AI/ML●●●Banger
Granite Switch - compose multiple LoRA adapters to one deployable model
Composing multiple LoRA adapters into one checkpoint solves the model sprawl nightmare.
Big BrainSolve My Problem
bignet
301mo ago
Back to browse
GitHub Repository
A memory and execution optimization architecture for AI models
4 starsPython
S2LC – 100 LoRA adapters in 3.59ms, zero HBM writes
by ai_spokesperson·Mar 22, 2026·1 point·0 comments
AI Analysis
●●●BangerWizardryBig Brain
3.59ms for 100 LoRA adapters with zero HBM writes—genuine GPU wizardry.
Strengths
- •Shared spectral basis via truncated SVD cuts memory 10.1× theoretically
- •Triton kernel reconstructs weights entirely in GPU register file—zero intermediate writes
- •CUDA Graph capture collapses 128 kernel dispatches into single replay call
Weaknesses
- •Narrow audience—only matters if you're serving many LoRA adapters at scale
- •Requires A100/H100 GPUs, limits accessibility for smaller teams
Category
Target Audience
ML infrastructure engineers, LLM serving teams
Similar To
LoRAX · vLLM · Punica
Similar Projects
AI/ML●Mid
LunaLora: Multi-LoRA System to Combat Catastrophic Forgetting
Stores memory in LoRA weights instead of cache, but lacks working benchmarks.
Bold BetShip It
SphericalCowww
122mo ago
AI/ML●●Solid
100cc - Roll your own Claude in 100 lines
Self-bootstrapping agent writes its own improvements in 100 lines of TypeScript.
Big BrainCozy
rapiz
10411d ago
Infrastructure●Mid
Deploybase CLI – Search GPU and LLM pricing from your terminal
Thin CLI wrapper around deploybase.ai when the website already shows this data.
Solve My Problem
grasper_
312mo ago
AI/ML●Mid
Adaptive Runtime – AI agent layer, no GPU, crash recovery
Agent runtime infra, but 0 stars and crowded with LangGraph and Temporal.
Ship ItBold Bet
StateflowsLabs
3115d ago
Productivity●●●Banger
Vocalinux // 100% offline voice typing for Linux
Linux finally gets offline voice typing; Ctrl-tap + Vulkan GPU support vs cloud-dependent alternatives.
Solve My ProblemDark Horse
jatinkrmalik
403mo ago