Research and Materials on Hardware implementation of Transformer Model
-
Updated
Feb 28, 2025 - Jupyter Notebook
Research and Materials on Hardware implementation of Transformer Model
PIMeval simulator and PIMbench suite
AIM, A Framework for High-throughput Sequence Alignment using Real Processing-in-Memory Systems, Bioinformatics, btad155, https://doi.org/10.1093/bioinformatics/btad155
A PIM instrumentation, compilation, execution, simulation, and evaluation repository for BLIMP-style architectures.
Source code & scripts for experimental characterization and demonstration of performing NOT and up to 16-input AND, NAND, OR, and NOR operations in real DDR4 DRAM chips. Described in our HPCA'24 paper by Yuksel et al. at https://arxiv.org/abs/2402.18736
UpPipe is an RNA abundance quantification design on a real processing-near-memory system (UPMEM DPU); the paper of this project is published in Design Automation Conference (DAC) 2023
Running state-of-the-art RNA-seq abundance quantification software "kallisto" on UPMEM DPU system
Aletheia is a memory-centric experimentation framework for exploring Processing near Memory (PNM) concepts on commodity hardware.
Processing-In-Memory acceleration of Breadth-First Search on DPUs.
Code for distributed inference of WNN on the UPMEM PiM System
Deyuan's fork with -print_tech_params, -print_decoder_breakdown, and macOS compatibility
A reference implementation of Matrix Multiplication algorithms for ML on UPMEM PIM - a processing-in-memory platform
A Basic Processing in Memory (PIM) System (Project of "Modern VLSI Design" Course)
Implementation of WNNs on the UPMEM PiM system. This implementation is optimizing for MRAM utilization.
Distributed Training on the UPMEM PiM system
Analytical model: video VLM serving is limited by KV cache capacity, not bandwidth. 113x fewer concurrent users than text on the same node.
Master's thesis research artifact for exact tensor-network quantum-circuit simulation on UPMEM Processing-in-Memory hardware.
Offloading RocksDB compaction to a near-memory (PNM) unit in full-system gem5 — +45–59% write throughput, −54 to −64% CPU L2 misses over two runs
We present an analog in-memory computing (AIMC) evaluation framework, providing SW/HW performance for LLM inference.
Upmem — independent third-party profile of a public API surface, by API Evangelist. UPMEM is a French fabless semiconductor company pioneering Processing-In-Memory (PIM). Its Data Processing Units (DPUs) are embedded directly in DRAM on standard DIMM modules, putting thousands of programmable cores inside memory to accelerate data-intensive workloa
To associate your repository with the processing-in-memory topic, visit your repo's landing page and select "manage topics."