ComponentsKVBM

Running KVBM in TensorRT-LLM

View as Markdown

This guide explains how to leverage KVBM (KV Block Manager) to manage KV cache and do KV offloading in TensorRT-LLM (trtllm).

To learn what KVBM is, please check here

  • Ensure that etcd and nats are running before starting.
  • KVBM only supports TensorRT-LLM’s PyTorch backend.
  • Disable partial reuse enable_partial_reuse: false in the LLM API config’s kv_connector_config to increase offloading cache hits.
  • KVBM requires TensorRT-LLM v1.1.0rc5 or newer.
  • Enabling KVBM metrics with TensorRT-LLM is still a work in progress.

Quick Start

To use KVBM in TensorRT-LLM, you can follow the steps below:

$# Start up etcd for KVBM leader/worker registration and discovery
$docker compose -f deploy/docker-compose.yml up -d
$
$# Build a dynamo TRTLLM container (KVBM is built in by default)
$./container/build.sh --framework trtllm
$
$# Launch the container
$./container/run.sh --framework trtllm -it --mount-workspace --use-nixl-gds
$
$# Configure KVBM cache tiers (choose one of the following options):
$
$# Option 1: CPU cache only (GPU -> CPU offloading)
$# 4 means 4GB of pinned CPU memory would be used
$export DYN_KVBM_CPU_CACHE_GB=4
$
$# Option 2: Both CPU and Disk cache (GPU -> CPU -> Disk tiered offloading)
$export DYN_KVBM_CPU_CACHE_GB=4
$# 8 means 8GB of disk would be used
$export DYN_KVBM_DISK_CACHE_GB=8
$
$# [Experimental] Option 3: Disk cache only (GPU -> Disk direct offloading, bypassing CPU)
$# NOTE: this option is only experimental and it might not give out the best performance.
$# NOTE: disk offload filtering is not supported when using this option.
$export DYN_KVBM_DISK_CACHE_GB=8
$
$# Note: You can also use DYN_KVBM_CPU_CACHE_OVERRIDE_NUM_BLOCKS or
$# DYN_KVBM_DISK_CACHE_OVERRIDE_NUM_BLOCKS to specify exact block counts instead of GB

When disk offloading is enabled, to extend SSD lifespan, disk offload filtering would be enabled by default. The current policy is only offloading KV blocks from CPU to disk if the blocks have frequency equal or more than 2. Frequency is determined via doubling on cache hit (init with 1) and decrement by 1 on each time decay step.

To disable disk offload filtering, set DYN_KVBM_DISABLE_DISK_OFFLOAD_FILTER to true or 1.

$# write an example LLM API config
$# Note: Disable partial reuse "enable_partial_reuse: false" in the LLM API config’s "kv_connector_config" to increase offloading cache hits.
$cat > "/tmp/kvbm_llm_api_config.yaml" <<EOF
$backend: pytorch
$cuda_graph_config: null
$kv_cache_config:
$  enable_partial_reuse: false
$  free_gpu_memory_fraction: 0.80
$kv_connector_config:
$  connector_module: kvbm.trtllm_integration.connector
$  connector_scheduler_class: DynamoKVBMConnectorLeader
$  connector_worker_class: DynamoKVBMConnectorWorker
$EOF
$
$# [DYNAMO] start dynamo frontend
$python3 -m dynamo.frontend --http-port 8000 &
$
$# [DYNAMO] To serve an LLM model with dynamo
$python3 -m dynamo.trtllm \
>  --model-path Qwen/Qwen3-0.6B \
>  --served-model-name Qwen/Qwen3-0.6B \
>  --extra-engine-args /tmp/kvbm_llm_api_config.yaml &
$
$# Make a call to LLM
$curl localhost:8000/v1/chat/completions   -H "Content-Type: application/json"   -d '{
>    "model": "Qwen/Qwen3-0.6B",
>    "messages": [
>    {
>        "role": "user",
>        "content": "In the heart of Eldoria, an ancient land of boundless magic and mysterious creatures, lies the long-forgotten city of Aeloria. Once a beacon of knowledge and power, Aeloria was buried beneath the shifting sands of time, lost to the world for centuries. You are an intrepid explorer, known for your unparalleled curiosity and courage, who has stumbled upon an ancient map hinting at ests that Aeloria holds a secret so profound that it has the potential to reshape the very fabric of reality. Your journey will take you through treacherous deserts, enchanted forests, and across perilous mountain ranges. Your Task: Character Background: Develop a detailed background for your character. Describe their motivations for seeking out Aeloria, their skills and weaknesses, and any personal connections to the ancient city or its legends. Are they driven by a quest for knowledge, a search for lost familt clue is hidden."
>    }
>    ],
>    "stream":false,
>    "max_tokens": 30
>  }'

Alternatively, can use “trtllm-serve” with KVBM by replacing the above two [DYNAMO] cmds with below:

$trtllm-serve Qwen/Qwen3-0.6B --host localhost --port 8000 --backend pytorch --extra_llm_api_options /tmp/kvbm_llm_api_config.yaml

Enable and View KVBM Metrics

Follow below steps to enable metrics collection and view via Grafana dashboard:

$# Start the basic services (etcd & natsd), along with Prometheus and Grafana
$docker compose -f deploy/docker-observability.yml up -d
$
$# Set env var DYN_KVBM_METRICS to true, when launch via dynamo
$# Optionally set DYN_KVBM_METRICS_PORT to choose the /metrics port (default: 6880).
$DYN_KVBM_METRICS=true \
>DYN_KVBM_CPU_CACHE_GB=20 \
>python3 -m dynamo.trtllm \
>  --model-path Qwen/Qwen3-0.6B \
>  --served-model-name Qwen/Qwen3-0.6B \
>  --extra-engine-args /tmp/kvbm_llm_api_config.yaml &
$
$# Optional if firewall blocks KVBM metrics ports to send prometheus metrics
$sudo ufw allow 6880/tcp

View grafana metrics via http://localhost:3000 (default login: dynamo/dynamo) and look for KVBM Dashboard

KVBM currently provides following types of metrics out of the box:

  • kvbm_matched_tokens: The number of matched tokens
  • kvbm_offload_blocks_d2h: The number of offload blocks from device to host
  • kvbm_offload_blocks_h2d: The number of offload blocks from host to disk
  • kvbm_offload_blocks_d2d: The number of offload blocks from device to disk (bypassing host memory)
  • kvbm_onboard_blocks_d2d: The number of onboard blocks from disk to device
  • kvbm_onboard_blocks_h2d: The number of onboard blocks from host to device
  • kvbm_host_cache_hit_rate: Host cache hit rate (0.0-1.0) from sliding window
  • kvbm_disk_cache_hit_rate: Disk cache hit rate (0.0-1.0) from sliding window

Troubleshooting

  1. If enabling KVBM does not show any TTFT perf gain or even perf degradation, one potential reason is not enough prefix cache hit on KVBM to reuse offloaded KV blocks. To confirm, please enable KVBM metrics as mentioned above and check the grafana dashboard Onboard Blocks - Host to Device and Onboard Blocks - Disk to Device. If observed large number of onboarded KV blocks as the example below, we can rule out this cause: Grafana Example

  2. Allocating large memory and disk storage can take some time and lead to KVBM worker initialization timeout. To avoid it, please set a longer timeout (default 1800 seconds) for leader–worker initialization.

$# 3600 means 3600 seconds timeout
$export DYN_KVBM_LEADER_WORKER_INIT_TIMEOUT_SECS=3600
  1. When offloading to disk is enabled, KVBM could fail to start up if fallocate is not supported to create the files. To bypass the issue, please use disk zerofill fallback.
$# Set to true to enable fallback behavior when disk operations fail (e.g. fallocate not available)
$export DYN_KVBM_DISK_ZEROFILL_FALLBACK=true

Benchmark KVBM

Once the model is loaded ready, follow below steps to use LMBenchmark to benchmark KVBM performance:

$git clone https://github.com/LMCache/LMBenchmark.git
$
$# Show case of running the synthetic multi-turn chat dataset.
$# We are passing model, endpoint, output file prefix and qps to the sh script.
$cd LMBenchmark/synthetic-multi-round-qa
$./long_input_short_output_run.sh \
>    "Qwen/Qwen3-0.6B" \
>    "http://localhost:8000" \
>    "benchmark_kvbm" \
>    1
$
$# Average TTFT and other perf numbers would be in the output from above cmd

More details about how to use LMBenchmark could be found here.

NOTE: if metrics are enabled as mentioned in the above section, you can observe KV offloading, and KV onboarding in the grafana dashboard.

To compare, you can remove the kv_connector_config section from the LLM API config and run trtllm-serve with the updated config as the baseline.

$cat > "/tmp/llm_api_config.yaml" <<EOF
$backend: pytorch
$cuda_graph_config: null
$kv_cache_config:
$  enable_partial_reuse: false
$  free_gpu_memory_fraction: 0.80
$EOF
$
$# Run trtllm-serve for the baseline for comparison
$trtllm-serve Qwen/Qwen3-0.6B --host localhost --port 8000 --backend pytorch --extra_llm_api_options /tmp/llm_api_config.yaml &

Developing Locally

Inside the Dynamo container, after changing KVBM related code (Rust and/or Python), to test or use it:

$cd /workspace/lib/bindings/kvbm
$uv pip install maturin[patchelf]
$maturin build --release --out /workspace/dist
$uv pip install --upgrade --force-reinstall --no-deps /workspace/dist/kvbm*.whl