velocitatem/PHANTOM
1
0
Fork 0
mirror of https://github.com/velocitatem/PHANTOM.git synced 2026-05-31 16:43:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

citing compute

Browse Source
This commit is contained in:
Daniel Rosel
2026-02-17 14:46:34 +01:00
parent 76c31a2abd
commit 244af9ac09
2 changed files with 69 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
paper/src/bib/references.bib
View File

@@ -585,3 +585,34 @@ Volume: 21},
year = {2026}, year = {2026},
file = {Snapshot:/home/velocitatem/Zotero/storage/DGW8PHMV/marc-andreessen-the-real-ai-boom.html:text/html}, file = {Snapshot:/home/velocitatem/Zotero/storage/DGW8PHMV/marc-andreessen-the-real-ai-boom.html:text/html},
} }
@misc{noauthor_tpu_2025,
title = {{TPU} v6e},
url = {https://cloud.google.com/tpu/docs/v6e},
language = {es-419-x-mtfrom-en},
urldate = {2026-02-17},
journal = {Google Cloud Documentation},
month = dec,
year = {2025},
file = {Snapshot:/home/velocitatem/Zotero/storage/RNMB32KD/v6e.html:text/html},
}
@misc{noauthor_tpu_2025-1,
title = {{TPU} v5e {\textbar} {Google} {Cloud} {Documentation}},
url = {https://cloud.google.com/tpu/docs/v5e},
language = {es-419-x-mtfrom-en},
urldate = {2026-02-17},
month = dec,
year = {2025},
file = {Snapshot:/home/velocitatem/Zotero/storage/BLLG9NZC/v5e.html:text/html},
}
@misc{noauthor_tpu_2026,
title = {{TPU} v4 {\textbar} {Google} {Cloud} {Documentation}},
url = {https://cloud.google.com/tpu/docs/v4},
language = {es-419-x-mtfrom-en},
urldate = {2026-02-17},
month = feb,
year = {2026},
file = {Snapshot:/home/velocitatem/Zotero/storage/N724QGF6/v4.html:text/html},
}

39
paper/src/chapters/03-methodology.tex
View File

@@ -198,7 +198,44 @@ The dynamic pricing mechanism elicited immediate behavioral adjustments. Partici
\subsubsection{Design of Training Factorial Study} \subsubsection{Design of Training Factorial Study}
The simulator has multiple configurable factors, including valuation distributions, demand parametrization, contamination ratio, and policy settings. We therefore design a multi-factor study (current grid estimate: $4\times4\times3\times2\times2$). While this scale is generally expensive for reinforcement learning, we execute it on a large TPU cluster to make the sweep tractable and logged with services provided by weights and biases. The simulator has multiple configurable factors, including valuation distributions, demand parametrization, contamination ratio, and policy settings. We therefore design a multi-factor study (current grid estimate: $4\times4\times3\times2\times2$). While this scale is generally expensive for reinforcement learning, we execute it on a large TPU cluster to make the sweep tractable.
Our training budget is provisioned through TPU Research Cloud and spans 384 chips across TPU v4, v5e, and v6e generations, with a spot-heavy allocation plus an on-demand reserve. At peak BF16 throughput this corresponds to approximately 160 PFLOPS of aggregate compute, which makes repeated seeds, ablations, and sensitivity sweeps feasible within practical wall-clock limits. We allocate v6e capacity to the highest-intensity policy training jobs, use v5e for wider hyperparameter exploration where throughput-per-dollar is favorable, and reserve on-demand v4 capacity for runs that should not be interrupted.
\begin{table}[ht]
\centering
\caption{Compact comparison of TPU generations used in the training stack.}
\label{tab:tpu_specs}
\begin{tabular}{@{}llll@{}}
\toprule
\textbf{Feature} & \textbf{TPU v4} & \textbf{TPU v5e} & \textbf{TPU v6e (Trillium)} \\
\midrule
Peak BF16 per chip (TFLOPS) & 275 & 197 & 918 \\
HBM capacity per chip (GB) & 32 & 16 & 32 \\
HBM bandwidth per chip (GB/s) & 1200 & 819 & 1600 \\
TensorCores per chip & 2 & 1 & 1 \\
Interconnect topology & 3D mesh/torus & 2D torus & 2D torus \\
Max pod size (chips) & 4096 & 256 & 256 \\
\bottomrule
\end{tabular}
\end{table}
\begin{table}[ht]
\centering
\caption{TPU allocation used for the factorial study.}
\label{tab:tpu_allocation}
\begin{tabular}{@{}llll@{}}
\toprule
\textbf{TPU Type} & \textbf{Total Chips} & \textbf{Zone(s)} & \textbf{Provisioning} \\
\midrule
v6e & 128 (64 + 64) & europe-west4-a, us-east1-d & Spot \\
v5e & 128 (64 + 64) & us-central1-a, europe-west4-b & Spot \\
v4 & 64 (32 + 32) & us-central2-b & 32 Spot + 32 On-demand \\
\bottomrule
\end{tabular}
\end{table}
For interactive monitoring from Madrid, we prioritize the europe-west4 allocation for latency-sensitive runs. All sweep metadata, model checkpoints, and reward traces are logged in Weights \& Biases. Hardware specifications are from the official Google Cloud TPU documentation \parencite{noauthor_tpu_2026,noauthor_tpu_2025-1,noauthor_tpu_2025}.
\subsubsection{Interaction Schema} \subsubsection{Interaction Schema}