Search results

perf: multi-NIC parallel RDMA pull — MX is single-NIC-pinned by default, leaves 75-87% of allocated bandwidth idle

Status: Open.

#449 In ai-dynamo/modelexpress;

· KavinKrishnan opened on Jun 23, 2026

filter no-rdma tensor

Status: Open.

#439 In ai-dynamo/modelexpress;

· gaogaoSpark opened on Jun 15, 2026

Standalone host-memory sources for ModelExpress

Status: Open.

#422 In ai-dynamo/modelexpress;

· athreesh opened on Jun 5, 2026

[RFC] Coordinator gRPC authorization using device assertions

Status: Open.

Feature

#298 In ai-dynamo/modelexpress;

· wseaton opened on May 25, 2026

[RFC] Proposal for a central transfer planner between peers

Status: Open.

#293 In ai-dynamo/modelexpress;

· wseaton opened on May 20, 2026

vLLM DP workers can collide on MX P2P serving ports when CUDA device is unindexed

Status: Open.

#266 In ai-dynamo/modelexpress;

· zhengluo-nv opened on May 6, 2026

[Bug] Lint PR workflow permission for fork PRs

Status: Open.

#263 In ai-dynamo/modelexpress;

· emricksini-h opened on May 5, 2026

[Feature] Optimize model downloads by skipping duplicate weight formats

Status: Open.

#173 In ai-dynamo/modelexpress;

· Jont828 opened on Mar 20, 2026

[Feature] Download acceleration: multi-file parallelism, retry resilience, and tunable download parameters

Status: Open.

#149 In ai-dynamo/modelexpress;

· baladithyab opened on Feb 25, 2026

[Feature] Distributed Downloading

Status: Open.

#78 In ai-dynamo/modelexpress;

· tlipoca9 opened on Sep 24, 2025