SearcharxivSearch

arXiv subjects

Ahmed Sharif

Publications and source records attributed to Ahmed Sharif.

2 recordsLinked to original sources

Mechanistic Insights into Enhanced Alkaline Oxygen Evolution on Zn-Al Alloy Electrodes

Electrochemical water electrolysis, which produces clean energy carriers to mitigate carbon emissions, lacks suitable, low-cost electrodes for efficient oxygen evolution reaction (OER) in alkaline water splitting. To address this challenge, we developed Zn-Al alloy electrodes with varying Al contents up to 20 wt.% via powder metallurgy method and conducted electrochemical measurements of the OER in alkaline solution to investigate their catalytic performance. We also performed first-principles calculations to examine their thermodynamic phase stability and electronic structures. Both theoretical and experimental results indicated that incorporating $\geq 20$ wt.% Al into Zn led to thermodynamic phase instability and secondary-phase segregation in Al-rich regions, limiting reaction kinetics and reducing catalytic efficiency. Although the Al content of 5 wt.% into Zn exhibited favorable thermodynamic and electronic characteristics, but its electrochemical performance was inefficient and poor due to inadequate reaction active sites on the surface. In contrast, the 10 wt.% and 15 wt.% Al into Zn showed approximately three- and two-fold increases in anodic exchange current density relative to pure Zn, respectively. Additionally, the anodic overpotential losses ($η_{0,a}$) measured at a current density of 12 mAcm$^{-2}$ were 0.240 V for Zn$_{0.9}$Al$_{0.1}$ and 0.5603 V for Zn$_{0.85}$Al$_{0.15}$, significantly lower than that of pure Zn ($η_{0,a} = 1.086$ V). While Zn$_{0.9}$Al$_{0.1}$ and Zn$_{0.85}$Al$_{0.15}$ showed similar charge transfer resistance ($R_{\rm CT}$), Zn$_{0.9}$Al$_{0.1}$ demonstrated superior reaction kinetics and lower $η_{0,a}$ across all samples tested. Furthermore, the improved kinetics and reduced overpotential of the Zn-Al alloys favorably compare with those of other transition-metal-based catalysts, including Fe-Co-Ni-Mo alloys and Fe-doped CuO.

cond-mat.mtrl-sci

Training LLMs with Fault Tolerant HSDP on 100,000 GPUs

Large-scale training systems typically use synchronous training, requiring all GPUs to be healthy simultaneously. In our experience training on O(100K) GPUs, synchronous training results in a low efficiency due to frequent failures and long recovery time. To address this problem, we propose a novel training paradigm, Fault Tolerant Hybrid-Shared Data Parallelism (FT-HSDP). FT-HSDP uses data parallel replicas as units of fault tolerance. When failures occur, only a single data-parallel replica containing the failed GPU or server is taken offline and restarted, while the other replicas continue training. To realize this idea at scale, FT-HSDP incorporates several techniques: 1) We introduce a Fault Tolerant All Reduce (FTAR) protocol for gradient exchange across data parallel replicas. FTAR relies on the CPU to drive the complex control logic for tasks like adding or removing participants dynamically, and relies on GPU to perform data transfer for best performance. 2) We introduce a non-blocking catch-up protocol, allowing a recovering replica to join training with minimal stall. Compared with fully synchronous training at O(100K) GPUs, FT-HSDP can reduce the stall time due to failure recovery from 10 minutes to 3 minutes, increasing effective training time from 44\% to 80\%. We further demonstrate that FT-HSDP's asynchronous recovery does not bring any meaning degradation to the accuracy of the result model.

cs.DC