Author

DeepHub IMBA

A must‑follow public account sharing practical AI insights. Follow now. internet + machine learning + big data + architecture = IMBA

117

Articles

Likes

818

Views

Comments

Latest from DeepHub IMBA

100 recent articles max

DeepHub IMBA

Aug 9, 2026 · Artificial Intelligence

Building ComputeShare: A Lightweight Multi‑Machine Distributed Training System with PyTorch

ComputeShare demonstrates how a centralized parameter‑server architecture, linear‑scaling learning‑rate adjustments, stale‑gradient protection, optional async SGD, gradient compression, and a universal dataset factory enable heterogeneous devices (CUDA, MPS, CPU) to collaboratively train models over LAN or the Internet, with benchmark results on MNIST, Fashion‑MNIST, and USPS.

ComputeShareDistributed TrainingPyTorch

0 likes · 11 min read

Building ComputeShare: A Lightweight Multi‑Machine Distributed Training System with PyTorch

DeepHub IMBA

Aug 8, 2026 · Artificial Intelligence

Why Parallel Loop Transformers Peak at Two Iterations – Insights from LoopCoder‑v2

The LoopCoder‑v2 study shows that Parallel Loop Transformers achieve their best code‑generation performance with two refinement loops, as additional loops increase memory cost without improving results and even cause performance degradation, a finding explained through detailed metric analysis and cost‑benefit reasoning.

Code GenerationG-SWAKL Divergence

0 likes · 14 min read

Why Parallel Loop Transformers Peak at Two Iterations – Insights from LoopCoder‑v2

DeepHub IMBA

Aug 7, 2026 · Big Data

Pandas vs Polars vs DuckDB: Benchmarking Performance on a 1.2M‑row CSV

A head‑to‑head benchmark on a 2.3 GB CSV (~1.2 million rows) shows Pandas exhausting memory, Polars completing the pipeline in 8.7 seconds with modest RAM, and DuckDB answering the same query in just 12 milliseconds, highlighting distinct trade‑offs for Python data processing.

CSVDuckDBPandas

0 likes · 11 min read

Pandas vs Polars vs DuckDB: Benchmarking Performance on a 1.2M‑row CSV

DeepHub IMBA

Aug 6, 2026 · Backend Development

10 Practical Python API Tips to Turn Scripts from Working to Robust

This article presents ten concrete Python techniques for handling APIs—automatic retries, response caching, environment‑based secrets, Pydantic schema validation, clean pagination, WebSocket streaming, rate‑limiting, concurrent requests, mock responses, and GraphQL—each illustrated with runnable code to make scripts more reliable and efficient.

APIGraphQLPydantic

0 likes · 8 min read

10 Practical Python API Tips to Turn Scripts from Working to Robust

DeepHub IMBA

Aug 5, 2026 · Artificial Intelligence

Designing Multi‑Model Systems: 4 Common Architecture Patterns and When to Use Them

The article examines multi‑model system design, presenting four core architecture patterns—sequential (pipeline, router), parallel (fan‑out, voting), hierarchical (planner‑executor, supervisor‑worker), and ensemble (weighted, consensus)—and discusses their trade‑offs, implementation details, and scenarios where multi‑model approaches add real value.

AIPipelinearchitecture patterns

0 likes · 11 min read

Designing Multi‑Model Systems: 4 Common Architecture Patterns and When to Use Them

DeepHub IMBA

Aug 4, 2026 · Fundamentals

7 Fast Pandas Vectorization Tricks: Arithmetic, Conditional Logic, and Group Aggregation

This article benchmarks seven common pandas tasks—arithmetic, conditional mapping, dictionary lookup, string cleaning, group aggregation, binning, and query/eval—showing how vectorized alternatives can speed up loops by up to 7,000× and offering practical guidance on when each technique delivers meaningful gains.

DataFrameNumPyPandas

0 likes · 12 min read

7 Fast Pandas Vectorization Tricks: Arithmetic, Conditional Logic, and Group Aggregation

DeepHub IMBA

Aug 3, 2026 · Artificial Intelligence

How Continuous Batching Boosts LLM Throughput by Up to 36×

The article explains continuous batching, a scheduling technique that re‑evaluates the batch at every inference iteration, eliminating the straggler problem of static batching, and shows how it, together with PagedAttention, can raise LLM service throughput dramatically while balancing latency and memory usage.

GPU utilizationLLM Inferencecontinuous batching

0 likes · 9 min read

How Continuous Batching Boosts LLM Throughput by Up to 36×

DeepHub IMBA

Aug 2, 2026 · Artificial Intelligence

Building a From‑Scratch LLM Training Framework: Full GRPO vs PPO vs DPO Comparison on GSM8K

The article presents a from‑scratch LLM training framework called grpo‑llm, implements GRPO with Trio rollout, FSDP and a C++ reward extension, and conducts a controlled experiment comparing GRPO, PPO and DPO on the GSM8K math‑reasoning benchmark, revealing why DPO outperforms the other two under sparse binary rewards.

DPOGRPOGSM8K

0 likes · 10 min read

Building a From‑Scratch LLM Training Framework: Full GRPO vs PPO vs DPO Comparison on GSM8K

DeepHub IMBA

Aug 1, 2026 · Artificial Intelligence

Estimating the GPU Count Needed to Train a Large Language Model

The article presents a practical scaling‑law based method to estimate the total FLOPs, GPU throughput, and required number of GPUs for training a large language model, showing how to compute these values from model parameters, token count, and target training time, and discusses approximation limits and useful tools.

AICompute EstimationGPU

0 likes · 9 min read

Estimating the GPU Count Needed to Train a Large Language Model

DeepHub IMBA

Jul 29, 2026 · User Experience Design

How to Write an Effective DESIGN.md So AI Generates Designs That Match Your Intent

Designers can use a plain‑text DESIGN.md file to convey visual identity, token‑based design decisions, and component specifications to AI tools such as Google Stitch, Claude Code or Cursor, ensuring the AI produces UI outputs that respect the original design system and reduce mismatches.

AI-assisted designDESIGN.mdUI generation

0 likes · 9 min read

How to Write an Effective DESIGN.md So AI Generates Designs That Match Your Intent