Valeria Solovyova

I train models and watch them learn to interpret the world. I share experiments, datasets, unexpected outcomes, and the messy beauty of applied math.

Joined on Feb 19, 2026

Valeria Solovyova

Mar 18

ICML's LLM Policy Breach: Addressing Fairness and Enforceability in Academic Review Processes

#ai #ethics #enforcement #fairness

13 min read

Valeria Solovyova

Mar 17

Fine-Tuning LLMs on Apple Silicon: New Tools Enable Local Prototyping, Reducing Cloud GPU Dependency

#llm #applesilicon #finetuning #mlx

19 min read

Valeria Solovyova

Mar 17

Understanding Primitive Layers in Small Language Models: Distinguishing Layer 0a and 0b, and Their Evolution with Scale

#ai #languagemodels #primitives #scaling

12 min read

Valeria Solovyova

Mar 16

Enhancing F1 Race Strategy Predictions with Physics Simulation and ML Residual Correction for Improved Accuracy

#f1 #machinelearning #physics #prediction

18 min read

Valeria Solovyova

Mar 16

Addressing Label Leakage in Machine Learning Datasets: Strategies for Valid Model Training and Evaluation

#machinelearning #labelleakage #preflight #dataintegrity

19 min read

Valeria Solovyova

Mar 15

Optimal Setup for ML Development on Windows 11 with RTX 5080: WSL2 vs. Dual Boot for Convenience and Performance

#machinelearning #windows11 #rtx5080 #wsl2

14 min read

Valeria Solovyova

Mar 13

Addressing LLM Benchmarking Obsolescence: Strategies for Timely and Relevant Model Evaluation

#llm #benchmarking #obsolescence #proprietary

13 min read

Valeria Solovyova

Mar 10

Addressing User Stress: Strategies for Sharing ARR Meta-Reviews in January 2026

#stress #community #reviews #deadlines

15 min read

Valeria Solovyova

Mar 5

Anonymous User Claims Proof of d^2 Complexity for Attention Mechanisms, Challenging Transformer Optimization

#transformers #attention #optimization #complexity

10 min read

Valeria Solovyova

Mar 4

ComfyUI Instability Prompts Search for Stable, User-Friendly Alternative Solutions

#comfyui #mindstudio #stability #workflow

12 min read

Valeria Solovyova

Mar 4

Addressing Neptune's Limitations: Developing an Efficient, User-Friendly ML Experiment Tracking Tool

#machinelearning #experimenttracking #goodseed #neptune

20 min read

Valeria Solovyova

Mar 3

PEP 827 Unveiled: How Python's New Type Manipulation Features Impact Ecosystem Usability and Adoption

#python #typing #pep827 #typevartuple

12 min read

Valeria Solovyova

Mar 2

Bridging the Semantic Gap in Neural Network Execution and Verification for Safety-Critical Systems

#neuralnetworks #verification #safetycritical #semanticgap

8 min read

Valeria Solovyova

Mar 1

Advancing Tiny Transformers: Achieving 100% Accuracy in 10-Digit Addition with Sub-100 Parameter Models Using Digit Tokenization

#ai #transformers #efficiency #tokenization

16 min read

DEV Community

Valeria Solovyova

Badges

Writing Debut

ICML's LLM Policy Breach: Addressing Fairness and Enforceability in Academic Review Processes

Fine-Tuning LLMs on Apple Silicon: New Tools Enable Local Prototyping, Reducing Cloud GPU Dependency

Understanding Primitive Layers in Small Language Models: Distinguishing Layer 0a and 0b, and Their Evolution with Scale

Enhancing F1 Race Strategy Predictions with Physics Simulation and ML Residual Correction for Improved Accuracy

Addressing Label Leakage in Machine Learning Datasets: Strategies for Valid Model Training and Evaluation

Optimal Setup for ML Development on Windows 11 with RTX 5080: WSL2 vs. Dual Boot for Convenience and Performance

Addressing LLM Benchmarking Obsolescence: Strategies for Timely and Relevant Model Evaluation

Addressing User Stress: Strategies for Sharing ARR Meta-Reviews in January 2026

Anonymous User Claims Proof of d^2 Complexity for Attention Mechanisms, Challenging Transformer Optimization

ComfyUI Instability Prompts Search for Stable, User-Friendly Alternative Solutions

Addressing Neptune's Limitations: Developing an Efficient, User-Friendly ML Experiment Tracking Tool

PEP 827 Unveiled: How Python's New Type Manipulation Features Impact Ecosystem Usability and Adoption

Bridging the Semantic Gap in Neural Network Execution and Verification for Safety-Critical Systems

Advancing Tiny Transformers: Achieving 100% Accuracy in 10-Digit Addition with Sub-100 Parameter Models Using Digit Tokenization