Fractal-Compute-Orchestrations /FractalWorkspace
A unified ecosystem and workspace for large-scale federated learning of Large Models & LLMS. Store compute as Wifi
★ 9,203⑂ 14👁 18⑃ 10● 1▤ 1
compute-orchestrationdistributed-systemsedge-computingedge-intelligenceedge-trainingfederated-learningiotmachine-learning
More from this author
-8
Daily
9,203
stars
14
forks
18
watchers
10
prs
1
issues
1
releases
README
!Fractal Banner
Fractal
Distributed Compute Orchestration and Edge Intelligence Framework
A unified distributed systems framework bridging centralized model orchestration with decentralized edge intelligence.
Overview | Key Results & Benchmarks | Industry & Community Recognition | Subsystems | Architecture | Operational Paradigms | Global Contracts | Academic Foundations | Getting Started
---
Overview
Executive Summary: The Core Breakthrough & Bottom Line
The Problem: Foundation models (TinyLlama, Llama-3, Mistral) cannot participate in decentralized edge federated learning because consumer mobile devices face three compounding physical bottlenecks:
1. The Uplink Bottleneck: Transmitting uncompressed model checkpoints (4.2 GB to 30.5 GB) over consumer Wi-Fi/LTE (5 to 10 Mbps) requires 5 to 10 minutes per client, triggering aggregator timeouts (120s cutoff) and dropping edge nodes.
2. The Dynamic Range Collapse: Standard 4-bit quantization breaks modern attention projections; emergent outliers inflate global scales, squashing 99.9% of normal weights into zero bins, collapsing Layer 5 cosine fidelity to 0.6990, and inducing catastrophic training divergence.
3. The Mobile Memory Wall (LMK): Checkpoint loading exceeds mobile RAM limits, triggering immediate Android Low Memory Killer (LMK) process evictions and battery overheating.
The FRACTAL Achievement: FRACTAL breaks all three physical walls through hardware-aligned algorithmic co-design:
7.11x Wire Payload Reduction (85.94% Wire Cut): Parameter payload drops from 4,196 MB down to 590 MB for TinyLlama-1.1B, and from 30.5 GB down to 4.29 GB for Llama-3 8B.
Sub-60s Transmission Guarantee: Slashes 10 Mbps upload duration from 321.2 seconds down to 46.2 seconds, beating the 120s federated aggregation deadline by 73.8 seconds.
Zero Accuracy Collapse (0.9878 Fidelity Preserved): Micro-block dynamic scaling (64-element tiles) confines attention projection outliers locally, sustaining 0.9878 mean cosine similarity and 4.3e-5 MSE across all 22 layers, matching lossless 16-bit convergence.
Zero Mobile LMK Crashes: Native memory-mapped execution (mmap) and an 85.6% reduction in radio transmission time keep mobile nodes thermally cool and crash-free.
Fractal is a high-performance distributed computing ecosystem engineered to harvest fragmented consumer compute across mobile edge devices while strictly preserving user privacy and device health. The framework treats edge devices not merely as remote processors, but as autonomous, resource-constrained nodes operating under human-centric constraints.
The Frontier Challenge
Centralized foundation model training and inference face compounding physical and financial bottlenecks:
1. The Centralized Compute Monopoly: Foundation model adaptation is concentrated in capital-intensive GPU clusters consuming megawatts, while billions of consumer edge devices with dedicated neural engines sit idle on chargers.
2Discussions
#2Why are PyTorch .bin weights ignored when provisioning the Llama 3 model?1 comments · 2026-06-27#1How to handle Hugging Face token authentication when deploying Fractal in headless Docker environments?1 comments · 2026-06-27
Latest release
FRACTAL Ecosystem v1.0.0 v1.0.0 2026-11-13Coordinated FRACTAL v1.0.0 release by Fractal Compute Orchestrations linking FractalCore v1.0.0 (fractal_quant INT4 SDK, CLI, hardened coordinator, slicer pipeline) and FractalAndroid v1.0.0 (wire-compatible Int4BlockCodec, OperationControl thermal gating, FractalPaths storage isolation, JVM unit tests, and pre-built debug APK), governed by the FRACTAL Proprietary Source-Available & Non-Commercial Restrictive License v3.1.