README.md

metadata

license: apache-2.0
task_categories:
  - tabular-regression
language:
  - en
tags:
  - llm-inference
  - benchmarking
  - gpu-profiling
  - vllm
  - sglang
  - agentic-workloads
size_categories:
  - 1K<n<10K
pretty_name: AgentPerfBench
version: '1.0'
configs:
  - config_name: trace_replay
    data_files:
      - split: summary
        path: trace_replay/summary.parquet
  - config_name: distributional
    data_files:
      - split: summary
        path: distributional/summary.parquet
  - config_name: mse_validation
    data_files:
      - split: summary
        path: mse_validation/summary.parquet
dataset_info:
  - config_name: trace_replay
    features:
      - name: run_id
        dtype: string
      - name: model
        dtype: string
      - name: model_family
        dtype: string
      - name: hardware
        dtype: string
      - name: engine
        dtype: string
      - name: tensor_parallelism
        dtype: int64
      - name: profile
        dtype: string
      - name: concurrency
        dtype: int64
      - name: num_requests
        dtype: int64
      - name: duration_s
        dtype: float64
      - name: successful_requests
        dtype: int64
      - name: failed_requests
        dtype: int64
      - name: request_throughput
        dtype: float64
      - name: input_token_throughput
        dtype: float64
      - name: output_token_throughput
        dtype: float64
      - name: total_token_throughput
        dtype: float64
      - name: mean_ttft_ms
        dtype: float64
      - name: median_ttft_ms
        dtype: float64
      - name: p90_ttft_ms
        dtype: float64
      - name: p99_ttft_ms
        dtype: float64
      - name: mean_tpot_ms
        dtype: float64
      - name: median_tpot_ms
        dtype: float64
      - name: p90_tpot_ms
        dtype: float64
      - name: p99_tpot_ms
        dtype: float64
      - name: mean_itl_ms
        dtype: float64
      - name: median_itl_ms
        dtype: float64
      - name: p90_itl_ms
        dtype: float64
      - name: p99_itl_ms
        dtype: float64
      - name: mean_e2el_ms
        dtype: float64
      - name: median_e2el_ms
        dtype: float64
      - name: p90_e2el_ms
        dtype: float64
      - name: p99_e2el_ms
        dtype: float64
    splits:
      - name: summary
        num_rows: 3147
        num_bytes: 694254
  - config_name: distributional
    features:
      - name: run_id
        dtype: string
      - name: model
        dtype: string
      - name: model_family
        dtype: string
      - name: hardware
        dtype: string
      - name: engine
        dtype: string
      - name: tensor_parallelism
        dtype: int64
      - name: profile
        dtype: string
      - name: concurrency
        dtype: int64
      - name: num_requests
        dtype: int64
      - name: duration_s
        dtype: float64
      - name: successful_requests
        dtype: int64
      - name: failed_requests
        dtype: int64
      - name: request_throughput
        dtype: float64
      - name: input_token_throughput
        dtype: float64
      - name: output_token_throughput
        dtype: float64
      - name: total_token_throughput
        dtype: float64
      - name: mean_ttft_ms
        dtype: float64
      - name: median_ttft_ms
        dtype: float64
      - name: p90_ttft_ms
        dtype: float64
      - name: p99_ttft_ms
        dtype: float64
      - name: mean_tpot_ms
        dtype: float64
      - name: median_tpot_ms
        dtype: float64
      - name: p90_tpot_ms
        dtype: float64
      - name: p99_tpot_ms
        dtype: float64
      - name: mean_itl_ms
        dtype: float64
      - name: median_itl_ms
        dtype: float64
      - name: p90_itl_ms
        dtype: float64
      - name: p99_itl_ms
        dtype: float64
      - name: mean_e2el_ms
        dtype: float64
      - name: median_e2el_ms
        dtype: float64
      - name: p90_e2el_ms
        dtype: float64
      - name: p99_e2el_ms
        dtype: float64
    splits:
      - name: summary
        num_rows: 245
        num_bytes: 70836
  - config_name: mse_validation
    features:
      - name: validation_id
        dtype: string
      - name: tier
        dtype: string
      - name: stage
        dtype: string
      - name: stage_label
        dtype: string
      - name: comparison_group
        dtype: string
      - name: raw_json_path
        dtype: string
      - name: per_turn_json_path
        dtype: string
      - name: replay_kind
        dtype: string
      - name: workload
        dtype: string
      - name: profile
        dtype: string
      - name: model
        dtype: string
      - name: hardware
        dtype: string
      - name: engine
        dtype: string
      - name: tensor_parallelism
        dtype: int64
      - name: concurrency
        dtype: int64
      - name: sessions
        dtype: int64
      - name: source_locked
        dtype: bool
      - name: prefix_caching_state
        dtype: string
      - name: chunked_prefill
        dtype: string
      - name: prefix_aware_synthetic
        dtype: bool
      - name: shared_prefix_tokens
        dtype: int64
      - name: shared_prefix_block_aligned
        dtype: bool
      - name: synthetic_filler_style
        dtype: string
      - name: synthetic_target_chars_per_token
        dtype: float64
      - name: max_model_len
        dtype: int64
      - name: success_rate
        dtype: float64
      - name: num_requests
        dtype: int64
      - name: duration_s
        dtype: float64
      - name: successful_requests
        dtype: int64
      - name: failed_requests
        dtype: int64
      - name: request_throughput
        dtype: float64
      - name: input_token_throughput
        dtype: float64
      - name: output_token_throughput
        dtype: float64
      - name: total_token_throughput
        dtype: float64
      - name: mean_ttft_ms
        dtype: float64
      - name: median_ttft_ms
        dtype: float64
      - name: p90_ttft_ms
        dtype: float64
      - name: p99_ttft_ms
        dtype: float64
      - name: mean_tpot_ms
        dtype: float64
      - name: median_tpot_ms
        dtype: float64
      - name: p90_tpot_ms
        dtype: float64
      - name: p99_tpot_ms
        dtype: float64
      - name: mean_itl_ms
        dtype: float64
      - name: median_itl_ms
        dtype: float64
      - name: p90_itl_ms
        dtype: float64
      - name: p99_itl_ms
        dtype: float64
      - name: mean_e2el_ms
        dtype: float64
      - name: median_e2el_ms
        dtype: float64
      - name: p90_e2el_ms
        dtype: float64
      - name: p99_e2el_ms
        dtype: float64
    splits:
      - name: summary
        num_rows: 28
        num_bytes: 40940

AgentPerfBench

LLM inference benchmark: 3,392 main sweep runs measuring TTFT, TPOT, ITL, and throughput across 9 models, up to 14 GPU configurations, and 2 serving engines (vLLM 0.19.0, SGLang 0.5.9). All models served in BF16 except gpt-oss, which uses mxfp4 for projection weights. The dataset also includes 28 curated MSE validation rows for the distributional synthetic replay generator.

Dataset configurations

The dataset provides three configurations. trace_replay replays exact input/output sequences from recorded agent sessions. distributional samples from statistical distributions fitted to those same workloads, trading fidelity for faster sweeps across the hardware matrix. mse_validation contains paired synthetic-vs-real validation runs and ablations for the final APC-aware synthetic generator.

trace_replay (3,147 rows)

Replays exact ISL/OSL sequences from recorded agent sessions (SWE-Bench, TerminalBench, OSWorld, ShareGPT). Covers 77 unique (model, hardware, engine) combinations across 17 profiles and 6 concurrency levels. The full 5-dimensional matrix is 12.2% filled; not all models run on all hardware.

17 profiles: chat-medium, chat-multiturn-long, chat-multiturn-medium, chat-multiturn-short, chat-short, chat-singleturn, coding-singleturn, decode-heavy, osworld-multiturn-long, osworld-multiturn-medium, osworld-multiturn-short, prefill-heavy, random-1k, swebench-multiturn-medium, swebench-multiturn-short, terminalbench-multiturn-medium, terminalbench-multiturn-short

distributional (245 rows)

Samples ISL/OSL from lognormal distributions fitted to real workload statistics. Covers 42 unique (model, hardware, engine) combinations across 6 profiles and 7 concurrency levels (3.0% matrix fill). gpt-oss-120b, 3090x8, and A100-40GBx8 are excluded from this configuration.

6 profiles: chat-multiturn, chat-singleturn, coding-singleturn, osworld-multiturn, swebench-multiturn, terminalbench-multiturn

mse_validation (28 rows)

Curated H100 / Llama-3.1-8B / vLLM validation artifacts for the distributional synthetic replay generator. The main rows keep paired MSE distributional replay and real trace replay runs with success rate at least 75%; supplementary rows preserve no-replacement and high-concurrency debug runs.

The headline SWE-bench C=5 source-locked cascade is:

Condition	Turn 10-19 E2EL delta
APC on, English filler, source-locked	+45.5%
APC on, code-morph filler, no shared prefix	+31.1%
APC off, code-morph filler, no shared prefix	+11.7%
APC on, code-morph filler, 1024-token shared prefix	-3.2%

See mse_validation/README.md, mse_validation/manifest.csv, and mse_validation/fidelity_deltas.csv for the filtering rule and raw JSON pointers.

Concurrency filtering

The benchmark harness capped actual concurrent connections at the session pool size. Rows where declared concurrency exceeded the pool were excluded:

• trace_replay: concurrency > 100 removed (session pool was 100). Remaining values: {1, 5, 10, 20, 40, 80}.
• distributional (pre-fix): concurrency > 10 removed (session pool was 10). Post-fix data has no cap. Remaining values: {1, 5, 10, 40, 80, 200, 320}.

Config	Rows
trace_replay	3,147
distributional	245
mse_validation	28
Total	3,420

Failed requests

Some runs produce request failures, typically at high concurrency where the engine hits memory or timeout limits. 30.8% of trace_replay rows and 42% of distributional rows have failed_requests > 0. Summary metrics (TTFT, TPOT, throughput) are computed from successful requests only.

Coverage

Hardware

All benchmarks collected on PyTorch 2.10.0, CUDA 12.8.

GPU	VRAM	HBM bandwidth	Peak half-precision TFLOPS
NVIDIA H100 SXM	80 GB	3.35 TB/s	989
NVIDIA A100 SXM4	40 GB	1.56 TB/s	312
NVIDIA RTX 3090	24 GB	936 GB/s	71
NVIDIA RTX 2080 Ti	11 GB	616 GB/s	27

Multi-GPU configurations: 1, 2, 4, or 8 GPUs with tensor parallelism. TP degree depends on model size and available GPUs.

Models

All models served in BF16 unless noted.

Model	Family	Parameters	Architecture	Notes
Llama-3.1-8B	Llama	8B	Dense
Llama-3.1-70B	Llama	70B	Dense
Llama-3.3-70B	Llama	70B	Dense
Qwen2.5-72B	Qwen	72B	Dense
Qwen3.5-9B	Qwen	9B	Dense
Qwen3.5-27B	Qwen	27B	Dense
Mixtral-8x7B	Mixtral	46.7B (12.9B active)	MoE
gpt-oss-20b	GPT-OSS	21B (3.6B active)	MoE	mxfp4 projections
gpt-oss-120b	GPT-OSS	117B (5.1B active)	MoE	mxfp4 projections

Model names in this table match the model column in the parquet files.

Engines

• vLLM 0.19.0
• SGLang 0.5.9

Schema

Each row in summary.parquet (both configs):

Column	Type	Description
run_id	string	Deterministic hash of run parameters
model	string	Model short name
model_family	string	Model family (llama, qwen, gpt-oss, mixtral)
hardware	string	GPU configuration (e.g., H100x4)
engine	string	Serving engine (vllm, sglang)
tensor_parallelism	int	TP degree
profile	string	Workload profile name
concurrency	int	Concurrent request level
num_requests	int	Total requests in run
duration_s	float	Total run duration
successful_requests	int	Completed requests
failed_requests	int	Failed requests
request_throughput	float	Requests/second
input_token_throughput	float	Input tokens/second
output_token_throughput	float	Output tokens/second
total_token_throughput	float	Total tokens/second
mean/median/p90/p99_ttft_ms	float	Time to first token
mean/median/p90/p99_tpot_ms	float	Time per output token
mean/median/p90/p99_itl_ms	float	Inter-token latency
mean/median/p90/p99_e2el_ms	float	End-to-end latency

Loading

from datasets import load_dataset

ds = load_dataset("agent-perf-bench/AgentPerfBench", "trace_replay")
# or "distributional" / "mse_validation"

Benchmark methodology

• Closed-loop concurrency with semaphore control.
• Concurrency levels: {1, 5, 10, 20, 40, 80} (trace_replay), {1, 5, 10, 40, 80, 200, 320} (distributional).
• 3-request warmup before each configuration.
• Metrics: TTFT, TPOT, ITL, E2EL, request throughput, token throughput.
• Summary statistics: mean, median, p90, p99.
• Collection period: March 2026 onwards.
• PyTorch 2.10.0, CUDA 12.8 on all machines. All models served in BF16 (gpt-oss: mxfp4 projection weights).

Future releases

• Full per-request and multi-turn granularity data for the main sweep (pending raw JSON availability from collection infrastructure). Curated raw JSONs are included for mse_validation.
• Per-kernel CUDA roofline profiles (PyTorch profiler, 2-layer forward passes, batch sizes 1/4/8/32/64).
• This is version 1.0. Updates will be tagged with semantic versions.

Intended uses

• Inference engine comparison under controlled conditions.
• Capacity planning for LLM deployments.
• TTFT scaling with context length in multi-turn sessions.

Limitations

• Results are specific to tested hardware and software versions (vLLM 0.19.0, SGLang 0.5.9, PyTorch 2.10.0, CUDA 12.8).
• Distributional profiles approximate but do not replicate production traffic patterns.
• No consumer GPUs beyond RTX 3090; no non-NVIDIA accelerators.
• Closed-loop concurrency only; no open-loop (Poisson) arrivals.
• The model-hardware-concurrency matrix is sparse (12.2% fill for trace_replay, 3.0% for distributional). Not all model-hardware combinations are represented.
• No model quality metrics. This is a systems benchmark.

Ethical considerations

No PII. Trace-replay profiles derive from open benchmarks (SWE-Bench MIT, TerminalBench, OSWorld). Synthetic profiles use random tokens.

License

Benchmark data released under Apache-2.0. Source datasets retain their original licenses.

Source datasets

• SWE-Bench (MIT)
• TerminalBench
• ShareGPT (Aeala/ShareGPT_Vicuna_unfiltered)
• OSWorld

Citation

@inproceedings{agentperfbench2026,
  title={AgentPerfBench: A Benchmarking and Evaluation Suite for Inference Performance of Agentic LLMs},
  author={Anonymous},
  booktitle={NeurIPS 2026 Evaluations and Datasets Track},
  year={2026}
}