v1.1.4

v1.1.4

Release Date: May 30, 2026
Hardware Profile: CPU (x86)

---

Model Architecture & Configuration

• Parameters: 6,684,497 (~6.68M)
• Batch Size: 16
• Block Size: 32
• Learning Rate: 1e-3
• Total Steps: 6,000

Training Performance & Metrics

• Best Validation Loss: 4.1319 (achieved at step 3,900)
• Total Training Time: 77m 16s
• Average Throughput: 791 tok/s (peaked at 885 tok/s during warmup)
• Average Step Time: 656.7 ms
• Evaluation Frequency: Every 100 steps

---

Note: The model achieved its best validation loss at step 3900. Beyond this point, the generalization gap began to widen, indicating the onset of overfitting in later steps.

What's Changed

• Refactor Dockerfile to use ARG for CUDA version by @codeaddict-119 in #57

Full Changelog: v1.1.3...v1.1.4

v1.1.3

v1.1.3
Docker / CUDA

• fix import torch failing at container startup caused by ENTRYPOINT resolving to system Python instead of the venv Python
• ENTRYPOINT now uses the absolute venv path (/app/venv/bin/python3) to avoid PATH resolution ambiguity
• Added libgl1 to runtime stage dependencies, required by torchvision

Rebuild your image to apply the fix, no other changes required.

What's Changed

• Enhance tensor management and CUDA utilities with benchmarks by @codeaddict-119 in #51
• feat :tensor management with benchmarks (#52) by @codeaddict-119 in #52

New Contributors

• @codeaddict-119 made their first contribution in #51

Full Changelog: v1.1.2...v1.1.3

v1.1.2

v1.1.2

Docker support

• Single-container build — FastAPI backend and React frontend run together via supervisord
• Multi-stage Dockerfile; CPU by default, CUDA-ready via BASE_IMAGE build arg
• Model weights mounted as a volume at runtime (/app/models)
• Published to GitHub Packages: ghcr.io/eamon2009/quadtrix.cpp:latest

docker pull ghcr.io/eamon2009/quadtrix.cpp:latest
docker run -v ./models:/app/models -p 3001:3001 -p 8080:8080 ghcr.io/eamon2009/quadtrix.cpp:latest

What's Changed

• Introduce Quadtrix benchmark suite with Python and C++ support by @Eamon2009 in #44
• Introduce Quadtrix benchmark suite with Python and C++ support (#44) by @Eamon2009 in #45

Full Changelog: v1.1.1...v1.1.2

v1.1.1

Quadtrix.cpp v1.1.1

Release Date: May 19, 2026

Minor maintenance release with bug fixes and tooling improvements.

What's Changed

Bug Fixes

• Fixed frontend session management edge cases
• Resolved backend request timeout handling
• Corrected API response formatting issues

New Features

• Added Python benchmark script for performance profiling
• Enhanced training metrics collection

Improvements

• Updated documentation
• Dependency version bumps

Setup

# Frontend
cd frontend
npm install
npm run build

# Backend
cd backend
pip install -r requirements.txt

---

Compatible with all v1.1.0 models and configurations.

What's Changed

• Implement Core Engine Entry Point (Training & Multi-Mode Inference) by @Eamon2009 in #7
• Integrate Python Inference Refactor and Frontend API Layer by @Eamon2009 in #8
• updating by @Eamon2009 in #9
• Refactor codebase, enhance documentation, and add new features by @Eamon2009 in #26
• Add benchmarking, documentation updates, and GitHub Packages workflow by @Eamon2009 in #30
• Revert dependency bump and enhance documentation and licensing by @Eamon2009 in #31
• Revert dependency bump and enhance documentation and licensing (#31) by @Eamon2009 in #38
• Implement core engine entry point and refactor Python inference by @Eamon2009 in #37
• Revert dependency bump and improve documentation for benchmarks by @Eamon2009 in #39
• Add Quadtrix benchmark dimensions and comparison tool by @Eamon2009 in #40
• Entry point for Python benchmark (#41) by @Eamon2009 in #41
• Core PyTorch benchmark suite for Quadtrix by @Eamon2009 in #42

Full Changelog: v1.1.0...v1.1.1

What's Changed

• Implement Core Engine Entry Point (Training & Multi-Mode Inference) by @Eamon2009 in #7
• Integrate Python Inference Refactor and Frontend API Layer by @Eamon2009 in #8
• updating by @Eamon2009 in #9
• Refactor codebase, enhance documentation, and add new features by @Eamon2009 in #26
• Add benchmarking, documentation updates, and GitHub Packages workflow by @Eamon2009 in #30
• Revert dependency bump and enhance documentation and licensing by @Eamon2009 in #31
• Revert dependency bump and enhance documentation and licensing (#31) by @Eamon2009 in #38
• Implement core engine entry point and refactor Python inference by @Eamon2009 in #37
• Revert dependency bump and improve documentation for benchmarks by @Eamon2009 in #39
• Add Quadtrix benchmark dimensions and comparison tool by @Eamon2009 in #40
• Entry point for Python benchmark (#41) by @Eamon2009 in #41
• Core PyTorch benchmark suite for Quadtrix by @Eamon2009 in #42

Full Changelog: v1.1.0...v1.1.1

v1.1.0

Quadtrix v1.1.0

Date: 2026-05-09  |  Device: NVIDIA T4 (CUDA / bf16)  |  Framework: PyTorch 2.10.0+cu128

---

Model Configuration

Parameter	Value
Layers	6
Heads	6
Embedding dim	100
Block size	190
Batch size	64
Dropout	0.2
Learning rate	3e-4
Total parameters	10,837,257

Training Details

Field	Value
Steps	8,000
Eval every	200 steps
Optimizer seed	1337
Train tokens	14,080,249
Val tokens	1,564,473
Precision	bf16
MFU	60.0%

Results

Metric	Value
Best val loss	2.3918
Final train loss	2.2825
Total loss drop	8.57
Peak throughput	19,602 tok/s
Mean throughput	18,756 tok/s
Peak grad norm	2.2504
Mean grad norm	1.6894
Training time	82m 43s
Checkpoint	best_model.pt

Notes

• Throughput ramps from ~279 tok/s at step 0 to a steady ~19,600 tok/s after the first eval interval, reflecting CUDA kernel warm-up.
• Gradient norms remain stable throughout training (mean 1.69), with no anomalous spikes observed.

What's Changed

• build(deps): bump @tanstack/react-query from 5.100.6 to 5.100.9 in /frontend by @dependabot[bot] in #23
• Revert "build(deps): bump @tanstack/react-query from 5.100.6 to 5.100.9 in /frontend" by @Eamon2009 in #25

Full Changelog: v1.01...v1.1.0

What's Changed

• build(deps): bump @tanstack/react-query from 5.100.6 to 5.100.9 in /frontend by @dependabot[bot] in #23
• Revert "build(deps): bump @tanstack/react-query from 5.100.6 to 5.100.9 in /frontend" by @Eamon2009 in #25

New Contributors

• @dependabot[bot] made their first contribution in #23

Full Changelog: v1.01...v1.1.0

Quadtrix v1.01

Quadtrix v1.0

Efficiency metrics

First release — token-level language model trained on CPU.

---

Training run

Metric	Value
Loss reduction	69.7% (10.82 → 3.25)
Best loss	3.252 (step 2510)
Peak throughput	435 tok/s
Wall time	~61 min

Loss curve

---

Model config

Parameter	Value
Parameters	6,684,497
Architecture	4 layers · 4 heads · 64d embedding
Batch · block size	16 · 32
Learning rate	1e-3
Dropout	0.1
Train tokens	7,065,137
Val tokens	785,016

---

How to run

python engine/main.py
python engine/inference.py

---

Notes

• Training ran on CPU (PyTorch 2.4.1) with steady 60% bf16 MFU throughout
• Loss converged from 10.82 → 3.25 over 2,690 steps in ~61 minutes
• Gradient norms stable; no spikes or divergence observed
• Checkpoint saved at step 2510 (best validation loss)