Muhammad Ibrahim

Affiliation:

Papers on Lattice

Total citations

Topics

Publication activitypapers/week, last 8 weeks

Research focus

Architecture Design (Transformers, SSMs, MoE) (1)Distributed Systems & Hardware (1)Inference & Quantization (1)

Frequent co-authors

M. Ibrahim (1)

Papers (1)

Oct 15, 2026

Affiliation:2d ago

When the KV Cache Is Slower Than the GPU: Scaling Remote KV Tiers for Context-Parallel Prefill

Remote KV caches can cripple LLM prefill latency by orders of magnitude if hit costs exceed recomputation, a problem solved by sharding and asynchronous loading.

M. Ibrahim, Muhammad Ibrahim

Architecture Design (Transformers, SSMs, MoE)Distributed Systems & Hardware Inference & Quantization

Muhammad Ibrahim

Publication activitypapers/week, last 8 weeks

Research focus

Frequent co-authors

Papers (1)

Search

Muhammad Ibrahim

Publication activitypapers/week, last 8 weeks

Research focus

Frequent co-authors

Papers (1)