RegisterKvCache

Applicable Products

Product	Supported (√/x)
Ascend 950PR/Ascend 950DT	√
Atlas A3 training products/Atlas A3 inference products	√
Atlas A2 training products/Atlas A2 inference products	√
Atlas 200I/500 A2 inference products	x
Atlas inference products	x
Atlas training products	x

Note: For Atlas A2 training products/Atlas A2 inference products, only the Atlas 800I A2 inference server and A200I A2 Box heterogeneous subrack are supported.

Function Description

Registers the local KV cache memory.

Prototype

Status RegisterKvCache(const CacheDesc &cache_desc,
                       const std::vector<uint64_t> &addrs,
                       const RegisterCfg &cfg,
                       int64_t &cache_id);

Parameters

Parameter	Input/Output	Description
cache_desc	Input	Description of the local cache.
addrs	Input	Addresses of the local cache. The number of addresses cannot exceed 240.
cfg	Input	Reserved.
cache_id	Output	ID of the registered cache. It can be used to construct a cache reference when KV transmission APIs are called.

Returns

LLM_SUCCESS: Success.
Other values: Failure.

Constraints

This API must be called after the Initialize API is called to complete initialization.

If the HDK version is earlier than 25.5.0, a maximum of 20 GB host memory can be registered. If the HDK version is 25.5.0 or later, a maximum of 1 TB host memory can be registered. Larger registered memory results in higher OS memory usage. This restriction applies to the following products:

Atlas A2 training product/Atlas A2 inference product
Atlas A3 training product/Atlas A3 inference product

Parent topic: LLM-DataDist APIs