nvidia-cufft-cu12==11.3.3.41
[pip3] nvidia-cufile-cu12==1.13.0.11
[pip3] nvidia-curand-cu12==10.3.9.55
[pip3] nvidia-cusolver-cu12==11.7.2.55
[pip3] nvidia-cusparse-cu12==12.5.7.53
[pip3] nvidia-cusparselt-cu12==0.6.3
[pip3] nvidia-nccl-cu12==2.26.2
[pip3] nvidia-nvjitlink-cu12==12.8.61
[pip3] nvidia-nvtx-cu12==12.8.55
[pip3] pyzmq==26.4.0
[pip3] torch==2.7.0+cu128
[pip3] torchaudio==2.7.0+cu128
[pip3] torchvision==0.22.0+cu128
[pip3] transformers==4.52.4
[pip3] triton==3.3.0
[conda] Could not collect

==============================
vLLM Info

ROCM Version : Could not collect
Neuron SDK Version : N/A
vLLM Version : 0.9.0.1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0 GPU1 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X PHB 0-31 0 N/A
GPU1 PHB X 0-31 0 N/A

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

==============================
Environment Variables

NCCL_CUMEM_ENABLE=0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

To reproduce the bug, install the latest 0.9.0.1 vllm and run examples/offline_inference/disaggregated_prefill.py with 2 T4 GPU and v0 engine
No matter setting enforce_eager to True or False, error message is shown similar stack as reported above.

[67lc]

I have the same problem,Did you solve the problem?

[fdingiit]

same problem, downgrade to 0.8.4 to pass by

[ziruiliu]

The bug is caused by mismatched tensor shape passed to reshape_and_cache_flash() in
vllm/distributed/kv_transfer/kv_connector/utils.py: put_kv_to_cache().
The key_cache and value_cache are expected to be 4-D but put_kv_to_cache() passes 2-D tensor in.
I believed this bug has been for long, but older version of reshape_and_cache_flash() does not explictly require 4-D
key/value_cache, which means no fault is reported but cache data is unpredictable, and so is decode result

I have a quick and dirty fix to put_kv_to_cache and get_kv_to_cache to make this PD disaggr on v0 engine work.
Since put_kv_to_cache() is only for v0 engine, I believe there wont be any official fix to this bug.

[ziruiliu]

Here's the simple workaround for this issue. It works well on T4 GPU with v0 engine.

diff --git a/vllm/distributed/kv_transfer/kv_connector/utils.py b/vllm/distributed/kv_transfer/kv_connector/utils.py
index 493235d724f4e389488d34111073ba1233b3e83b..dc30dff1d9c9f2b110c5eac103f2c77e794c345e 100644
--- a/vllm/distributed/kv_transfer/kv_connector/utils.py
+++ b/vllm/distributed/kv_transfer/kv_connector/utils.py
@@ -1,93 +1,127 @@
 
     def get_kv_from_cache(self, kv_cache, num_heads, head_size):
-        if self.is_deepseek_mla and self.use_mla_opt:
+        """Return key and value tensors in a flattened view.
+
+        The layout of ``kv_cache`` depends on the attention backend. 3-D
+        tensors correspond to the paged attention layout of shape
+        ``[2, num_blocks, block_size * num_heads * head_size]``. In this case
+        we first split the cache and then reshape so that the first dimension
+        indexes tokens directly. Other layouts (4-D/5-D) follow the previous
+        behavior.
+        """
+
+        if kv_cache.dim() == 3:
+            key_cache, value_cache = PagedAttention.split_kv_cache(
+                kv_cache, num_heads, head_size)
+            key_cache = key_cache.permute(0, 3, 1, 2, 4).reshape(
+                -1, num_heads, head_size)
+            value_cache = value_cache.permute(0, 3, 1, 2).reshape(
+                -1, num_heads, head_size)
+        elif self.is_deepseek_mla and self.use_mla_opt:
             key_cache = kv_cache.reshape(-1, num_heads, head_size)
             value_cache = kv_cache.reshape(-1, num_heads, head_size)
         else:
             key_cache = kv_cache[0].reshape(-1, num_heads, head_size)
             value_cache = kv_cache[1].reshape(-1, num_heads, head_size)
+
         return key_cache, value_cache
 
     def put_kv_to_cache(self, model_executable: torch.nn.Module, keys, values,
                         layer, kv_cache, slot_mapping, start_pos, end_pos):
 
         model_config = model_executable.model.config
+        num_heads, head_size = self.get_model_args(model_executable)
 
-        if self.is_deepseek_mla and self.use_mla_opt:
+        if kv_cache.dim() == 3:
+            # 3-D paged attention cache: [2, num_blocks, block_size * num_heads * head_size]
+            key_cache, value_cache = PagedAttention.split_kv_cache(
+                kv_cache, num_heads, head_size)
+            PagedAttention.write_to_paged_cache(
+                keys.to(kv_cache.device),
+                values.to(kv_cache.device),
+                key_cache,
+                value_cache,
+                slot_mapping[start_pos:end_pos],
+                layer.self_attn.attn.kv_cache_dtype,
+                layer.self_attn.attn._k_scale,
+                layer.self_attn.attn._v_scale,
+            )
+        elif self.is_deepseek_mla and self.use_mla_opt:
             layer.self_attn.attn = layer.self_attn.mla_attn
             k_c_normed_k_pe = keys.squeeze(1)
             k_c_normed = k_c_normed_k_pe[:, :model_config.kv_lora_rank]
             k_pe = k_c_normed_k_pe[:, model_config.kv_lora_rank:]
             ops.concat_and_cache_mla(
                 k_c_normed.to(kv_cache.device),
                 k_pe.to(kv_cache.device),
                 kv_cache,
                 slot_mapping[start_pos:end_pos],
                 layer.self_attn.attn.kv_cache_dtype,
                 layer.self_attn.attn._k_scale,
             )
         else:
             key_cache, value_cache = kv_cache[0], kv_cache[1]
             ops.reshape_and_cache_flash(
                 keys.to(key_cache.device),
                 values.to(value_cache.device),
                 key_cache,
                 value_cache,
                 slot_mapping[start_pos:end_pos],
                 layer.self_attn.attn.kv_cache_dtype,
                 layer.self_attn.attn._k_scale,
                 layer.self_attn.attn._v_scale,
             )