DeepSeek-V4

DeepSeek-V4#

Introduction#

DeepSeek-V4 is introducing several key upgrades over DeepSeek-V3. (Currently, vllm-ascend temporarily only supports DeepSeek-V4-FLASH)

The Manifold-Constrained Hyper-Connections (mHC) to strengthen conventional residual connections;
A hybrid attention architecture, which greatly improves long-context efficiency through Compress-4-Attention and Compress-128-Attention. For the Mixture-of Experts (MoE) components, it still adopt the DeepSeekMoE architecture, with only minor adjustments.

This document will show the main verification steps of the model, including supported features, feature configuration, environment preparation, single-node and multi-node deployment, accuracy and performance evaluation.

Environment Preparation#

Model Weight#

DeepSeek-V4-Flash-w8a8-mtp(Quantized version): require 1 Atlas 800 A3 (128G × 8) node or 1 Atlas 800 A2 (64G × 8) node. Download model weight

It is recommended to download the model weight to the shared directory of multiple nodes, such as /root/.cache/

Verify Multi-node Communication(Optional)#

If you want to deploy multi-node environment, you need to verify multi-node communication according to verify multi-node communication environment.

Installation#

You can using our official docker image to run DeepSeek-V4 directly. Currently, DeepSeek-V4 is integrated in image v0.13.0rc3.

A2 series

Start the docker image on your each node.

export IMAGE=quay.io/ascend/vllm-ascend:v0.13.0rc3
export NAME=vllm-ascend
docker run --rm \
    --name $NAME \
    --net=host \
    --shm-size=1g \
    --device /dev/davinci0 \
    --device /dev/davinci1 \
    --device /dev/davinci2 \
    --device /dev/davinci3 \
    --device /dev/davinci4 \
    --device /dev/davinci5 \
    --device /dev/davinci6 \
    --device /dev/davinci7 \
    --device /dev/davinci_manager \
    --device /dev/devmm_svm \
    --device /dev/hisi_hdc \
    -v /usr/local/dcmi:/usr/local/dcmi \
    -v /usr/local/Ascend/driver/tools/hccn_tool:/usr/local/Ascend/driver/tools/hccn_tool \
    -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
    -v /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/ \
    -v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info \
    -v /etc/ascend_install.info:/etc/ascend_install.info \
    -v /etc/hccn.conf:/etc/hccn.conf \
    -v /mnt/sfs_turbo/.cache:/root/.cache \
    -it $IMAGE bash

A3 series

Start the docker image on your each node.

export IMAGE=quay.io/ascend/vllm-ascend:v0.13.0rc3-a3
export NAME=vllm-ascend
docker run --rm \
    --name $NAME \
    --net=host \
    --shm-size=1g \
    --device /dev/davinci0 \
    --device /dev/davinci1 \
    --device /dev/davinci2 \
    --device /dev/davinci3 \
    --device /dev/davinci4 \
    --device /dev/davinci5 \
    --device /dev/davinci6 \
    --device /dev/davinci7 \
    --device /dev/davinci8 \
    --device /dev/davinci9 \
    --device /dev/davinci10 \
    --device /dev/davinci11 \
    --device /dev/davinci12 \
    --device /dev/davinci13 \
    --device /dev/davinci14 \
    --device /dev/davinci15 \
    --device /dev/davinci_manager \
    --device /dev/devmm_svm \
    --device /dev/hisi_hdc \
    -v /usr/local/dcmi:/usr/local/dcmi \
    -v /usr/local/Ascend/driver/tools/hccn_tool:/usr/local/Ascend/driver/tools/hccn_tool \
    -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
    -v /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/ \
    -v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info \
    -v /etc/ascend_install.info:/etc/ascend_install.info \
    -v /etc/hccn.conf:/etc/hccn.conf \
    -v /mnt/sfs_turbo/.cache:/root/.cache \
    -it $IMAGE bash

In addition, if you don’t want to use the docker image as above, you can also build all from source:

Install vllm-ascend from source, refer to installation. If you want to deploy multi-node environment, you need to set up environment on each node.

Note

Please use the v0.13.0rc3 code to install vllm-ascend.

Deployment#

Note

In this tutorial, we suppose you downloaded the model weight to /root/.cache/. Feel free to change it to your own path.

Single-node Deployment#

DeepSeek-V4-Flash-w8a8-mtp: can be deployed on 1 Atlas 800 A3 (128G × 8) or 1 Atlas 800 A2 (64G × 8).

Run the following scripts on each node respectively.

A2 series

Run the following script to execute online inference.

export USE_MULTI_BLOCK_POOL=1
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export ACL_OP_INIT_MODE=1
export TRITON_ALL_BLOCKS_PARALLEL=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
  --host 0.0.0.0 \
  --max_model_len 65536 \
  --max-num-batched-tokens 8192 \
  --served-model-name ds \
  --gpu-memory-utilization 0.9 \
  --max-num-seqs 16 \
  --data-parallel-size 1 \
  --tensor-parallel-size 8 \
  --enable-expert-parallel \
  --quantization ascend \
  --port 8006 \
  --block-size 128 \
  --chat-template /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp/chat_template.jinja \
  --async-scheduling \
  --additional-config '{"enable_cpu_binding": "true", "multistream_overlap_shared_expert": true}' \
  --speculative-config '{"num_speculative_tokens": 1,"method": "deepseek_mtp"}' \
  --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY"}'

A3 series

Run the following script to execute online inference.

export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10  
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True  
export ACL_OP_INIT_MODE=1
export ASCEND_A3_ENABLE=1
export USE_MULTI_BLOCK_POOL=1
export HCCL_BUFFSIZE=1024
export VLLM_ASCEND_ENABLE_FUSED_MC2=1
export VLLM_ASCEND_ENABLE_FLASHCOMM1=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
    --host 0.0.0.0 \
    --max_model_len 65536 \
    --max-num-batched-tokens 8192 \
    --served-model-name deepseek_v4 \
    --gpu-memory-utilization 0.9 \
    --max-num-seqs 16 \
    --data-parallel-size 2 \
    --tensor-parallel-size 8 \
    --enable-expert-parallel \
    --quantization ascend \
    --chat-template /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp/chat_template.jinja \
    --port 8005 \
    --block-size 128 \
    --async-scheduling \
    --compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}'\
    --speculative-config '{"num_speculative_tokens": 1,"method": "deepseek_mtp"}' \
    --additional-config '{"enable_cpu_binding": "true","multistream_overlap_shared_expert": false}'

Multi-node Deployment#

DeepSeek-V4-Flash-w8a8-mtp: require at least 2 Atlas 800 A2 (64G × 8). Run the following scripts on two nodes respectively.

A2 series

Node0

# this obtained through ifconfig
# nic_name is the network interface name corresponding to local_ip of the current node
nic_name="xxx"
local_ip="xxx"

# The value of node0_ip must be consistent with the value of local_ip set in node0 (master node)
node0_ip="xxxx"

export HCCL_OP_EXPANSION_MODE="AIV"

export HCCL_IF_IP=$local_ip
export GLOO_SOCKET_IFNAME=$nic_name
export TP_SOCKET_IFNAME=$nic_name
export HCCL_SOCKET_IFNAME=$nic_name
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export HCCL_BUFFSIZE=200
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_CONNECT_TIMEOUT=120
export HCCL_INTRA_PCIE_ENABLE=1
export HCCL_INTRA_ROCE_ENABLE=0
export ACL_OP_INIT_MODE=1
export TRITON_ALL_BLOCKS_PARALLEL=1
export USE_MULTI_BLOCK_POOL=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
--host 0.0.0.0 \
--port 8005 \
--data-parallel-size 2 \
--data-parallel-size-local 1 \
--data-parallel-address $node0_ip \
--data-parallel-rpc-port 13389 \
--tensor-parallel-size 8 \
--quantization ascend \
--seed 1024 \
--served-model-name deepseek-v4-flash \
--enable-expert-parallel \
--max-num-seqs 64 \
--max-model-len 131072 \
--max-num-batched-tokens 8192 \
--trust-remote-code \
--async-scheduling \
--no-enable-prefix-caching \
--chat-template /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp/chat_template.jinja \
--gpu-memory-utilization 0.94 \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--additional-config '{"enable_cpu_binding": "true", "multistream_overlap_shared_expert": true}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp"}'

Node1

# this obtained through ifconfig
# nic_name is the network interface name corresponding to local_ip of the current node
nic_name="xxx"
local_ip="xxx"

# The value of node0_ip must be consistent with the value of local_ip set in node0 (master node)
node0_ip="xxxx"

export HCCL_OP_EXPANSION_MODE="AIV"

export HCCL_IF_IP=$local_ip
export GLOO_SOCKET_IFNAME=$nic_name
export TP_SOCKET_IFNAME=$nic_name
export HCCL_SOCKET_IFNAME=$nic_name
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export HCCL_BUFFSIZE=200
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_CONNECT_TIMEOUT=120
export HCCL_INTRA_PCIE_ENABLE=1
export HCCL_INTRA_ROCE_ENABLE=0
export ACL_OP_INIT_MODE=1
export TRITON_ALL_BLOCKS_PARALLEL=1
export USE_MULTI_BLOCK_POOL=1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
--host 0.0.0.0 \
--port 8005 \
--headless \
--data-parallel-size 2 \
--data-parallel-size-local 1 \
--data-parallel-start-rank 1 \
--data-parallel-address $node0_ip \
--data-parallel-rpc-port 13389 \
--tensor-parallel-size 8 \
--quantization ascend \
--seed 1024 \
--served-model-name deepseek-v4-flash \
--enable-expert-parallel \
--max-num-seqs 64 \
--max-model-len 131072 \
--max-num-batched-tokens 8192 \
--trust-remote-code \
--async-scheduling \
--no-enable-prefix-caching \
--gpu-memory-utilization 0.94 \
--chat-template /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp/chat_template.jinja \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
--additional-config '{"enable_cpu_binding": "true", "multistream_overlap_shared_expert": true}' \
--speculative-config '{"num_speculative_tokens": 3, "method": "deepseek_mtp"}'

Prefill-Decode Disaggregation#

We’d like to show the deployment guide of DeepSeek-V4 on Atlas 800 A3 (128G × 8) multi-node environment with 2P1D for better performance.

Before you start, please

prepare the script launch_online_dp.py on each node.

import argparse
import multiprocessing
import os
import subprocess
import sys

def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--dp-size",
        type=int,
        required=True,
        help="Data parallel size."
    )
    parser.add_argument(
        "--tp-size",
        type=int,
        default=1,
        help="Tensor parallel size."
    )
    parser.add_argument(
        "--dp-size-local",
        type=int,
        default=-1,
        help="Local data parallel size."
    )
    parser.add_argument(
        "--dp-rank-start",
        type=int,
        default=0,
        help="Starting rank for data parallel."
    )
    parser.add_argument(
        "--dp-address",
        type=str,
        required=True,
        help="IP address for data parallel master node."
    )
    parser.add_argument(
        "--dp-rpc-port",
        type=str,
        default=12345,
        help="Port for data parallel master node."
    )
    parser.add_argument(
        "--vllm-start-port",
        type=int,
        default=9000,
        help="Starting port for the engine."
    )
    return parser.parse_args()

args = parse_args()
dp_size = args.dp_size
tp_size = args.tp_size
dp_size_local = args.dp_size_local
if dp_size_local == -1:
    dp_size_local = dp_size
dp_rank_start = args.dp_rank_start
dp_address = args.dp_address
dp_rpc_port = args.dp_rpc_port
vllm_start_port = args.vllm_start_port

def run_command(visiable_devices, dp_rank, vllm_engine_port):
    command = [
        "bash",
        "./run_dp_template.sh",
        visiable_devices,
        str(vllm_engine_port),
        str(dp_size),
        str(dp_rank),
        dp_address,
        dp_rpc_port,
        str(tp_size),
    ]
    subprocess.run(command, check=True)

if __name__ == "__main__":
    template_path = "./run_dp_template.sh"
    if not os.path.exists(template_path):
        print(f"Template file {template_path} does not exist.")
        sys.exit(1)

    processes = []
    num_cards = dp_size_local * tp_size
    for i in range(dp_size_local):
        dp_rank = dp_rank_start + i
        vllm_engine_port = vllm_start_port + i
        visiable_devices = ",".join(str(x) for x in range(i * tp_size, (i + 1) * tp_size))
        process = multiprocessing.Process(target=run_command,
                                        args=(visiable_devices, dp_rank,
                                                vllm_engine_port))
        processes.append(process)
        process.start()

    for process in processes:
        process.join()

prepare the script run_dp_template.sh on each node.

Prefill node 1

nic_name="xxxx" # change to your own nic name
local_ip=xx.xx.xx.1 # change to your own ip

export HCCL_OP_EXPANSION_MODE="AIV"

export HCCL_IF_IP=$local_ip
export GLOO_SOCKET_IFNAME=$nic_name
export TP_SOCKET_IFNAME=$nic_name
export HCCL_SOCKET_IFNAME=$nic_name

export VLLM_RPC_TIMEOUT=3600000
export VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=30000
export HCCL_EXEC_TIMEOUT=204
export HCCL_CONNECT_TIMEOUT=120

export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=2560
export TASK_QUEUE_ENABLE=1

export ASCEND_BUFFER_POOL=4:8
export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
export USE_MULTI_BLOCK_POOL=1

export ASCEND_RT_VISIBLE_DEVICES=$1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
    --host 0.0.0.0 \
    --port $2 \
    --data-parallel-size $3 \
    --data-parallel-rank $4 \
    --data-parallel-address $5 \
    --data-parallel-rpc-port $6 \
    --tensor-parallel-size $7 \
    --enable-expert-parallel \
    --seed 1024 \
    --served-model-name deepseek_v4 \
    --max-model-len 65536 \
    --max-num-batched-tokens 8192 \
    --max-num-seqs 4 \
    --no-disable-hybrid-kv-cache-manager \
    --no-enable-prefix-caching \
    --trust-remote-code \
    --gpu-memory-utilization 0.85 \
    --quantization ascend \
    --chat-template /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp/chat_template.jinja \
    --speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}' \
    --enforce-eager \
    --additional_config '{"enable_cpu_binding": "true"}' \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeConnectorV1",
    "kv_role": "kv_producer",
    "kv_port": "30000",
    "engine_id": "0",
    "kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
    "kv_connector_extra_config": {
                "prefill": {
                        "dp_size": 16,
                        "tp_size": 1
                },
                "decode": {
                        "dp_size": 32,
                        "tp_size": 1
                }
        }
    }'

Prefill node 2

nic_name="xxxx" # change to your own nic name
local_ip=xx.xx.xx.2 # change to your own ip

export HCCL_OP_EXPANSION_MODE="AIV"

export HCCL_IF_IP=$local_ip
export GLOO_SOCKET_IFNAME=$nic_name
export TP_SOCKET_IFNAME=$nic_name
export HCCL_SOCKET_IFNAME=$nic_name

export VLLM_RPC_TIMEOUT=3600000
export VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=30000
export HCCL_EXEC_TIMEOUT=204
export HCCL_CONNECT_TIMEOUT=120

export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=2560
export TASK_QUEUE_ENABLE=1

export ASCEND_BUFFER_POOL=4:8
export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
export USE_MULTI_BLOCK_POOL=1

export ASCEND_RT_VISIBLE_DEVICES=$1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
    --host 0.0.0.0 \
    --port $2 \
    --data-parallel-size $3 \
    --data-parallel-rank $4 \
    --data-parallel-address $5 \
    --data-parallel-rpc-port $6 \
    --tensor-parallel-size $7 \
    --enable-expert-parallel \
    --seed 1024 \
    --served-model-name deepseek_v4 \
    --max-model-len 65536 \
    --max-num-batched-tokens 8192 \
    --max-num-seqs 4 \
    --no-disable-hybrid-kv-cache-manager \
    --no-enable-prefix-caching \
    --trust-remote-code \
    --gpu-memory-utilization 0.85 \
    --quantization ascend \
    --chat-template /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp/chat_template.jinja \
    --speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}' \
    --enforce-eager \
    --additional_config '{"enable_cpu_binding": "true"}' \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeConnectorV1",
    "kv_role": "kv_producer",
    "kv_port": "30100",
    "engine_id": "1",
    "kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
    "kv_connector_extra_config": {
                "prefill": {
                        "dp_size": 16,
                        "tp_size": 1
                },
                "decode": {
                        "dp_size": 32,
                        "tp_size": 1
                }
        }
    }'

Decode node (Same as another D node)

nic_name="xxxx" # change to your own nic name
local_ip=xx.xx.xx.xx # change to your own ip

export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
export HCCL_OP_EXPANSION_MODE="AIV"
export TASK_QUEUE_ENABLE=1

export VLLM_RPC_TIMEOUT=3600000
export VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=30000
export HCCL_EXEC_TIMEOUT=2000
export HCCL_CONNECT_TIMEOUT=1200

export HCCL_IF_IP=$local_ip
export GLOO_SOCKET_IFNAME=$nic_name
export TP_SOCKET_IFNAME=$nic_name
export HCCL_SOCKET_IFNAME=$nic_name

export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export HCCL_BUFFSIZE=1024
export ASCEND_BUFFER_POOL=4:8

export USE_MULTI_BLOCK_POOL=1
export VLLM_ASCEND_ENABLE_FUSED_MC2=1
export ASCEND_RT_VISIBLE_DEVICES=$1

vllm serve /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp \
    --host 0.0.0.0 \
    --port $2 \
    --data-parallel-size $3 \
    --data-parallel-rank $4 \
    --data-parallel-address $5 \
    --data-parallel-rpc-port $6 \
    --tensor-parallel-size $7 \
    --enable-expert-parallel \
    --seed 1024 \
    --served-model-name deepseek_v4 \
    --max-model-len 65536 \
    --max-num-batched-tokens 144 \
    --max-num-seqs 48 \
    --async-scheduling \
    --no-disable-hybrid-kv-cache-manager \
    --no-enable-prefix-caching \
    --trust-remote-code \
    --gpu-memory-utilization 0.88 \
    --quantization ascend \
    --chat-template /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp/chat_template.jinja \
    --speculative-config '{"num_speculative_tokens": 2, "method":"deepseek_mtp"}' \
    --compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY","cudagraph_capture_sizes":[144]}' \
    --kv-transfer-config \
    '{"kv_connector": "MooncakeConnectorV1",
    "kv_role": "kv_consumer",
    "kv_port": "30200",
    "engine_id": "2",
    "kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
    "kv_connector_extra_config": {
                "prefill": {
                        "dp_size": 16,
                        "tp_size": 1
                },
                "decode": {
                        "dp_size": 32,
                        "tp_size": 1
                }
        }
    }' \
    --additional_config '{"enable_cpu_binding": "true", "multistream_overlap_shared_expert": false, "multistream_dsa_preprocess": false}'

Once the preparation is done, you can start the server with the following command on each node:

Prefill node 0

# change ip to your own
python launch_online_dp.py --dp-size 16 --tp-size 1 --dp-size-local 16 --dp-rank-start 0 --dp-address xx.xx.xx.1 --dp-rpc-port 12321 --vllm-start-port 7100

Prefill node 1

# change ip to your own
python launch_online_dp.py --dp-size 16 --tp-size 1 --dp-size-local 16 --dp-rank-start 0 --dp-address xx.xx.xx.2 --dp-rpc-port 12321 --vllm-start-port 7100

Decode node 0

# change ip to your own
python launch_online_dp.py --dp-size 32 --dp-size-local 16 --dp-rank-start 0 --dp-address xx.xx.xx.3 --dp-rpc-port 12321 --vllm-start-port 7100

Decode node 1

# change ip to your own
python launch_online_dp.py --dp-size 32 --dp-size-local 16 --dp-rank-start 16 --dp-address xx.xx.xx.3 --dp-rpc-port 12321 --vllm-start-port 7100

Finally, Refer to Prefill-Decode Disaggregation (Deepseek) to deploy the P-D disaggregation proxy.

Functional Verification#

Once your server is started, you can query the model with input prompts:

curl http://<node0_ip>:<port>/v1/chat/completions \
    -H "Content-Type: application/json" \
    -d '{
        "model": "deepseek_v4",
        "messages": [
            {
                "role": "user",
                "content": "Who are you?"
            }
        ],
        "max_tokens": 256,
        "temperature": 0
    }'

Accuracy Evaluation#

Here are two accuracy evaluation methods.

Using AISBench#

Refer to Using AISBench for details.
After execution, you can get the result.

Using Language Model Evaluation Harness#

As an example, take the gsm8k dataset as a test dataset, and run accuracy evaluation of DeepSeek-V4 in online mode.

Refer to Using lm_eval for lm_eval installation.
Run lm_eval to execute the accuracy evaluation.

lm_eval \
  --model local-completions \
  --model_args model=/root/.cache/Eco-Tech/DeepSeek-V4-Flash-w8a8-mtp,base_url=http://127.0.0.1:8006/v1/completions,tokenized_requests=False,trust_remote_code=True \
  --tasks gsm8k \
  --output_path ./

After execution, you can get the result.

Performance#

Using AISBench#

Refer to Using AISBench for performance evaluation for details.

Using vLLM Benchmark#

Run performance evaluation of DeepSeek-V4-Flash-w8a8-mtp as an example.

Refer to vllm benchmark for more details.

There are three vllm bench subcommand:

latency: Benchmark the latency of a single batch of requests.
serve: Benchmark the online serving throughput.
throughput: Benchmark offline inference throughput.

Take the serve as an example. Run the code as follows.

export VLLM_USE_MODELSCOPE=true
vllm bench serve --model /root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V4-Flash-w8a8-mtp  --dataset-name random --random-input 200 --num-prompt 200 --request-rate 1 --save-result --result-dir ./

DeepSeek-V4

Contents

DeepSeek-V4#

Introduction#

Environment Preparation#

Model Weight#

Verify Multi-node Communication(Optional)#

Installation#

Deployment#

Single-node Deployment#

Multi-node Deployment#

Prefill-Decode Disaggregation#

Functional Verification#

Accuracy Evaluation#

Using AISBench#

Using Language Model Evaluation Harness#

Performance#

Using AISBench#

Using vLLM Benchmark#