EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[P/D][main]Offline the llmdatadist connector related parts of the code and files. (#4780)

Browse Source 
### What this PR does / why we need it?
As support for the mooncake connector is now available, the llmdatadist
connector is no longer being maintained, so the llmdatadist-related
files need to be retired.

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
By ci

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

---------

Signed-off-by: wangxiaoteng <wangxiaoteng@huawei.com>
Signed-off-by: liziyu <liziyu16@huawei.com>
Co-authored-by: liziyu <liziyu16@huawei.com>
This commit is contained in:
wangxiaoteng888
2025-12-09 22:36:43 +08:00
committed by GitHub
parent 848419d1ba
commit a77045f355
19 changed files with 188 additions and 1819 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
.github/workflows/vllm_ascend_test_pr_light.yaml vendored
View File

@@ -139,7 +139,6 @@ jobs:
--ignore tests/ut/model_loader/netloader/test_netloader_elastic.py \
--ignore tests/ut/kv_connector/test_remote_prefill_lifecycle.py \
--ignore tests/ut/kv_connector/test_remote_decode_lifecycle.py \
--ignore tests/ut/kv_connector/test_llmdatadist_connector.py \
--ignore tests/ut/core/test_scheduler_dynamic_batch.py
- name: Upload coverage to Codecov

2
docs/source/faqs.md
View File

@@ -104,7 +104,7 @@ vllm-ascend is a hardware plugin for vLLM. Basically, the version of vllm-ascend
### 8. Does vllm-ascend support Prefill Disaggregation feature?
Yes, vllm-ascend supports Prefill Disaggregation feature with LLMdatadist, Mooncake backend. Take [official tutorial](https://vllm-ascend.readthedocs.io/en/latest/tutorials/multi_node_pd_disaggregation_llmdatadist.html) for example.
Yes, vllm-ascend supports Prefill Disaggregation feature with Mooncake backend. Take [official tutorial](https://vllm-ascend.readthedocs.io/en/latest/tutorials/multi_node_pd_disaggregation_mooncake.html) for example.
### 9. Does vllm-ascend support quantization method?

2
docs/source/tutorials/index.md
View File

@@ -9,7 +9,7 @@ single_npu_qwen2_audio
single_npu_qwen3_embedding
single_npu_qwen3_quantization
single_npu_qwen3_w4a4
single_node_pd_disaggregation_llmdatadist
single_node_pd_disaggregation_mooncake
multi_npu_qwen3_next
multi_npu
multi_npu_moge

54
docs/source/tutorials/single_node_pd_disaggregation_llmdatadist.md → docs/source/tutorials/single_node_pd_disaggregation_mooncake.md
View File

@@ -1,4 +1,4 @@
# Prefill-Decode Disaggregation Llmdatadist Verification (Qwen2.5-VL)
# Prefill-Decode Disaggregation Mooncake Verification (Qwen2.5-VL)
## Getting Start
@@ -69,10 +69,8 @@ export HCCL_IF_IP=192.0.0.1 # node ip
export GLOO_SOCKET_IFNAME="eth0" # network card name
export TP_SOCKET_IFNAME="eth0"
export HCCL_SOCKET_IFNAME="eth0"
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH="/path/to/your/generated/ranktable.json"
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export VLLM_ASCEND_LLMDD_RPC_PORT=5959
vllm serve /model/Qwen2.5-VL-7B-Instruct \
--host 0.0.0.0 \
@@ -85,14 +83,22 @@ vllm serve /model/Qwen2.5-VL-7B-Instruct \
--max-num-batched-tokens 40000 \
--trust-remote-code \
--gpu-memory-utilization 0.9 \
--kv-transfer-config \
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
"kv_role": "kv_producer",
"kv_parallel_size": 1,
"kv_port": "20001",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
--kv-transfer-config \
'{"kv_connector": "MooncakeConnector",
"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": 1,
"tp_size": 1
},
"decode": {
"dp_size": 1,
"tp_size": 1
}
}
}'
```
@@ -106,10 +112,8 @@ export HCCL_IF_IP=192.0.0.1 # node ip
export GLOO_SOCKET_IFNAME="eth0" # network card name
export TP_SOCKET_IFNAME="eth0"
export HCCL_SOCKET_IFNAME="eth0"
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH="/path/to/your/generated/ranktable.json"
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export VLLM_ASCEND_LLMDD_RPC_PORT=5979
vllm serve /model/Qwen2.5-VL-7B-Instruct \
--host 0.0.0.0 \
@@ -122,14 +126,22 @@ vllm serve /model/Qwen2.5-VL-7B-Instruct \
--max-num-batched-tokens 40000 \
--trust-remote-code \
--gpu-memory-utilization 0.9 \
--kv-transfer-config \
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
--kv-transfer-config \
'{"kv_connector": "MooncakeConnector",
"kv_role": "kv_consumer",
"kv_parallel_size": 1,
"kv_port": "20001",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
"kv_port": "30100",
"engine_id": "1",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 1,
"tp_size": 1
},
"decode": {
"dp_size": 1,
"tp_size": 1
}
}
}'
```
@@ -137,7 +149,7 @@ vllm serve /model/Qwen2.5-VL-7B-Instruct \
:::::
If you want to run "2P1D", please set ASCEND_RT_VISIBLE_DEVICES, VLLM_ASCEND_LLMDD_RPC_PORT and port to different values for each P process.
If you want to run "2P1D", please set ASCEND_RT_VISIBLE_DEVICES and port to different values for each P process.
## Example Proxy for Deployment

238
examples/disaggregated_prefill_v1/README.md
View File

@@ -1,238 +0,0 @@
# Disaggregated Prefill-Decode Deployment Guide
## Overview
This demo document provides instructions for running a disaggregated vLLM-ascend service with separate prefill and decode stages across 4 nodes, uses 16 Ascend NPUs for two prefill nodes (P1/P2) and 16 Ascend NPUS for two decode nodes (D1/D2).
## Prerequisites
- Ascend NPU environment with vLLM 0.9.1 installed
- Network interfaces configured for distributed communication (eg: eth0)
- Model weights located at `/models/deepseek_r1_w8a8`
## Rank table generation
The rank table is a JSON file that specifies the mapping of Ascend NPU ranks to nodes. The following command generates a rank table for all nodes with 16 cards prefill and 16 cards decode:
Run the following command on every node to generate the rank table:
```shell
cd /vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/
bash gen_ranktable.sh --ips 172.19.32.175 172.19.241.49 172.19.123.51 172.19.190.36 \
--npus-per-node 8 --network-card-name eth0 --prefill-device-cnt 16 --decode-device-cnt 16
```
Rank table will generated at `/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json`
## Start disaggregated vLLM-ascend service
For demonstration purposes, we will utilize the quantized version of Deepseek-R1. Recommended Parallelization Strategies:
- P-node: DP2-TP8-EP16 (Data Parallelism 2, Tensor Parallelism 8, Expert Parallelism 16)
- D-node: DP4-TP4-EP16 (Data Parallelism 4, Tensor Parallelism 4, Expert Parallelism 16)
Execution Sequence
- 4 configured node ip are: 172.19.32.175 172.19.241.49 172.19.123.51 172.19.190.36
- Start Prefill on Node 1 (P1)
- Start Prefill on Node 2 (P2)
- Start Decode on Node 1 (D1)
- Start Decode on Node 2 (D2)
- Start proxy server on Node1
Run prefill server P1 on first node:
```shell
export HCCL_IF_IP=172.19.32.175 # node ip
export GLOO_SOCKET_IFNAME="eth0" # network card name
export TP_SOCKET_IFNAME="eth0"
export HCCL_SOCKET_IFNAME="eth0"
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export VLLM_ASCEND_LLMDD_RPC_PORT=5559
vllm serve /models/deepseek_r1_w8a8 \
--host 0.0.0.0 \
--port 20002 \
--data-parallel-size 2 \
--data-parallel-size-local 1 \
--api-server-count 2 \
--data-parallel-address 172.19.32.175 \
--data-parallel-rpc-port 13356 \
--tensor-parallel-size 8 \
--enable-expert-parallel \
--seed 1024 \
--served-model-name deepseek \
--max-model-len 32768 \
--max-num-batched-tokens 32768 \
--max-num-seqs 256 \
--trust-remote-code \
--enforce-eager \
--gpu-memory-utilization 0.9 \
--kv-transfer-config \
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
"kv_role": "kv_producer",
"kv_parallel_size": 1,
"kv_port": "20001",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}'
```
Run prefill server P2 on second node:
```shell
export HCCL_IF_IP=172.19.241.49
export GLOO_SOCKET_IFNAME="eth0"
export TP_SOCKET_IFNAME="eth0"
export HCCL_SOCKET_IFNAME="eth0"
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export VLLM_ASCEND_LLMDD_RPC_PORT=5659
vllm serve /models/deepseek_r1_w8a8 \
--host 0.0.0.0 \
--port 20002 \
--headless \
--data-parallel-size 2 \
--data-parallel-start-rank 1 \
--data-parallel-size-local 1 \
--data-parallel-address 172.19.32.175 \
--data-parallel-rpc-port 13356 \
--tensor-parallel-size 8 \
--enable-expert-parallel \
--seed 1024 \
--served-model-name deepseek \
--max-model-len 32768 \
--max-num-batched-tokens 32768 \
--max-num-seqs 256 \
--trust-remote-code \
--enforce-eager \
--gpu-memory-utilization 0.9 \
--kv-transfer-config \
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
"kv_role": "kv_producer",
"kv_parallel_size": 1,
"kv_port": "20001",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}'
```
Run decode server d1 on third node:
* In the D node, the `max-num-batched-tokens` parameter can be set to a smaller value since the D node processes at most `max-num-seqs` batches concurrently. As the `profile_run` only needs to handle `max-num-seqs` sequences at a time, we can safely set `max-num-batched-tokens` equal to `max-num-seqs`. This optimization will help reduce activation memory consumption.
```shell
export HCCL_IF_IP=172.19.123.51
export GLOO_SOCKET_IFNAME="eth0"
export TP_SOCKET_IFNAME="eth0"
export HCCL_SOCKET_IFNAME="eth0"
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export VLLM_ASCEND_LLMDD_RPC_PORT=5759
vllm serve /models/deepseek_r1_w8a8 \
--host 0.0.0.0 \
--port 20002 \
--data-parallel-size 4 \
--data-parallel-size-local 2 \
--api-server-count 2 \
--data-parallel-address 172.19.123.51 \
--data-parallel-rpc-port 13356 \
--tensor-parallel-size 4 \
--enable-expert-parallel \
--seed 1024 \
--served-model-name deepseek \
--max-model-len 32768 \
--max-num-batched-tokens 256 \
--max-num-seqs 256 \
--trust-remote-code \
--gpu-memory-utilization 0.9 \
--kv-transfer-config \
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
"kv_role": "kv_consumer",
"kv_parallel_size": 1,
"kv_port": "20001",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}' \
--additional-config \
'{"torchair_graph_config": {"enabled":true}}'
```
Run decode server d2 on last node:
```shell
export HCCL_IF_IP=172.19.190.36
export GLOO_SOCKET_IFNAME="eth0"
export TP_SOCKET_IFNAME="eth0"
export HCCL_SOCKET_IFNAME="eth0"
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export VLLM_ASCEND_LLMDD_RPC_PORT=5859
vllm serve /models/deepseek_r1_w8a8 \
--host 0.0.0.0 \
--port 20002 \
--headless \
--data-parallel-size 4 \
--data-parallel-start-rank 2 \
--data-parallel-size-local 2 \
--data-parallel-address 172.19.123.51 \
--data-parallel-rpc-port 13356 \
--tensor-parallel-size 4 \
--enable-expert-parallel \
--seed 1024 \
--served-model-name deepseek \
--max-model-len 32768 \
--max-num-batched-tokens 256 \
--max-num-seqs 256 \
--trust-remote-code \
--gpu-memory-utilization 0.9 \
--kv-transfer-config \
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
"kv_role": "kv_consumer",
"kv_parallel_size": 1,
"kv_port": "20001",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
}' \
--additional-config \
'{"torchair_graph_config": {"enabled":true}}'
```
Run proxy server on the first node:
```shell
cd /vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1
python load_balance_proxy_server_example.py --host 172.19.32.175 --port 1025 --prefiller-hosts 172.19.241.49 --prefiller-port 20002 --decoder-hosts 172.19.123.51 --decoder-ports 20002
```
Verification
Check service health using the proxy server endpoint:
```shell
curl http://localhost:1025/v1/completions \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek",
"prompt": "Who are you?",
"max_tokens": 100,
"temperature": 0
}'
```
Performance
Test performance with vllm benchmark:
```shell
cd /vllm-workspace/vllm/benchmarks
python3 benchmark_serving.py \
--backend vllm \
--dataset-name random \
--random-input-len 4096 \
--random-output-len 1536 \
--num-prompts 256 \
--ignore-eos \
--model deepseek \
--tokenizer /models/deepseek_r1_w8a8 \
--host localhost \
--port 1025 \
--endpoint /v1/completions \
--max-concurrency 4 \
--request-rate 4
```

144
examples/disaggregated_prefill_v1/gen_ranktable.py
View File

@@ -1,144 +0,0 @@
import argparse
import json
import os
import torch.distributed as dist
from vllm_ascend.utils import AscendDeviceType, get_ascend_device_type
parser = argparse.ArgumentParser(
description="Arguments of rank table generator", )
parser.add_argument("--local-host", type=str, required=True, help="local ip")
parser.add_argument("--prefill-device-cnt",
type=int,
required=True,
help="number of prefill devices")
parser.add_argument("--decode-device-cnt",
type=int,
required=True,
help="number of decode devices")
parser.add_argument("--local-device-ids",
type=str,
required=False,
help="local device ids")
parser.add_argument("--ranktable-path",
type=str,
default="./ranktable.json",
help="output rank table path")
args = parser.parse_args()
local_host = args.local_host
prefill_device_cnt = args.prefill_device_cnt
decode_device_cnt = args.decode_device_cnt
print("enter py")
hccn_tool_path = os.environ.get("HCCN_TOOL_PATH",
"/usr/local/Ascend/driver/tools/hccn_tool")
master_addr = os.environ.get("MASTER_ADDR")
master_port = os.environ.get("MASTER_PORT")
rank = os.environ.get("RANK")
local_rank = os.environ.get("LOCAL_RANK")
# This variable is set by torchrun,
# and is different from WORLD_SIZE in gen_rank_table.sh.
world_size = os.environ.get("WORLD_SIZE")
device_type = get_ascend_device_type()
def get_cmd_stdout(cmd):
import subprocess
return subprocess.run(cmd, capture_output=True,
shell=True).stdout.decode("utf-8").strip()
print(f"local_host: {local_host}")
print("gen ranktable.json")
num_cards = get_cmd_stdout("npu-smi info -l | grep \"Total Count\"").split(
":")[1].strip()
num_cards = int(num_cards)
chips_per_card = get_cmd_stdout("npu-smi info -l | grep \"Chip Count\"").split(
"\n")[0].split(":")[1].strip()
chips_per_card = int(chips_per_card)
if args.local_device_ids:
try:
local_device_ids = [int(id_str) for id_str in args.local_device_ids.split(',')]
except ValueError:
print(f"Error: --local-device-ids must be a comma-separated list of integers. Received: '{args.local_device_ids}'")
exit(1)
else:
local_device_ids = []
for card_id in range(num_cards):
for chip_id in range(chips_per_card):
device_id = card_id * chips_per_card + chip_id
local_device_ids.append(device_id)
# generate local device list for local rank 0, and gather it to all ranks
local_device_list: list[dict[str, str]] = list()
if local_rank == "0":
super_pod_id = "0"
for idx in range(len(local_device_ids)):
device_id = local_device_ids[idx]
chip_id = device_id % chips_per_card
card_id = device_id // chips_per_card
if device_type == AscendDeviceType._910_93:
device_ip = get_cmd_stdout(
f"{hccn_tool_path} -i {device_id} -vnic -g | grep ipaddr"
).split(":")[1].strip()
super_device_id = get_cmd_stdout(
f"npu-smi info -t spod-info -i {card_id} -c {chip_id} | grep SDID"
).split(":")[1].strip()
super_pod_id = get_cmd_stdout(
f"npu-smi info -t spod-info -i {card_id} -c {chip_id} | grep \"Super Pod ID\""
).split(":")[1].strip()
else:
device_ip = get_cmd_stdout(
f"{hccn_tool_path} -i {device_id} -ip -g | grep ipaddr"
).split(":")[1].strip()
device_info = {
"server_id": local_host,
"device_id": str(device_id),
"device_ip": str(device_ip),
}
if device_type == AscendDeviceType._910_93:
device_info.update({
"super_pod_id": str(super_pod_id),
"super_device_id": str(super_device_id)
})
local_device_list.append(device_info)
dist.init_process_group(backend=dist.Backend.GLOO)
global_device_list = [None] * dist.get_world_size()
dist.all_gather_object(global_device_list, local_device_list)
global_device_list = [
device_info for device_list in global_device_list
for device_info in device_list # type: ignore[attr-defined]
]
cnt = 1
for device_info in global_device_list: # type: ignore[assignment]
device_info["cluster_id"] = str(cnt)
cnt += 1
assert (prefill_device_cnt + decode_device_cnt) <= len(global_device_list), \
"prefill_device_cnt + decode_device_cnt must be less than or equal to number of all devices in cluster"
ranktable = {
"version":
"1.2",
"server_count":
str(world_size),
"prefill_device_list":
global_device_list[:prefill_device_cnt],
"decode_device_list":
global_device_list[prefill_device_cnt:prefill_device_cnt +
decode_device_cnt],
"status":
"completed"
}
if local_rank == '0':
os.makedirs(os.path.dirname(args.ranktable_path), exist_ok=True)
with open(args.ranktable_path, "w") as f:
json.dump(ranktable, f, indent=4)
print("gen ranktable.json done")

89
examples/disaggregated_prefill_v1/gen_ranktable.sh
View File

@@ -1,89 +0,0 @@
#!/bin/bash
source /usr/local/Ascend/ascend-toolkit/set_env.sh
export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/opp/vendors/customize/op_api/lib/:${LD_LIBRARY_PATH}
NPUS_PER_NODE=8
while [[ $# -gt 0 ]]; do
case "$1" in
--ips)
shift
while [[ $# -gt 0 && ! "$1" == --* ]]; do
IPs+=("$1")
shift
done
;;
--npus-per-node)
shift
NPUS_PER_NODE="$1"
shift
;;
--network-card-name)
shift
NETWORK_CARD_NAME="$1"
shift
;;
--prefill-device-cnt)
shift
PREFILL_DEVICE_CNT="$1"
shift
;;
--decode-device-cnt)
shift
DECODE_DEVICE_CNT="$1"
shift
;;
--local-device-ids)
shift
LOCAL_DEVICE_IDS="$1"
shift
;;
esac
done
LOCAL_HOSTS=($(hostname -I))
LOCAL_HOST="127.0.0.1"
MASTER_ADDR=${IPs[0]}
MASTER_PORT=6657
NNODES=${#IPs[@]}
NODE_RANK="8"
for i in "${!IPs[@]}"; do
ip="${IPs[$i]}"
for local_host in "${LOCAL_HOSTS[@]}"; do
if [[ "$local_host" == "$ip" ]]; then
LOCAL_HOST=$local_host
NODE_RANK=$i
break 2
fi
done
done
if [[ $NODE_RANK == "" ]];then
echo "[Error] para \"NODE_RANK\" must be defined"
exit 1
fi
WORLD_SIZE=$(($NPUS_PER_NODE * $NNODES))
RANKSTART=`expr $NPUS_PER_NODE \* $NODE_RANK`
echo "========>param:"
echo "LOCAL_HOST": $LOCAL_HOST
echo "WORLD_SIZE: " $WORLD_SIZE
echo "RANKSTART": $RANKSTART
echo "NNODES": $NNODES
echo "NODE_RANK": $NODE_RANK
echo "==============="
if [ -n "$LOCAL_DEVICE_IDS" ]; then
OPTIONAL_SECTION=" --local-device-ids $LOCAL_DEVICE_IDS"
fi
if [[ -n "${GEN_RANKTABLE}" || ! -e ${PWD}/ranktable.json ]]; then
timeout 180s \
GLOO_SOCKET_IFNAME=$NETWORK_CARD_NAME torchrun \
--nproc_per_node 1 \
--nnodes ${NNODES} \
--node_rank ${NODE_RANK} \
--master_addr ${MASTER_ADDR} \
--master_port ${MASTER_PORT} \
gen_ranktable.py --local-host $LOCAL_HOST --prefill-device-cnt $PREFILL_DEVICE_CNT --decode-device-cnt $DECODE_DEVICE_CNT $OPTIONAL_SECTION
fi

30
examples/disaggregated_prefill_v1/run_server.sh
View File

@@ -1,30 +0,0 @@
export HCCL_IF_IP=141.61.39.117
export GLOO_SOCKET_IFNAME="enp48s3u1u1"
export TP_SOCKET_IFNAME="enp48s3u1u1"
export HCCL_SOCKET_IFNAME="enp48s3u1u1"
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=path-to-rank-table
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
vllm serve model_path \
--host 0.0.0.0 \
--port 20002 \
--tensor-parallel-size 1\
--seed 1024 \
--served-model-name dsv3 \
--max-model-len 2000 \
---max-num-batched-tokens 2000 \
--trust-remote-code \
--gpu-memory-utilization 0.9 \
--kv-transfer-config \
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
"kv_role": "kv_consumer",
"kv_parallel_size": 1,
"kv_port": "20001",
"engine_id": 0,
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_connector_v1_a3"
}' \
--additional-config \
'{"enable_graph_mode": "True"}'\

45
examples/offline_disaggregated_prefill_npu.py
View File

@@ -24,6 +24,7 @@ from multiprocessing import Event, Process
os.environ["VLLM_USE_MODELSCOPE"] = "True"
os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
def clean_up():
import gc
@@ -37,9 +38,6 @@ def clean_up():
def run_prefill(prefill_done, process_close):
# ranktable.json needs be generated using gen_ranktable.sh
# from the examples/disaggregated_prefill_v1 in the main branch.
os.environ['DISAGGREGATED_PREFILL_RANK_TABLE_PATH'] = "./ranktable.json"
os.environ["ASCEND_RT_VISIBLE_DEVICES"] = "0"
from vllm import LLM, SamplingParams
@@ -51,9 +49,22 @@ def run_prefill(prefill_done, process_close):
]
sampling_params = SamplingParams(temperature=0, top_p=0.95, max_tokens=1)
ktc = KVTransferConfig(kv_connector="LLMDataDistCMgrConnector", kv_buffer_device="npu", kv_role="kv_producer",
kv_parallel_size=1,
kv_connector_module_path="vllm_ascend.distributed.llmdatadist_c_mgr_connector")
ktc = KVTransferConfig(
kv_connector="MooncakeConnector",
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": 1,
"tp_size": 1
},
"decode": {
"dp_size": 1,
"tp_size": 1
}
})
# Set NPU memory utilization to 0.8
llm = LLM(model="deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
kv_transfer_config=ktc,
@@ -79,10 +90,6 @@ def run_prefill(prefill_done, process_close):
def run_decode(prefill_done):
os.environ['VLLM_ASCEND_LLMDD_RPC_PORT'] = '6634'
# ranktable.json needs be generated using gen_ranktable.sh
# from the examples/disaggregated_prefill_v1 module in the main branch.
os.environ['DISAGGREGATED_PREFILL_RANK_TABLE_PATH'] = "./ranktable.json"
os.environ["ASCEND_RT_VISIBLE_DEVICES"] = "1"
from vllm import LLM, SamplingParams
@@ -94,8 +101,22 @@ def run_decode(prefill_done):
]
sampling_params = SamplingParams(temperature=0, top_p=0.95)
ktc = KVTransferConfig(kv_connector="LLMDataDistCMgrConnector", kv_buffer_device="npu", kv_role="kv_consumer",
kv_parallel_size=1, kv_connector_module_path="vllm_ascend.distributed.llmdatadist_c_mgr_connector")
ktc = KVTransferConfig(
kv_connector="MooncakeConnector",
kv_role="kv_consumer",
kv_port="30100",
engine_id="1",
kv_connector_module_path="vllm_ascend.distributed.mooncake_connector",
kv_connector_extra_config={
"prefill": {
"dp_size": 1,
"tp_size": 1
},
"decode": {
"dp_size": 1,
"tp_size": 1
}
})
llm = LLM(model="deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
kv_transfer_config=ktc,

72
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8-EPLB.yaml
View File

@@ -41,13 +41,21 @@ deployment:
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
'{"kv_connector": "MooncakeConnector",
"kv_role": "kv_producer",
"kv_parallel_size": 1,
"kv_port": "20001",
"kv_port": "30000",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
@@ -71,13 +79,21 @@ deployment:
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
'{"kv_connector": "MooncakeConnector",
"kv_role": "kv_producer",
"kv_parallel_size": 1,
"kv_port": "20001",
"kv_port": "30100",
"engine_id": "1",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
@@ -102,13 +118,21 @@ deployment:
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
'{"kv_connector": "MooncakeConnector",
"kv_role": "kv_consumer",
"kv_parallel_size": 1,
"kv_port": "20001",
"kv_port": "30200",
"engine_id": "2",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'
@@ -132,13 +156,21 @@ deployment:
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
'{"kv_connector": "MooncakeConnector",
"kv_role": "kv_consumer",
"kv_parallel_size": 1,
"kv_port": "20001",
"kv_port": "30200",
"engine_id": "2",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true,"dynamic_eplb":true,"num_iterations_eplb_update":2048,"num_wait_worker_iterations":200}'

78
tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8.yaml
View File

@@ -40,13 +40,21 @@ deployment:
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
'{"kv_connector": "MooncakeConnector",
"kv_role": "kv_producer",
"kv_parallel_size": 1,
"kv_port": "20001",
"kv_port": "30000",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true}'
@@ -70,13 +78,21 @@ deployment:
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
'{"kv_connector": "MooncakeConnector",
"kv_role": "kv_producer",
"kv_parallel_size": 1,
"kv_port": "20001",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
"kv_port": "30100",
"engine_id": "1",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"torchair_graph_config":{"enabled":false,"enable_multistream_shared_expert":false},"enable_prefill_optimizations":true,"enable_weight_nz_layout":true}'
@@ -101,13 +117,21 @@ deployment:
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
'{"kv_connector": "MooncakeConnector",
"kv_role": "kv_consumer",
"kv_parallel_size": 1,
"kv_port": "20001",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
"kv_port": "30200",
"engine_id": "2",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true}'
@@ -131,13 +155,21 @@ deployment:
--gpu-memory-utilization 0.9
--speculative-config '{"num_speculative_tokens": 1, "method":"mtp"}'
--kv-transfer-config
'{"kv_connector": "LLMDataDistCMgrConnector",
"kv_buffer_device": "npu",
'{"kv_connector": "MooncakeConnector",
"kv_role": "kv_consumer",
"kv_parallel_size": 1,
"kv_port": "20001",
"engine_id": "0",
"kv_connector_module_path": "vllm_ascend.distributed.llmdatadist_c_mgr_connector"
"kv_port": "30200",
"engine_id": "2",
"kv_connector_module_path": "vllm_ascend.distributed.mooncake_connector",
"kv_connector_extra_config": {
"prefill": {
"dp_size": 2,
"tp_size": 8
},
"decode": {
"dp_size": 32,
"tp_size": 1
}
}
}'
--additional-config
'{"torchair_graph_config":{"enabled":true,"enable_multistream_mla":true,"graph_batch_sizes":[28],"use_cached_graph":true,"enable_super_kernel":false},"multistream_overlap_shared_expert":true}'

3
tests/e2e/nightly/multi_node/config/multi_node_config.py
View File

@@ -127,9 +127,6 @@ class MultiNodeConfig:
master_ip = self.master_ip
if self.disaggregated_prefill:
self.envs[
"DISAGGREGATED_PREFILL_RANK_TABLE_PATH"] = self.disaggregated_prefill.get(
"ranktable_path")
if self.cur_index < self.decode_start_index:
# For prefiller nodes, use the default master ip(index==0) as DP master
master_ip = self.master_ip

25
tests/e2e/pd_disaggreate/run_edge_case_test.sh
View File

@@ -16,17 +16,6 @@ GIT_ROOT=$(git rev-parse --show-toplevel)
# Trap the SIGINT signal (triggered by Ctrl+C)
trap 'kill $(jobs -pr)' SIGINT SIGTERM EXIT
# Gen ranktable
RANKTABLE_PATH=${GIT_ROOT}/examples/disaggregate_prefill_v1/ranktable.json
if [ -f "$RANKTABLE_PATH" ]; then
rm "$RANKTABLE_PATH"
fi
cd ${GIT_ROOT}/examples/disaggregate_prefill_v1
LOCAL_HOST=`hostname -I|awk -F " " '{print$1}'`
bash gen_ranktable.sh --ips $LOCAL_HOST --network-card-name enp189s0f0 --prefill-device-cnt 1 --decode-device-cnt 1
cd -
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH="$RANKTABLE_PATH"
# Waits for vLLM to start.
wait_for_server() {
local port=$1
@@ -69,12 +58,14 @@ run_tests_for_model() {
# Start prefill instance
PREFILL_PORT=8001
BASE_CMD="ASCEND_RT_VISIBLE_DEVICES=0 VLLM_ASCEND_LLMDD_RPC_PORT=5559 vllm serve $model_name \
BASE_CMD="ASCEND_RT_VISIBLE_DEVICES=0 vllm serve $model_name \
--port $PREFILL_PORT \
--seed 1024 \
--enforce-eager \
--disable-log-requests \
--gpu-memory-utilization 0.8 \
--kv-transfer-config '{\"kv_connector\":\"LLMDataDistCMgrConnector\",\"kv_role\":\"kv_producer\",\"kv_buffer_device\":\"npu\",\"kv_parallel_size\":\"1\",\"kv_port\":\"20001\",\"engine_id\":\"0\",\"kv_connector_module_path\":\"vllm_ascend.distributed.llmdatadist_c_mgr_connector\"}'"
--distributed-executor-backend mp \
--kv-transfer-config '{\"kv_connector\":\"MooncakeConnector\",\"kv_role\":\"kv_producer\",\"kv_port\":\"30000\",\"engine_id\":\"0\",\"kv_connector_module_path\":\"vllm_ascend.distributed.mooncake_connector\",\"kv_connector_extra_config\":{\"prefill\":{\"dp_size\":1,\"tp_size\":1},\"decode\":{\"dp_size\":1,\"tp_size\":1}}}'"
if [ -n "$model_args" ]; then
FULL_CMD="$BASE_CMD $model_args"
@@ -88,12 +79,14 @@ run_tests_for_model() {
DECODE_PORT=8002
# Build the command with or without model-specific args
BASE_CMD="ASCEND_RT_VISIBLE_DEVICES=1 VLLM_ASCEND_LLMDD_RPC_PORT=6000 vllm serve $model_name \
BASE_CMD="ASCEND_RT_VISIBLE_DEVICES=1 vllm serve $model_name \
--port $DECODE_PORT \
--seed 1024 \
--enforce-eager \
--disable-log-requests \
--gpu-memory-utilization 0.8 \
--kv-transfer-config '{\"kv_connector\":\"LLMDataDistCMgrConnector\",\"kv_role\":\"kv_consumer\",\"kv_buffer_device\":\"npu\",\"kv_parallel_size\":\"1\",\"kv_port\":\"20001\",\"engine_id\":\"0\",\"kv_connector_module_path\":\"vllm_ascend.distributed.llmdatadist_c_mgr_connector\"}'"
--distributed-executor-backend mp \
--kv-transfer-config '{\"kv_connector\":\"MooncakeConnector\",\"kv_role\":\"kv_consumer\",\"kv_port\":\"30100\",\"engine_id\":\"1\",\"kv_connector_module_path\":\"vllm_ascend.distributed.mooncake_connector\",\"kv_connector_extra_config\":{\"prefill\":{\"dp_size\":1,\"tp_size\":1},\"decode\":{\"dp_size\":1,\"tp_size\":1}}}'"
if [ -n "$model_args" ]; then
FULL_CMD="$BASE_CMD $model_args"
@@ -111,7 +104,7 @@ run_tests_for_model() {
# Build the command for the proxy server with all the hosts and ports
PROXY_PORT=8192
PROXY_CMD="python ${GIT_ROOT}/examples/disaggregate_prefill_v1/toy_proxy_server.py --port $PROXY_PORT"
PROXY_CMD="python ${GIT_ROOT}/examples/disaggregated_prefill_v1/load_balance_proxy_server_example.py --port $PROXY_PORT"
PROXY_CMD+=" --prefiller-ports ${PREFILL_PORT}"
PROXY_CMD+=" --decoder-ports ${DECODE_PORT}"
# Start the proxy server

98
tests/ut/kv_connector/test_llmdatadist_connector.py
View File

@@ -1,98 +0,0 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
import os
import types
from tests.ut.kv_connector.utils import (create_request, create_scheduler,
create_vllm_config)
from vllm_ascend.distributed.llmdatadist_c_mgr_connector import (
LLMDataDistCMgrConnectorMetadata, LLMDataDistCMgrConnectorWorker, LLMRole)
def test_basic_inferface():
"""Unit test for basic LLMDataDistCMgrConnector interface functionality."""
vllm_config = create_vllm_config()
scheduler = create_scheduler(vllm_config)
# 2 Full Blocks and 1 Half Block.
BLOCK_SIZE = vllm_config.cache_config.block_size
NUM_EXTERNAL_FULL_BLOCKS = 2
NUM_TOKENS = int(BLOCK_SIZE * (NUM_EXTERNAL_FULL_BLOCKS + 0.5))
request = create_request(request_id=1,
num_tokens=NUM_TOKENS,
do_remote_prefill=True)
request_id = request.request_id
scheduler.add_request(request)
# Remote Prefill, triggers LLMDataDistCMgrConnectorMetadata.
scheduler_output = scheduler.schedule()
kv_connector_metadata = scheduler_output.kv_connector_metadata
assert kv_connector_metadata is not None
assert isinstance(kv_connector_metadata, LLMDataDistCMgrConnectorMetadata)
assert len(kv_connector_metadata.requests) == 1
assert request_id in kv_connector_metadata.requests
req_meta = kv_connector_metadata.requests[request_id]
for block_id, block in zip(
req_meta.local_block_ids, scheduler.kv_cache_manager.coordinator.
single_type_managers[0].req_to_blocks[request_id]):
assert block_id == block.block_id
def test_read_agent_metadata():
rank_table = {
"version":
"1.2",
"server_count":
"2",
"prefill_device_list": [{
"server_id": "192.168.1.1",
"device_id": "0",
"device_ip": "10.30.0.1",
"cluster_id": "0",
}, {
"server_id": "192.168.1.1",
"device_id": "1",
"device_ip": "10.30.0.2",
"cluster_id": "1",
}, {
"server_id": "192.168.1.2",
"device_id": "0",
"device_ip": "10.30.0.3",
"cluster_id": "2",
}, {
"server_id": "192.168.1.2",
"device_id": "1",
"device_ip": "10.30.0.4",
"cluster_id": "3",
}]
}
def get_device_ip(worker_local_ip, worker_tp_rank, worker_visible_devices):
old_visible_devices = os.environ.get("ASCEND_RT_VISIBLE_DEVICES", "")
worker = types.SimpleNamespace()
worker.local_ip = worker_local_ip
worker.tp_rank = worker_tp_rank
worker.llm_datadist_role = LLMRole.PROMPT
worker.pcp_rank = 0
worker.tp_size = worker_tp_rank + 1
os.environ["ASCEND_RT_VISIBLE_DEVICES"] = worker_visible_devices
agent_metadata = LLMDataDistCMgrConnectorWorker.read_agent_metadata(
worker, rank_table)
os.environ["ASCEND_RT_VISIBLE_DEVICES"] = old_visible_devices
return agent_metadata.device_ip
assert get_device_ip("192.168.1.1", 0, "0") == "10.30.0.1"
assert get_device_ip("192.168.1.1", 0, "1") == "10.30.0.2"
assert get_device_ip("192.168.1.2", 0, "0") == "10.30.0.3"
assert get_device_ip("192.168.1.2", 0, "1") == "10.30.0.4"
assert get_device_ip("192.168.1.1", 0, "0,1") == "10.30.0.1"
assert get_device_ip("192.168.1.1", 1, "0,1") == "10.30.0.2"
assert get_device_ip("192.168.1.1", 0, "") == "10.30.0.1"
assert get_device_ip("192.168.1.1", 1, "") == "10.30.0.2"

5
tests/ut/kv_connector/utils.py
View File

@@ -78,10 +78,9 @@ def create_vllm_config(
enable_prefix_caching=True,
)
kv_transfer_config = KVTransferConfig(
kv_connector="LLMDataDistCMgrConnector",
kv_connector="MooncakeConnector",
kv_role="kv_both",
kv_connector_module_path=
"vllm_ascend.distributed.llmdatadist_c_mgr_connector")
kv_connector_module_path="vllm_ascend.distributed.mooncake_connector")
return VllmConfig(scheduler_config=scheduler_config,
model_config=model_config,
cache_config=cache_config,

5
vllm_ascend/distributed/__init__.py
View File

@@ -20,11 +20,6 @@ from vllm.distributed.kv_transfer.kv_connector.factory import \
def register_connector():
KVConnectorFactory.register_connector(
"LLMDataDistCMgrConnector",
"vllm_ascend.distributed.llmdatadist_c_mgr_connector",
"LLMDataDistCMgrConnector")
KVConnectorFactory.register_connector(
"MooncakeConnectorV1", "vllm_ascend.distributed.mooncake_connector",
"MooncakeConnector")

1095
vllm_ascend/distributed/llmdatadist_c_mgr_connector.py
View File

File diff suppressed because it is too large Load Diff

17
vllm_ascend/envs.py
View File

@@ -103,23 +103,6 @@ env_variables: Dict[str, Callable[[], Any]] = {
"VLLM_ASCEND_ENABLE_TOPK_TOPP_OPTIMIZATION":
lambda: bool(
int(os.getenv("VLLM_ASCEND_ENABLE_TOPK_TOPP_OPTIMIZATION", '1'))),
# `LLMDataDistCMgrConnector` required variable. `DISAGGREGATED_PREFILL_RANK_TABLE_PATH` is
# used for llmdatadist to build the communication topology for kv cache transfer, it is
# a required variable if `LLMDataDistCMgrConnector` is used as kv connector for disaggregated
# pd. The rank table can be generated by adopting the script `gen_ranktable.sh`
# in vllm_ascend's example folder.
"DISAGGREGATED_PREFILL_RANK_TABLE_PATH":
lambda: os.getenv("DISAGGREGATED_PREFILL_RANK_TABLE_PATH", None),
# `LLMDataDistCMgrConnector` required variable. `VLLM_ASCEND_LLMDD_RPC_IP` is used as the
# rpc communication listening ip, which will be used to receive the agent metadata from the
# remote worker.
"VLLM_ASCEND_LLMDD_RPC_IP":
lambda: os.getenv("VLLM_ASCEND_LLMDD_RPC_IP", "0.0.0.0"),
# `LLMDataDistCMgrConnector` required variable. `VLLM_ASCEND_LLMDD_RPC_PORT` is used as the
# rpc communication listening port, which will be used to receive the agent metadata from the
# remote worker.
"VLLM_ASCEND_LLMDD_RPC_PORT":
lambda: int(os.getenv("VLLM_ASCEND_LLMDD_RPC_PORT", 5557)),
# Whether to enable mla_pa for deepseek mla decode, this flag will be removed after its available torch_npu is public accessible
# and the mla_pa will be the default path of deepseek decode path.
"VLLM_ASCEND_MLA_PA":

4
vllm_ascend/worker/model_runner_v1.py
View File

@@ -3398,7 +3398,7 @@ class NPUModelRunner(LoRAModelRunnerMixin, ECConnectorModelRunnerMixin):
# init kv cache tensors
kv_cache_raw_tensors: dict[str, Union[torch.Tensor,
Optional[torch.Tensor]]] = {}
# llmdatadist need the addr of cache tensor be aligned with 2M
# prefill disaggregation need the addr of cache tensor be aligned with 2M
alignment = 2 * 1024 * 1024
for kv_cache_tensor in kv_cache_config.kv_cache_tensors:
# TODO: REFACTOR ME to sharing hybrid cache
@@ -3426,7 +3426,7 @@ class NPUModelRunner(LoRAModelRunnerMixin, ECConnectorModelRunnerMixin):
elif "attn" in layer_name and layer_name not in kv_cache_raw_tensors.keys(
):
# NOTE: We need to init k cache tensor (nope cache tensor in mla) and
# v cache tensor (rope cache tensor in mla) separately to support llmdatadist,
# v cache tensor (rope cache tensor in mla) separately to support prefill disaggregation,
# as it only support the 0-dim of kv_cache is `num_blocks`.
# For deepseek mla, we need to spilt cache tensor accrodding to the nope head dim
# and rope head dim.