Soroush Tabesh

12 papers A* 3Journal 9
YearRankTypeTitle / Venue / Authors
2026 J jnl
Quartet II: Accurate LLM Pre-Training in NVFP4 by Improved Unbiased Gradient Estimation.
CoRR
Andrei Panferov, Erik Schultheis, Soroush Tabesh, Dan Alistarh
2025 J jnl
ASIDE: Architectural Separation of Instructions and Data in Language Models.
CoRR
Egor Zverev, Evgenii Kortukov, Alexander Panfilov, Soroush Tabesh, Alexandra Volkova, Sebastian Lapuschkin, Wojciech Samek, Christoph H. Lampert
2025 J jnl
CAGE: Curvature-Aware Gradient Estimation For Accurate Quantization-Aware Training.
CoRR
Soroush Tabesh, Mher Safaryan, Dan Alistarh
2025 A* conf
Can LLMs Separate Instructions From Data? And What Do We Even Mean By That?
ICLR
Egor Zverev, Sahar Abdelnabi, Soroush Tabesh, Mario Fritz, Christoph H. Lampert
2025 J jnl
HALO: Hadamard-Assisted Lower-Precision Optimization for LLMs.
CoRR
Saleh Ashkboos, Mahdi Nikdan, Soroush Tabesh, Roberto L. Castro, Torsten Hoefler, Dan Alistarh
2025 A* conf
QuEST: Stable Training of LLMs with 1-Bit Weights and Activations.
ICML
Andrei Panferov, Jiale Chen, Soroush Tabesh, Mahdi Nikdan, Dan Alistarh
2025 J jnl
QuEST: Stable Training of LLMs with 1-Bit Weights and Activations.
CoRR
Andrei Panferov, Jiale Chen, Soroush Tabesh, Roberto L. Castro, Mahdi Nikdan, Dan Alistarh
2025 J jnl
Quartet: Native FP4 Training Can Be Optimal for Large Language Models.
CoRR
Roberto L. Castro, Andrei Panferov, Soroush Tabesh, Oliver Sieberling, Jiale Chen, Mahdi Nikdan, Saleh Ashkboos, Dan Alistarh
2025 J jnl
TACO Vision Models Can Be Efficiently Specialized via Few-Shot Task-Aware Compression.
Trans. Mach. Learn. Res.
Denis Kuznedelev, Soroush Tabesh, Kimia Noorbakhsh, Elias Frantar, Sara Beery, Eldar Kurtic, Dan Alistarh
2024 A* conf
RoSA: Accurate Parameter-Efficient Fine-Tuning via Robust Adaptation.
ICML
Mahdi Nikdan, Soroush Tabesh, Elvir Crncevic, Dan Alistarh
2024 J jnl
RoSA: Accurate Parameter-Efficient Fine-Tuning via Robust Adaptation.
CoRR
Mahdi Nikdan, Soroush Tabesh, Dan Alistarh
2023 J jnl
Vision Models Can Be Efficiently Specialized via Few-Shot Task-Aware Compression.
CoRR
Denis Kuznedelev, Soroush Tabesh, Kimia Noorbakhsh, Elias Frantar, Sara Beery, Eldar Kurtic, Dan Alistarh