Pith. sign in

REVIEW 3 major objections 5 minor 18 references

Smoothing the Ramp, Not the Peak: Scheduling-Induced Power Dynamics of LLM Inference and Their Grid-Scale Consequences

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Chunked prefill scheduling cuts GPU power ramp rates by about a third at heavy load while leaving peak power unchanged, and translates into roughly one-fifth less fast-ramping reserve capacity for the grid.

desk verdict The single-GPU measurement is solid and worth knowing about, but the headline reserve-reduction number is a wide, independence-assuming extrapolation that the paper itself flags. read the letter →

arxiv 2608.01250 v1 pith:YI53UIEN submitted 2026-08-02 eess.SY cs.SY

Pan Li , Yize Chen , Xia Miao , Dai Wang This is my paper · ORCID
classification eess.SYcs.SY
keywords LLMinferenceservingdatacenterpowerdemandrampratechunkedprefillschedulingregulationreserveprocurementcoincidencefactorGPUtracesdemand-sidemanagement
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper establishes that chunked prefill scheduling—a technique already used in production LLM serving to protect tail latency—also acts as a no-cost demand-shaping lever for GPU power draw. Measured power traces show that splitting a long prompt's prefill into many small scheduler steps lowers the mean ramp rate of power draw by about 35% at heavy load (roughly 46 to 30 W/s) while leaving peak power essentially unchanged (differences no more than 1.5%). The benefit is not fixed: it grows with server saturation along two independent axes, from 7.0% mean-ramp reduction at light concurrency to 34.6% at heavy concurrency, and from statistically flat to 42.6% as long-prompt ('whale') request load rises. At fleet scale, a model-free bootstrap over real measured traces gives an estimated 20.3–22.7% reduction in the fast-ramping reserve capacity a grid operator would need to procure, across 95% to 99.9% reliability levels. This matters because ramp rate, not peak size, is what interconnection agreements and reserve procurement commonly regulate.

What carries the argument

The central object is chunked prefill: dividing a long prompt's prefill pass into fixed token-budget chunks (for example 512 tokens) that are executed across many scheduler iterations rather than one dominant, latency-blocking iteration, and interleaved with decode steps from concurrent short requests. This mechanism stretches the same total prefill work over more wall-clock time—measured energy per output token stays nearly constant (+0.0% at a 2048-token budget, +2.0% at 512)—so the GPU spends a larger fraction of time near its power ceiling with fewer and shallower excursions, lowering the mean instantaneous ramp rate while leaving peak power essentially fixed. The fleet-scale translation uses a second machinery: a model-free, two-level bootstrap that resamples real 50 ms power traces to synthesize 10,000 virtual servers with random phase shifts, computes the realized peak ramp of the aggregate demand, and solves the chance-constrained reserve-procurement problem from those empirical quantiles.

What would settle it

Instrument a real multi-GPU serving fleet at 50 ms resolution under production traffic, alternating chunked and unchunked scheduling on the same workload, and compare the 99th percentile of aggregate ramp rate. If chunked prefill does not show a materially lower aggregate ramp than unchunked scheduling at comparable utilization, or if the observed aggregate ramp exceeds the independent-bootstrap prediction once correlation is accounted for, the central claim would be refuted.

Watch

Extended reading notes

Core claim

Contrary to the intuition that slicing a long prompt into smaller steps should flatten its power spike, chunked prefill leaves peak power near the GPU ceiling in all configurations (463.9 W mono versus 470.9 W at a 512-token budget, within 1.5%) while lowering the mean instantaneous ramp rate |dP/dt| from about 45.6 to 29.8 W/s. The mechanism is temporal interleaving: each whale prompt's chunks are spread across more scheduler iterations and interleaved with decode work from concurrent short requests, replacing frequent deep power drops with longer near-ceiling plateaus. The ramp-rate reduction is not a fixed policy property but grows monotonically with saturation—from 7.0% at concurrency 6 to 34.6% at concurrency 20, and from roughly flat at low whale incidence to 42.6% at high whale fraction and size. Translated through a model-free bootstrap of 10,000 virtual servers built from 100 measured traces, chunked prefill requires an estimated 20.3–22.7% less fast-ramping reserve capacity than unchunked scheduling at a representative heavily-loaded operating point.

Load-bearing premise

The fleet-scale reserve numbers assume that many GPUs ramp independently, so their power spikes mostly cancel; if real data centers synchronize their ramps through shared scheduler ticks or coordinated batch triggers, the estimated 20% reserve saving could erode.

Editorial extensions

If this is right

  • Operators of LLM-serving facilities can reduce fast-ramping reserve requirements by roughly one-fifth without new hardware, by relying on the chunked-prefill scheduling already deployed by default in production serving systems.
  • The ramp-shaping benefit is largest at exactly the heavy-load operating points where data centers run hottest and grid stress is most salient.
  • Facilities serving retrieval-augmented or agentic traffic, which inject many long prompts, stand to gain more from chunking than ordinary chat workloads.
  • Peak power does not move under chunking, so this lever should be used for ramp management, not for shaving demand peaks.
  • The reserve saving and the whale-request time-to-first-token cost arrive together, so any aggressive chunk budget should be chosen with that latency tradeoff in view.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the mechanism transfers from the consumer-class GPU measured here to datacenter-class accelerators, as the paper expects but does not measure, the reserve saving could generalize to large fleets; a multi-GPU trace study on such hardware would settle it.
  • The independent-arrival bootstrap is an optimistic baseline: real fleets with shared scheduler ticks or coordinated batch triggers could see aggregate ramps closer to the appendix's correlated-arrival threshold, so measuring site-level correlation is the natural next test.
  • The appendix's robustness checks suggest the direction holds under Poisson arrivals, Pareto-tailed whale sizes, tensor parallelism, and sparse mixture-of-experts models, but the effect size varies; a production fleet measurement could turn these directional checks into a quantitative operating curve.
  • Because the headline reserve number is calibrated on one GPU class and a 100-trial trace library, operators should recalibrate on their own measured traces before treating 20% as a site-specific saving.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. This paper uses controlled single-GPU measurements of LLM inference to characterize the power-shape effects of chunked prefill scheduling. Replaying identical request sequences across mono/16384, 2048, and 512 token chunk budgets with locked GPU clocks and 50 ms power sampling, it finds that mean ramp rate falls from about 46 to 30 W/s (roughly 35%) at concurrency 20, while peak power moves by at most 1.5%, and that near-ceiling duty cycle rises. It reports that this ramp reduction increases with saturation along concurrency (7.0% to 34.6%) and whale-request load (up to 42.6%). It then constructs a fleet-scale aggregate from N=10,000 independent phase-shifted resamples of a 100-trial trace library and solves a chance-constrained reserve procurement problem, reporting a 20.3–22.7% reduction in fast-ramping reserve requirements across 95–99.9% reliability. Appendices document robustness checks, a parametric coincidence-factor model, and explicit caveats about correlated arrivals and heavy-tailed ramps.

Significance. The single-GPU measurement program is a definite strength: identical request sequences are replayed across configurations, GPU clocks are locked, power is sampled at 50 ms, ten trials per arm are reported, and energy conservation is checked directly. If the fleet-scale extrapolation can be put on a more robust footing, the paper would provide the power-systems community with a concrete, no-cost demand-shaping insight and a useful application of coincidence-factor formalism to measured LLM traces. The saturation-dependence and reserve-procurement claims are, however, currently supported by point estimates and by an independence assumption that the manuscript itself flags in Appendix A-A.

major comments (3)
  1. [Section VI; Appendix A-A] The fleet-scale reserve reduction is built from N=10,000 independent, randomly phase-shifted, approximately 30-second windows resampled from a 100-trial single-GPU library and summed as D(t)=sum_i P_i(t). Real fleets can exhibit correlated ramp events through shared scheduler ticks, coordinated batch triggers, viral long prompts, or common GPU power-management behavior. The manuscript itself states in Appendix A-A that any reserve number in this paper, main body included, is best read as an independent-arrival baseline, and Appendix C-A shows that even s=0.10 pushes the level coincidence factor near its ceiling; for ramp rate, the appendix reports smooth sensitivity to s but gives no quantitative estimate of the effect on the Section VI headline number. This makes the 20.3–22.7% reduction an independent-arrival baseline rather than a fleet-realistic estimate, and the paper should either add a correlation sensitivity analysis to the Section VI bootstrap or explicitly reword the grid-scale claim as conditional on independent arrivals.
  2. [Section VI, Table IV] The reported reserve reductions carry a nested-bootstrap standard deviation of 11.9 to 12.3 percentage points against mean reductions of 20.3 to 22.7 percentage points. Under a standard error interpretation, a one-standard-deviation band reaches values near 8%, and a 95% band would include reductions close to zero, so the statement that chunking requires an estimated 20.3–22.7% less ramping reserve is a point estimate with very wide spread. Please report confidence intervals for R*(epsilon) and for the percentage reduction, and clarify how the 30 outer bootstrap replicates represent between-trial uncertainty; without this, the operational significance of the grid-scale number is not yet established.
  3. [Section V, Tables II and III] The saturation-dependence claim is a central contribution, but Tables II and III report only point estimates of mean and p99 ramp reduction, with no standard deviations, confidence intervals, or statistical tests. The text asserts that the effect is statistically flat at low whale incidence (Table III), yet no test is shown, and the claimed monotonic increase from 7.0% to 34.6% across concurrency levels is not accompanied by any measure of uncertainty. Please report per-arm variability and provide at least confidence intervals, or a formal test, for monotonicity along both the concurrency and whale-load axes.
minor comments (5)
  1. [Section IV-B] The sentence 'each of independent trials per configuration discards an initial warmup burst' appears to be missing the trial count; please state the number of trials and the duration of the discarded warmup.
  2. [Section III, Eq. (1)] Equation (1) and the surrounding text define R in MW/min, but the measured traces are sampled in W/s; please specify the ramp interval used for |D'(t)| and the unit conversion applied when evaluating the condition P(|D'(t)|>R).
  3. [Table I] The 'near-ceiling' threshold used for the duty-cycle columns is not defined in the caption; state the threshold (for example, a fraction of the measured peak power) and how it was chosen.
  4. [Section VI, Method] The phase-shift and window-sampling procedure for the bootstrap is underspecified; state the number of phase offsets, their distribution, and whether windows drawn from the same trial may overlap.
  5. [Figure 2] The annotation that the decode segment is sub-pixel makes the central schematic difficult to verify; please provide a zoomed inset or an explicit numeric scale for the decode component.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the single-GPU finding is a direct measurement, the saturation dependence is an empirical sweep, and the fleet-scale reserve estimate is a bootstrap functional of measured traces rather than a fitted parameter renamed as a prediction.

full rationale

The paper's central claims are empirical measurements, not derivations from fitted parameters. Table I reports directly measured mean ramp rates (mono 45.6±2.6 W/s vs. chunk=512 29.8±3.2 W/s) and peak-power differences of at most 1.5%; Tables II and III report direct concurrency and whale-load sweeps. Section VI's fleet-scale reserve number is produced by resampling a 100-trial library of real traces into N=10,000 phase-shifted virtual servers and reading off the (1−ε)-quantile of the aggregate ramp; the reduction is an emergent statistic of those traces, not a quantity fitted to equal the single-GPU ramp reduction. The paper itself flags the main extrapolation assumption: "any reserve number in this paper, main body included, is best read as an independent-arrival baseline" (Appendix A-A). That is an untested modeling assumption, hence a correctness or robustness risk, but not circularity. The secondary OU model (Appendix B) calibrates per-policy arrival timing and mean-reversion rate from the same traces, and Table IX notes the mean ramp "matches by construction," but the model's reserve reduction (Table X) is an aggregate tail quantity that is not equal to that calibration target, and the headline result deliberately uses the model-free bootstrap instead. No load-bearing self-citations or imported uniqueness theorems appear; references to chunked-prefill prior work are background, not premises that force the result.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

The central single-GPU finding is a direct measurement and does not hide fitted parameters in the headline claim. The free-parameter burden appears mainly in the fleet extrapolation: chosen concurrency, the exaggerated whale workload, bootstrap design choices, and the Gumbel tail fit. The secondary parametric coincidence-factor model calibrates per-policy timing to the very traces it then evaluates, but the main body uses a model-free bootstrap instead. The most load-bearing domain assumption is that independent draws from a single-GPU library represent a 10,000-server fleet with no correlation. No new physical entities are introduced.

free parameters (6)
  • Concurrency for headline grid result = 20
    The 20.3 to 22.7% reserve reduction is computed only at this heavily loaded operating point. Lighter concurrency shows smaller single-GPU ramp reductions, so the grid number is conditional on this chosen load level.
  • Whale workload mixture = 15% whale fraction, 18k-50k character whales
    This bimodal mixture is deliberately exaggerated relative to real ShareGPT-derived traffic to expose the mechanism. The magnitude of the ramp-reduction and reserve effects depends on this choice.
  • Bootstrap design parameters = N=10,000 servers, 100 traces/policy, ~30 s windows, 30x30 nested bootstrap replicates
    These Monte Carlo choices define the fleet-scale ramp distribution. They are not fitted to data, but the wide reported uncertainty (about plus or minus 12 percentage points) shows sensitivity to the resampling design.
  • Per-policy ON/OFF timing parameters in Appendix B = mono: p=0.561, mean duration 1.63 s; chunk=512: p=0.696, mean duration 3.68 s
    Calibrated separately to each policy's own measured traces for the parametric coincidence-factor model. This model is secondary and the paper itself flags its Gaussian tail as lighter than real hardware.
  • OU/first-order-lag time constant tau in Appendix B = Fit per policy to match measured mean ramp rate
    Used to make the parametric model reproduce the measured single-server mean ramp. This is a fitted parameter in the secondary model, not in the headline model-free bootstrap.
  • Gumbel tail fit for the 99.9% reliability level = Location and scale fitted to the 95% and 99% empirical reserve quantiles
    The 99.9% reserve reduction is extrapolated beyond the measured 100-trial library using a Gumbel tail fit. The paper calls this directional rather than fully supported.
assumptions (7)
  • domain assumption Independent-server aggregation in the fleet bootstrap
    Section VI builds D(t) as the sum of 10,000 virtual servers, each independently drawing a randomly phase-shifted ~30 s window from a 100-trial single-GPU library. This assumes no cross-server correlation in ramp events, which the appendix shows can be consequential for coincidence factors.
  • domain assumption RTX 4090 power behavior represents datacenter-class GPUs
    Section IV states the qualitative mechanism should carry over because it follows from scheduler slicing, but ramp-rate and duty-cycle magnitudes are hardware-specific. The fleet reserve numbers are nevertheless produced from these consumer-GPU traces.
  • domain assumption The chosen bimodal workload represents production LLM serving
    Section IV uses a deliberately exaggerated 15% whale fraction with 18k-50k character prompts, grounded in ShareGPT and agentic traffic references. The absolute effect sizes depend on this mixture.
  • domain assumption Chunked prefill is already the default in production vLLM/Sarathi-Serve
    The paper's 'no additional cost' framing relies on the claim that chunked prefill is already deployed by default. This is supported by citations to Sarathi-Serve and standard practice, but is a background assumption about production systems.
  • standard math Functional central limit theorem and Gaussian process max scaling
    Appendix C uses FCLT and the scale-invariance of Gaussian process maxima to derive CF_ramp(N) ~ sigma*M/(R*sqrt(N)). This is standard probability theory and is used for the secondary parametric model.
  • standard math Campbell's theorem for the random telegraph process
    Appendix C derives the single-server ramp-rate variance in Eq. (6) by passing the ON/OFF power target through a first-order lag and applying Campbell's theorem. Standard stochastic-process tooling.
  • standard math Gumbel extreme-value distribution for tail extrapolation
    Section VI and Appendix B extrapolate the 99.9% reserve quantile with a Gumbel fit to the 95% and 99% empirical quantiles. This is a standard but assumption-laden statistical tool, and the paper acknowledges the real tail is heavier.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Smoothing the Ramp, Not the Peak: Scheduling-Induced Power Dynamics of LLM Inference and Their Grid-Scale Consequences." pith.science (2026). https://pith.science/paper/YI53UIEN

@misc{pith2026260801250,
  author       = {Pith},
  title        = {Pith review of: Smoothing the Ramp, Not the Peak: Scheduling-Induced Power Dynamics of LLM Inference and Their Grid-Scale Consequences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YI53UIEN}},
  note         = {Machine review of arXiv:2608.01250}
}
read the original abstract

Large language model (LLM) inference serving is a fast-growing electricity load whose power dynamics remain uncharacterized from a grid-planning perspective. Using real, measured GPU power traces, we show that chunked prefill scheduling, a latency-motivated technique already deployed by default in production LLM serving, is a controllable knob that regulates power ramp rate without touching peak power. Contrary to the intuitive hypothesis that splitting a long prompt's computation into smaller steps should flatten its power spike, peak power stays relatively the same while mean ramp rate falls substantially. Critically, this ramp-rate benefit is not a fixed property of the policy: it grows monotonically with system saturation, and we confirm this along two independent axes: concurrency (7.0% at light load to 34.6% at heavy load, mean-ramp reduction) and long-prompt ("whale") request load (from statistically flat at low whale incidence to 42.6% at high whale fraction/size). We translate this single-GPU mechanism into an operational grid quantity, regulation-reserve procurement, posed and solved as a chance-constrained problem using a model-free bootstrap directly resampling real measured power traces. At a representative operating point, this translates to an estimated 20.3-22.7% reduction in the fast-ramping reserve capacity a grid operator would need to provision, across reliability levels from 95% to 99.9%. Together, these results give grid operators a no-cost demand-shaping tool available today, whose benefit is largest precisely when data centers run hottest and grid stress is most salient.

Figures

Figures reproduced from arXiv: 2608.01250 by the authors.

Figure 1
Figure 1. Overview of the paper’s argument from single GPU to grid-operations decision (illustrative). [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustrative schematics of chunked-prefill batch composition. Across a representative 13,000-token whale’s full prefill, mono finishes in one large iteration [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Request arrival/prefill-size timeline (top) against the GPU power trace (bottom) for the same seeded request sequence under mono/16384, chunk/2048, [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Appendix check: widened distribution, open-loop Poisson arrivals. Chunk=512 shows an especially long, sustained near-ceiling plateau with very few [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Appendix check: genuine Pareto-tailed whale-size distribution ( [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: The ramp-rate/power-shape benefit is bundled with the already-known short-request tail-latency protection that motivates chunked prefill (panel a), at [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: A single representative realization of the simulated aggregate ramp rate [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Ramp-rate coincidence factor, normalized by each configuration’s own single-server rate. (a) Both curves shrink as [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Level coincidence factor. (a) Converges to the baseline-corrected floor as [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

18 extracted references · 10 canonical work pages

  1. [1]

    Energy and AI,

    International Energy Agency, “Energy and AI,” IEA, Tech. Rep., 2025, data centre electricity demand projected to more than double by 2030

    work page 2025

  2. [2]

    Electricity demand and grid impacts of AI data centers: Challenges and prospects,

    X. Chen, X. Wang, A. Colacelli, M. Lee, and L. Xie, “Electricity demand and grid impacts of AI data centers: Challenges and prospects,”arXiv preprint arXiv:2509.07218, 2026

    arXiv 2026

  3. [3]

    Power stabilization for AI training datacenters,

    E. Choukseet al., “Power stabilization for AI training datacenters,”arXiv preprint arXiv:2508.14318, 2025, microsoft, OpenAI, and NVIDIA

    arXiv 2025

  4. [4]

    The Llama 3 herd of models,

    Llama Team, AI @ Meta, “The Llama 3 herd of models,” 2024, arXiv preprint arXiv:2407.21783

    arXiv 2024

  5. [5]

    SARATHI: Efficient LLM inference by piggybacking decodes with chunked prefills,

    A. Agrawal, A. Panwar, J. Mohan, N. Kwatra, B. S. Gulavani, and R. Ramjee, “SARATHI: Efficient LLM inference by piggybacking decodes with chunked prefills,”arXiv preprint arXiv:2308.16369, 2023

    arXiv 2023

  6. [6]

    Taming throughput-latency tradeoff in LLM inference with Sarathi-Serve,

    A. Agrawal, N. Kedia, A. Panwar, J. Mohan, N. Kwatra, B. S. Gulavani, A. Tumanov, and R. Ramjee, “Taming throughput-latency tradeoff in LLM inference with Sarathi-Serve,” inProc. 18th USENIX Symposium on Operating Systems Design and Implementation (OSDI), 2024

    work page 2024

  7. [7]

    Wide-area power system oscillations from large-scale AI workloads,

    M.-S. Ko and H. Zhu, “Wide-area power system oscillations from large-scale AI workloads,”arXiv preprint arXiv:2508.16457, 2025

    arXiv 2025

  8. [8]

    From servers to sites: Compositional power trace generation of LLM inference for infrastructure planning,

    G. Wilkins, F. Kazhamiaka, and R. Rajagopal, “From servers to sites: Compositional power trace generation of LLM inference for infrastructure planning,” arXiv preprint arXiv:2603.18383, 2026

    arXiv 2026

Show all 18 references
  1. [9]

    Workload composition smooths aggregate power demand while sustaining short-horizon ramps in AI data centers,

    S. Majumder, M. Yu, and L. Xie, “Workload composition smooths aggregate power demand while sustaining short-horizon ramps in AI data centers,” arXiv preprint arXiv:2604.10769, 2026

    2026 arXiv

  2. [10]

    Inference as flexibility: Ramp management for transmission-connected AI data centres,

    Z. Liang, “Inference as flexibility: Ramp management for transmission-connected AI data centres,”arXiv preprint arXiv:2606.21833, 2026, reports a 10 MW/min point-of-connection ramp-rate limit imposed by the Alberta Electric System Operator (AESO) on transmission-connected data centres

    2026 arXiv

  3. [11]

    Coincidence factors for domestic EV charging from driving and plug-in behavior,

    J. Bollerslev, P. B. Andersen, T. V . Jensen, M. Marinelli, A. Thingvad, L. Calearo, and T. Weckesser, “Coincidence factors for domestic EV charging from driving and plug-in behavior,”IEEE Transactions on Transportation Electrification, vol. 8, no. 1, pp. 808–819, 2022

    2022

  4. [12]

    The case for energy-proportional computing,

    L. A. Barroso and U. H ¨olzle, “The case for energy-proportional computing,”IEEE Computer, vol. 40, no. 12, pp. 33–37, 2007

    2007

  5. [13]

    The energy cost of execution-idle in GPU clusters,

    Y . Lei, J. Fernandez, V . Kypriotis, D. Skarlatos, E. Strubell, J. Sherry, and D. V osler, “The energy cost of execution-idle in GPU clusters,”arXiv preprint arXiv:2604.04745, 2026

    2026 arXiv

  6. [14]

    Ancillary services,

    Federal Energy Regulatory Commission, “Ancillary services,” https://www.ferc.gov/ancillary-services, 2024

    2024

  7. [15]

    ShareGPT Vicuna unfiltered,

    anon8231489123, “ShareGPT Vicuna unfiltered,” https://huggingface.co/datasets/anon8231489123/ShareGPT Vicuna unfiltered, 2023, huggingFace dataset;ShareGPT_V3_unfiltered_cleaned_split.json, the file used in this paper’s experiments

    2023

  8. [16]

    Burstgpt: A real-world workload dataset to optimize llm serving systems,

    Y . Wang, Y . Chen, Z. Li, X. Kang, Y . Fang, Y . Zhou, Y . Zheng, Z. Tang, X. He, R. Guo, X. Wang, Q. Wang, A. C. Zhou, and X. Chu, “Burstgpt: A real-world workload dataset to optimize llm serving systems,” inProceedings of the 31st ACM SIGKDD Conference on Knowledge Discover...

    2025

  9. [17]

    Tracelab: Characterizing coding agent workloads for llm serving,

    K. Zhu, M. Jacob, C. Ma, Y . Pan, S. Wang, A. Krishnamurthy, and B. Kasikci, “Tracelab: Characterizing coding agent workloads for llm serving,”arXiv preprint arXiv:2606.30560, 2026

    2026 arXiv

  10. [18]

    Distribution power system reliability assessment considering cold load pickup events,

    A. Al-Nujaimi, M. A. Abido, and M. Al-Muhaini, “Distribution power system reliability assessment considering cold load pickup events,”IEEE Transactions on Power Systems, vol. 33, no. 4, pp. 4197–4206, 2018. APPENDIXA ROBUSTNESS ANDSCOPECHECKS The main body’s Section V already ...

    2018

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.