Education●●●Banger
PaperDraw.dev – A System Design Simulator
System design meets flight simulator—watch traffic propagate, bottlenecks emerge, failures cascade.
Solve My ProblemEye CandyRabbit Hole
vhatkarpratap
205mo ago
Back to browse
GitHub Repository
A distributed hardware-software co-design fabric for MoE models (DeepSeek-V3, Mixtral) that eradicates NCCL All-to-All communication stalls via distributed RoCEv2 RDMA virtual address MUX and JAX/XLA SPMD sharding.
0 starsPython
Hardware-software co-design for MoE models to bypass NCCL bottlenecks
by PJHkorea·Aug 4, 2026·1 point·1 comment
AI Analysis
●MidBold Bet
Speculative infra for MoE training bottlenecks with zero stars and no benchmarks yet.
Strengths
- •Attempts to bridge RDMA virtual addresses with JAX sharding for zero-copy routing.
- •Targets a real pain point in distributed MoE training communication overhead.
Weaknesses
- •PoC with no stars, forks, or performance benchmarks to validate the approach.
- •Highly niche audience and unproven whether this actually improves training speed.
Category
Target Audience
ML engineers working on distributed MoE training
Similar To
DeepSpeed · FairScale · JAX distributed
Similar Projects
AI/ML●●●Banger
NVFP4 on Desktop Blackwell – 122B MoE on a Single RTX PRO 6000 31 tok/s
Bypasses NVIDIA's artificial FP4 lock—122B MoE on single desktop GPU at 31 tok/s.
WizardryDark Horse
jcartu
204mo ago
Security●●●Banger
deterministic oracle for hardware designs with replayable proofs
Fingerprints hardware interaction graphs to catch trojans functional testing misses.
Big BrainWizardry
suprastructure
103mo ago
AI/ML●Mid
Tok/s on a 35B MoE model using a $100 AMD crypto APU and Vulkan
Clever hardware hack but this is a config guide, not a shipped tool.
Niche Gem
akandr
214mo ago
AI/ML●●●Banger
SwiftLM – Qwen Chat on iPhone, 100B+ Moe on M5 Pro 64GB (Native Swift)
Native Swift inference with SSD streaming runs 100B MoE models without kernel panics.
WizardryNiche Gem
aegis_camera
124mo ago
AI/ML●●Solid
Tokenstead, find AI models for your hardware
Enterprise AI deployment tracker with source-cited cases beats generic model compatibility charts.
Dark HorseSolve My Problem
cdnsteve
3227d ago