Local ML Telemetry Monitor.
Visualized and Explained.

ModelSight is a visual dashboard extension that hooks into local PyTorch or TensorFlow training loops. Stream training metrics, detect overfitting alerts, and explain traceback errors in plain English—completely locally.

Experience ModelSight Live

Click the simulation buttons below to start a live mockup training run, inspect overfitting triggers, or paste tracebacks to test the diagnostics layer.

ModelSight Telemetry ×
train_resnet.py ~2.4 KB
                    
                  
Status: Idle
Active: train_resnet.py
Current Loss
-
Accuracy
-
Learning Rate
-
GPU Usage
-
Tesla T4 Temp: -
RAM Usage
-
Overfitting Risk
Low

Interactive Hyperparameter Tuning

Simulation Progress Calculating ETA...
0% Complete Epochs 0/30

Telemetry Curves

Traceback Diagnostics Sandbox

Paste a PyTorch/TensorFlow error stack to explain the exception locally.

Run History Logs

  • No training logs recorded. Run a simulation to start tracking runs.

Checkpoints & Events

  • No timeline logs recorded yet.

Dataset Metadata

ImageNet_Sample_Train
Total Samples
50,000
Feature Vector Shape
[3, 224, 224]

Class Counts

Class Label Samples Percentage
0 - Airplane5,00010%
1 - Automobile5,00010%
2 - Bird5,00010%
3 - Cat5,00010%
4 - Deer5,00010%

Label Distribution Chart

Balanced Label Set (10% per class)

Run Comparison Matrix

Inspect validation losses, hyperparameters, and run status side-by-side.

Parameter / Result Active Run (Simulated) Baseline Run (Run_082) Test Run (Run_081)
Epochs-30/3012/30 (Crashed)
Learning Rate-0.0010.005
Batch Size323264
OptimizersAdamWAdamWSGD
Final Loss-0.1250.542
Accuracy-94.8%81.2%
Risk Assessment-OptimalHigh Skew (NaN)
Python Exec stdout stdout
Welcome to ModelSight. Run a simulation to start streaming Python logs here...
Telemetry Visualizer Features

ModelSight delivers high-resolution analytics and diagnostics directly inside your text editor workspace.

📈

Real-time Streaming

Streams loss, accuracy, perplexity, and learning rates with low-overhead JSON sockets in real-time.

🛡️

Overfitting Detection

Flags divergence alerts early if validation loss curves separate from training gradients.

📟

Hardware Monitors

Monitors hardware constraints including GPU power, core temperatures, VRAM, and host RAM.

🔬

Traceback Diagnostician

Decodes Python tracebacks into descriptive error summaries and step-by-step remediation checklists.

Offline Diagnostics Library

ModelSight parses core ML exceptions offline with rule-based heuristics. Examples of handled issues:

CUDA Out of Memory (OOM)

Triggered when tensor sizes or batch weights exceed physical VRAM. Remediation suggests gradient accumulation or batch size reductions.

Dimension / Shape Mismatches

Identifies matrix multiplication shape conflicts or classification target indices mismatch (e.g. 1-indexed targets instead of 0-indexed targets).

Numerical Instability (NaN Loss)

Detects exploding gradients or division by zero in custom losses, recommending learning rate decays and gradient clipping.

Install ModelSight Extension

Streamline your deep learning feedback loop. Get the extension inside VS Code.

VS Code Marketplace
ext install CODExGAMERZ.modelsight