AI/ML●●Solid
NeuralForge – Fine-Tune LLMs on Your Mac Using Apple Neural Engine
Fine-tune LLMs on Apple Neural Engine using reverse-engineered private frameworks — genuinely novel approach.
WizardryBig BrainNiche Gem
khaeldur
112mo ago
Back to browse
GitHub Repository
LoRA fine-tuning gradients on Apple Neural Engine via private APIs. Conv-as-matmul, spatial constraints, MLX integration.
8 starsPython
LoRA gradients on Apple's Neural Engine at 2.8W
by jmanhype·Mar 6, 2026·6 points·1 comment
AI Analysis
●●●BangerWizardryBig BrainDark Horse
First LoRA gradients on ANE; matmul doesn't work, conv-as-matmul does.
Strengths
- •Discovers undocumented ANE failure modes (matmul silently broken, spatial dimension constraint) via rigorous testing.
- •Concrete conv-as-matmul mapping for 4-operation gradient flow with verified loss convergence across steps.
- •Subprocess isolation workaround solves compile handle leak — practical production insight.
Weaknesses
- •Relies on private Apple APIs; zero stability or forward-compatibility guarantees.
- •Marginal energy win (2.8W) vs. battery capacity; real-world training speedup not quantified.
Category
Target Audience
ML engineers on Apple Silicon, LLM fine-tuning practitioners, reverse-engineering enthusiasts.
Similar To
MLX · CoreML Tools · maderix/ANE
Similar Projects
Developer Tools●●Solid
Courtyard – Open-source macOS app for local MLX fine-tuning Text
Tauri GUI wrapper around mlx-lm—useful for Mac users, but local fine-tuning UIs already exist.
Solve My ProblemShip It
tuwenbo0120
103mo ago
AI/ML●●●Banger
Orion – Native Training LLMs on the Apple Neural Engine Without CoreML
Direct ANE access bypasses CoreML to enable training—genuinely novel Apple Silicon unlock.
WizardryZero to OneBig Brain
mechramc
213mo ago
AI/ML●●Solid
I fine-tuned Qwen 3.5 (0.8B–4B) on a Mac for text-to-SQL – 2B beats 12B
Unified memory trick lets a 2B model beat 12B; trains on MacBook with zero cloud costs.
Ship ItNiche GemBig Brain
sciences44
713mo ago
Health●●Solid
Health optimization as agent-guiding gradient descent
Research-as-loss-function lets stale-knowledge agents guide health optimization via gradients.
Big BrainNiche Gem
dingmuti
103mo ago
Developer Tools●●●Banger
GEKO (up to 80% compute savings on LLM fine-tuning)
Mountain Curriculum routing: 5× compute to hard samples, skip mastered ones.
Big BrainWizardryShip It
SyedAbdurR2hman
113mo ago