### What this PR does / why we need it?
Add user guide of speculative decoding that includes n-grams, EAGLE,
MTP, and suffix.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
Signed-off-by: zhaomingyu <zhaomingyu13@h-partners.com>
### What this PR does / why we need it?
This PR introduces the initial integration of **UCM (Unified Cache
Management)** into the vllm-ascend distributed KV-cache system.
Specifically, it adds:
- A new `UCMConnector` implementation under the distributed KV-transfer
framework.
- Support for offloading KV-cache blocks to external UCM backends (DRAM
/ NFS / Localdisk), depending on UCM configuration).
- Integration with vLLM V1 KV connector interface, including metadata
handling and role registration.
**Why it is needed:**
- UCM provides a unified, high-performance storage layer for KV-cache
externalization.
- This enables vllm-ascend to support out-of-core KV-cache workloads,
improve memory efficiency, and leverage hardware-accelerated storage
paths (RDMA / NFS / hybrid modes).
- This connector is a required component to allow future work on
multi-node inference + UCM-based scaling.
---
### Does this PR introduce _any_ user-facing change?
Yes, but limited:
- A new `kv_connector=UCMConnector` option becomes available through the
configuration interface.
- When selected, vllm-ascend workers may initialize UCM and offload
KV-cache blocks externally.
- No default behaviors are changed. Users must explicitly enable this
connector.
This PR does **not** modify:
- existing APIs,
- default execution paths,
- model runner behavior,
- user workflow unless `UCMConnector` is configured.
---
### How was this patch tested?
---
### Prefix Caching Benchmark
We provide preliminary measurements for TTFT (ms) under VLLM benchmark.
Tests run on 2 * Ascend 910B3, vllm-ascend 0.11.0, Tensor Parallel size
2, with UCM (Localdisk) enabled.
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
Signed-off-by: UnifiedCacheManager <unifiedcachem@163.com>
### What this PR does / why we need it?
Correct more doc mistakes
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
Signed-off-by: lilinsiman <lilinsiman@gmail.com>
### What this PR does / why we need it?
1.In short, we renamed the existing MooncakeStoreConnector to
AscendStoreConnector and extracted the storage engine interaction logic
into a new Backend class.
Associated RFC：https://github.com/vllm-project/vllm-ascend/issues/4329
2.Fixed the issue where the number of input parameters for the connector
was incorrect, introduced in vllm 0.11.2
### Does this PR introduce _any_ user-facing change?
change MooncakeStoreConnector to AscendStoreConnector
### How was this patch tested?
- vLLM version: v0.11.2
---------
Signed-off-by: fems14 <1804143737@qq.com>
### What this PR does / why we need it?
While using the LLM Compressor quantization tool from the VLLM community
to generate quantized weights, the VLLM Ascend engine needs to be
adapted to support the compressed tensors quantization format.
1. Add AscendCompressedTensorsConfig to replace CompressedTensorsConfig
in vllm.
2. Support CompressedTensorsW8A8 static weight.
- weight: per-channel, int8, symmetric; activation: per-tensor, int8,
symmetric.
4. Support CompressedTensorsW8A8Dynamic weight.
- weight: per-channel, int8, symmetric; activation: per-token, int8,
symmetric, dynamic.
5. Modify the override_quantization_method in AscendQuantConfig.
Co-authored-by: taoqun110 taoqun@huawei.com
Co-authored-by: chenxi-hh chen464822955@163.com
- vLLM version: v0.11.2
---------
Signed-off-by: LHXuuu <scut_xlh@163.com>
Signed-off-by: chenxi-hh <chen464822955@163.com>
Signed-off-by: chenxi-hh <32731611+chenxi-hh@users.noreply.github.com>
Co-authored-by: chenxi-hh <chen464822955@163.com>
Co-authored-by: chenxi-hh <32731611+chenxi-hh@users.noreply.github.com>
### What this PR does / why we need it?
This PR adds a load-balance dp proxy server which can be used in
external DP scenario without Disaggregated-Prefill enabled. What's more,
add a doc of external dp and load-balance dp proxy server.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
See the new doc.
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
---------
Signed-off-by: whx-sjtu <2952154980@qq.com>
### What this PR does / why we need it?
This PR introduces a new model loader called Netloader, which leverages
high-bandwidth P2P direct transfer between NPU cards to achieve weight
loading. Netloader is implemented as a plugin through the newly added
'register_model_loader' function in vLLM 0.10. It facilitates the
process of weight loading by sending weights from a pre-loaded model
(server) to an empty model of a newly started instance (client). The
server operates concurrently with normal inference tasks through
sub-threads and the 'stateless_init_torch_distributed_process_group' in
vLLM. The client initiates a transfer request after verifying that the
model and partitioning method are the same as the server's, and uses
HCCL's collective communication (send/recv) to load the weights in the
order they are stored in the model.
Application Scenarios:
1. Significantly Reduces Inference Instance Startup Time By reusing the
weights of already loaded instances and performing high-speed transfers
directly between computing cards, this method reduces model loading
latency compared to traditional remote/local pull methods.
2. Reduces Network and Storage Pressure Avoids the need to repeatedly
download weight files from remote repositories, reducing the impact on
centralized storage and network traffic, thereby enhancing overall
system stability and service quality.
3. Improves Resource Utilization and Reduces Costs Accelerating the
loading process reduces reliance on redundant computing pools, allowing
computing resources to be elastically scaled and reclaimed as needed.
4. Enhances Business Continuity and High Availability In fault recovery
scenarios, new instances can quickly take over existing services,
avoiding prolonged business interruptions and improving the system's
high availability and user experience.
### Does this PR introduce _any_ user-facing change?
Netloader utilizes the existing --load-format=netloader and
--model-loader-extra-config to be activated. The
model-loader-extra-config needs to be input as a JSON string (as it is
now)
Afterwards, you can check whether the outputs for the same sentence are
consistent when the temperature is set to 0.
Signed-off-by: destinysky <kangrui10@126.com>
- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0
---------
Signed-off-by: destinysky <kangrui10@126.com>
### Motivation
Currently dynamically experts balancing would stop-the-world.
Asynchronously expert load balancing would be better without flowing
problems:
Host-bound latency:
There are many cpu operations during EPLB such as
eplb-algorithm、creating p2p ops、and log2phy expert converting would
spend long cpu time, as ~1s.
Communication latency: The transfer time would cost much in the
situation without nvlink. As the weight of an expert maybe transfer to
multiple new positions, thus N times send/recv for one expert, with
result long latency. We had tested that batch_isend_irecv cost more
100ms for 16 experts weight transmission in A2 server of ascend.
SwiftBalancer would not stop-the-world anymore, in out test on NPU 1~2ms
cost for each layer while benefit 5ms-8ms decode latency with ep_size =
64.
The following updates have been made:
1、expert distribution recording with lower cost.
2、async cpu computing for eplb algo and other python operator.
3、new eplb algo with less expert rebalancing while almost the same
effect.
### Proposed Change
We will gradually migrate the EPLB logic to the VLLM community and
implement a generalized design. Relevant RFC:
https://github.com/vllm-project/vllm/issues/22246
The overall workflow involves:
<img width="801" height="302"
alt="474430541-23b06f58-23bc-44a3-a1be-00f268aeb15c"
src="https://github.com/user-attachments/assets/1d73a459-1b23-4b0a-812a-bf0a75debfed"
/>
1. Record experts distribution during forward. We using expert_token_num
after disptach instead of topk_ids, thus we got much smaller tensor
shape to reduce cost of hbm recording and add-operator.
2. Do all-gather for experts distribution. Using all-gather instead of
all-reduce as less traffic volume.
3. Wake up eplb worker process with experts distribution when
num_iterations comes. Run eplb algorithm in eplb worker.
4. Generate p2p send/recv ops and other operator such as log2phy would
cost long cpu time.
5. Lanch ibatch_send_recv in async_stream before forward.
6. After forward, wait for the ibatch_send_recv finish, then do uapte
expert map and expert weights.
### Co-author
Co-authored-by: raindaywhu raindaywhu@raindaywhu@ 163.con
Co-authored-by: njuyuan yuanjl19@smail.nju.edu.cn
Co-authored-by: qmkakaxi wjh1594260677@qq.com
Co-authored-by: Skywalker-EP 173723846@qq.com
- vLLM version: v0.10.2
- vLLM main:
567939953b
---------
Signed-off-by: offline0806 <z00858301@china.huawei.com>
Co-authored-by: offline0806 <z00858301@china.huawei.com>
Add user doc index to make the user guide more clear
- vLLM version: v0.9.1
- vLLM main:
49e8c7ea25
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
