LLM Training Dashboard - (AI-Augmented Software Engineering)
Manage Training Workflow, 3D Visualizations, Ollama Integration and Model Artifact Clean Up
Architecture, Design and Development by Franz Ayestaran / Enhanced Pair Programming with Claude Code & OpenAI GPT
Loading workflow documentation...
📊 LLM Training Pipeline Presentation
Download the complete PowerPoint presentation covering the LLM creation and Ollama deployment pipeline
🚀 Train Your Model
📂 Upload Training Data
Upload a .txt, .json, or .jsonl training file. JSON datasets can use records like {"instruction": "...", "output": "..."}.
⚙️ Training Configuration
LoRA Configuration
🔄 Resume from Checkpoint (Optional)
Continue training from a saved checkpoint instead of starting from scratch.
Recommended resets the form to the dashboard baseline. Export saves the current dashboard settings as a versioned JSON config. Load applies a previously saved config back into the form and restores the referenced training file when available.
💻 System Status
CPU
0 coresMemory (RAM)
0 GB totalGPU
Checking...Device Information
NVIDIA
Utilization Details
CPU Cores
Disk I/O
Network I/O
🧪 Phase-1 ML/DL And Classical ML Training
Train Phase-1 deep learning models, Phase-2 classical ML models, and the new Phase-3 Vision Transformer and small text transformer models from one dashboard page. The sections stay stacked in order so Phase 3 appears below the existing classical workflow.
A. Model Selector
B. Dataset Upload Panel
Choose a model to see its supported dataset format.
C. Hyperparameter Panel
Training Actions
📈 Phase-2 Classical ML Models
Train Random Forest, SVM, and Logistic Regression models on CSV datasets with schema preview, target-column control, optional ONNX export, telemetry, and downloadable `.pkl` artifacts.
A. Classical Model Selector
B. CSV Dataset Upload
Choose a classical model to see its supported dataset format.
C. Hyperparameter Panel
Training Actions
🛰️ Phase-3 Vision Transformers and Small Transformers
Train Vision Transformer plus tiny causal LLaMA text models with image-folder or JSONL datasets, live telemetry, ONNX export, ONNX Runtime inference support, and GGUF/Ollama export for the Phase-3 text models.
A. Transformer Model Selector
B. Dataset Upload Panel
Choose a Phase-3 model to see its supported dataset format.
C. Hyperparameter Panel
Training Actions
🤖 Chat with Your Model
💬 Generate Text
🦙 Import Model to Ollama
📦 Import GGUF Model
ℹ️ About: This will import your trained model into Ollama, allowing you to run it locally using the ollama run command.
⚠️ No GGUF file? If you don't see any GGUF files below, you need to convert your trained model first.
ollama run <name>
☁️ Cloud deploy: Use the second button to upload the selected GGUF and create the model on ollama.ayestaran.dev. This requires SSH access from the dashboard host to the server.
☁️ Ollama Cloud Server Admin
ℹ️ About: Manage models on ollama.ayestaran.dev. Listing uses the remote tags endpoint, and deletion uses SSH access from the dashboard host.
📊 Training Artifacts
Loading artifacts...
📊 3D Model Visualizations
Interactive 3D visualizations of your model's training process, checkpoints, embeddings, layer structures, and internal LLM topology.
Checking for visualizations...
🧠 3D LLM Attention Explorer
A dedicated scene for inspecting how query, key, and value paths move through a transformer block. The layout is intentionally closer to a cinematic architecture diagram than the existing topology plot, with Q/K/V weight slabs, vector stages, layer norm, attention scoring, and animated token flow.
Separate Q, K, and V stages, token score routing, residual handoff, and downstream attention aggregation in one navigable 3D scene.
Orbit, pan, and zoom the scene. Click a subsystem to pin metadata and use the scene controls to jump straight to Q, K, V, or the attention matrix.
The page uses a hand-built scene rather than Plotly so the composition can read more like a technical explainer, matching the reference style more closely.
🧬 3D Neural Brain Journey
A dedicated brain-shaped scene for travelling through your trained model, now with a semantic Embedding Galaxy mode. Switch between the structural 3D brain and a token-star universe where clusters become constellations, dense regions become nebulae, and model-specific concepts form their own colored territories.
Brain mode shows the residual spine, attention branches, and sampled learned pathways from the real checkpoint. Galaxy mode remaps sampled token embeddings into a 3D star field.
Use the built-in waypoints to jump from input cortex to mid-layer reasoning and then into the output crown. Click any node to pin its layer, tensor path, and sampled row metrics.
This atlas is sampled, not literal-all-edges. Rendering every trained connection in a browser is not feasible at LLM scale, so the scene uses the strongest learned pathways to stay explorable.
🛠️ Technical Stack
Overview of the Phase-1 ML/DL, Phase-2 classical ML, and Phase-3 transformer workflows that power this dashboard across NVIDIA CUDA and Apple Silicon MLX/MPS environments.
🧠 Core Deep Learning Framework
PyTorch 2.0+
Primary runtime for Phase-1 CNN/MLP/RNN training, custom transformer experiments, Phase-3 Vision Transformer pipelines, and shared optimization and export flows.
Transformers 4.30+
Hugging Face stack for tokenization, pre-trained checkpoints, compact text transformers, and PEFT-based LoRA fine-tuning workflows.
CUDA / MPS / MLX
CUDA accelerates NVIDIA training and quantized fine-tuning workloads.
Apple Silicon (MPS / MLX): Native GPU acceleration for M-series processors via Metal Performance Shaders, with MLX-LM enabling Apple-native quantized LoRA and QLoRA training and fused export flows.
🏗️ Model Coverage & Training Workflows
Model Coverage (transformer.py + backend/ml_models)
- Phase 1 ML/DL Models - CNN, MLP, and RNN training flows with CSV, JSONL, and image dataset support
- Custom Transformer Blocks - Attention, positional encoding, feed-forward layers, and decoder-oriented experiments
- Phase 2 Classical ML - Random Forest, SVM, and Logistic Regression registries with CSV schema inspection
- Phase 3 Models - Vision Transformer, MicroLLaMA, and MiniLLaMA workflows with optional ONNX export plus GGUF conversion for the text-model path
- Shared Dataset Pipeline - Built-in upload, sample-dataset generation, preview, validation, and run history flows
Training Workflows (finetune.py / finetune_mlx.py / backend/ml_models)
- Gradient Accumulation & MPS Safety - Automatic batch-size and sequence-length adjustments for constrained Apple Silicon memory
- LoRA / QLoRA Backends - PEFT plus bitsandbytes on CUDA and MLX-native quantized fine-tuning on Apple Silicon
- Resume, Validation & Scheduling - Cosine scheduling, gradient checkpointing, resume-from-checkpoint, and epoch validation samples
- Export Paths - SafeTensors, GGUF, ONNX, and artifact bundles for local deployment and comparison
- Telemetry & History - Training status, logs, history, and downloadable artifacts surfaced directly in the dashboard
🌐 Web Dashboard & Visualization
Flask 3.0+
Flask serves the unified Phase-1, Phase-2, and Phase-3 dashboard, training APIs, dataset upload routes, telemetry, and artifact export endpoints.
Plotly 5.20+
Interactive 3D visualizations for checkpoints, embeddings, layer structures, neural atlases, and interpretability scenes.
Marked.js
Markdown parsing for rendering documentation and workflow guides directly in the dashboard.
📊 Data Processing & System Monitoring
NumPy 1.24+
Array-centric processing for tabular, text, image, and geometry workflows across ML and deep-learning training.
PSUtil 5.9+
Real-time system and process monitoring for CPU, memory, GPU, and disk usage tracking.
TQDM 4.65+
Progress reporting for fine-tuning, classical ML jobs, dataset conversion, and export pipelines with ETA estimates.
🦙 Model Export & Deployment
SafeTensors 0.4+
Checkpoint and adapter serialization for transformer fine-tuning, resume bundles, and model conversion pipelines.
Ollama Integration
GGUF export path for PyTorch and Apple-native MLX outputs, with direct Ollama import for local inference.
GGUF / ONNX Runtime
llama.cpp quantization supports GGUF deployment, while ONNX and ONNX Runtime cover Phase-2 exports plus optional ONNX and GGUF paths for the Phase-3 text models.
🔧 ML, Tracking & Inference Tooling
TensorBoard 2.12+
Loss, learning-rate, and gradient tracking for deep-learning runs and checkpoint inspection.
Weights & Biases
Experiment tracking and collaboration platform for machine learning projects.
Scikit-learn 1.2+
Phase-2 classical ML stack for Random Forest, SVM, and Logistic Regression training, metrics, and optional skl2onnx export.
🔬 Transformer Interpretability Stack
LLM Training Dashboard progressively exposes the internal logic of large language models through twelve hierarchical layers, moving from high-level semantic geometry down to neuron-level concept discovery.
- Embedding Galaxy - Constructs the top-level semantic space, mapping tokens into a unified representational geometry.
- Brain Atlas - Defines the macro-architecture of the model, showing how major regions such as attention, MLP, and embedding blocks interconnect.
- Tensor Microarchitecture - Visualizes tensor statistics and heatmaps for fine-grained inspection of weight distributions and activation patterns.
- Head-Aware Q/K/V Decomposition - Separates attention heads to analyze norms, sparsity, and per-head behavior.
- Activation-Path Visualization - Traces query, key, and value activations through attention weights to reveal how information flows within a layer.
- Multi-Head Interaction Map - Examines horizontal structure through head-to-head similarity, clustering, redundancy, and specialization, including syntax, induction, and negation heads.
- Layer-to-Layer Activation Flow - Explores vertical structure by showing how outputs propagate across layers to form emergent circuits and conceptual hierarchies.
- MLP Neuron Concept Discovery - Discovers concept neurons and feature detectors in MLP layers that respond to interpretable patterns such as numbers, names, and emotions.
- Feature-Space Geometry - Visualizes principal components, concept directions, neuron clusters, and subspaces for specific behaviors.
- Time Drift Visualisation - Tracks how embeddings, attention patterns, and neuron behaviours shift across checkpoints to reveal when representations diverge or capabilities emerge.
- Gradient Flow & Influence Maps - Reveals why a model chose its output by tracing token‑level gradients, attribution signals, and causal influence pathways.
- Mechanistic Circuits & Subgraph Extraction - A multi‑level framework that reveals how transformer models represent, transform, and reason through their internal mechanisms.
🎯 System Architecture
Data Pipeline: CSV / JSONL / image-folder input → preview and validation → dataset builders → training loaders
Training Loop: Phase-1 deep learning loops / Phase-2 classical fit / Phase-3 transformer fine-tuning → telemetry → checkpoints and artifacts
Model Export: SafeTensors / .pkl / ONNX / GGUF → Ollama / ONNX Runtime / local artifact downloads
Dashboard: Flask API ↔ Phase-1/2/3 controls ↔ real-time updates ↔ 3D visualizations ↔ system monitoring
🚀 Built for Production-Ready LLM Training
Architecture, Design and Development by Franz Ayestaran / Enhanced Pair Programming with Claude Code & OpenAI GPT