All issues

Search results

4 unit tests fail on any platform: test_flashinfer_attention_backend.py fixtures use head_dim 8/16 vs {64,128} FlashInfer gate

Status: Open.

#196 In EfficientMoE/MoE-Infinity;

· mfethe1 opened on Sep 3, 2026

Serving: DeepSeek-V2/V3 continuous-batching decode is broken — paged-attention shim exists only in tests (silent garbage output)

Status: Open.

#191 In EfficientMoE/MoE-Infinity;

· drunkcoding opened on Sep 2, 2026

[BUG] OLMoE crashes on inference: fused_moe_ffn_into hidden dim mismatch - up_proj/down_proj swapped via wrong expert-type mapping

Status: Open.

#123 In EfficientMoE/MoE-Infinity;

· danielesalpietro opened on Jul 19, 2026

[BUG] moe-infinity on PyPI is a 0.0.1.dev0 placeholder — no real wheels published

bugSomething isn't workingSomething isn't working

Status: Open.

#114 In EfficientMoE/MoE-Infinity;

· drunkcoding opened on Jul 5, 2026

[Feature Request] Improve cold start latency with ServerlessLLM sllm_store

enhancementNew feature or requestNew feature or request

Status: Open.

#65 In EfficientMoE/MoE-Infinity;

· drunkcoding opened on Jul 14, 2025

Question: Support for Continuous Batching and Asynchronous Requests

enhancementNew feature or requestNew feature or request

Status: Open.

#25 In EfficientMoE/MoE-Infinity;

· Msiavashi opened on Jun 12, 2024