Search results

R3 accuracy bug

Status: Open.

#1186 In radixark/miles;

· guapisolo opened on May 23, 2026

SGLang custom all-reduce v2 CUDA graph capture fails under Miles colocate rollout

bugSomething isn't workingSomething isn't working

help wantedExtra attention is neededExtra attention is needed

Status: Open.

#1176 In radixark/miles;

· yueming-yuan opened on May 22, 2026

PYTHONBUFFERED seems like a mistake. It should be PYTHONUNBUFFERED

Status: Open.

#1134 In radixark/miles;

· timgianitsos opened on May 15, 2026

[RFC] SendRecv-Broadcast-Based Disaggregated Weight Sync

Status: Open.

#1120 In radixark/miles;

· zyzshishui opened on May 12, 2026

DeepSeek-V4 support roadmap on ROCm

Status: Open.

#1113 In radixark/miles;

· lizamd opened on May 11, 2026

Miles CI gap Between ROCm & CUDA

Status: Open.

#1105 In radixark/miles;

· indianspeedster opened on May 11, 2026

R3 not working for Qwen3.5 PD disaggregated

Status: Open.

#1085 In radixark/miles;

· jthomy opened on May 7, 2026

Deepseek V4 RL Roadmap

Status: Open.

#1046 In radixark/miles;

· yueming-yuan opened on Apr 24, 2026

R3 replay: RuntimeError 'Split sizes doesn't match total dim 0 size' in Megatron all_to_all_single on MoE compute_log_prob

Status: Open.

#1002 In radixark/miles;

· DavidBellamy opened on Apr 17, 2026

support for qwen3.5 397b

Status: Open.

#983 In radixark/miles;

· bhaktatejas922 opened on Apr 15, 2026

[multi_agent] rollout generation stalls for a long time before any /generate fires

Status: Open.

#979 In radixark/miles;

· GuanxingLu opened on Apr 14, 2026

Feature request: bounded-uneven partitioning mode for --balance-data

Status: Open.

#970 In radixark/miles;

· mvillmow opened on Apr 10, 2026