Guide: Running Gemma 4 on a Tesla T4 GPU via Google Cloud with vLLM

·1 views

A developer has published a step-by-step guide for deploying Google's Gemma 4 2B model on a Tesla T4 GPU using Google Cloud's Compute Engine. The setup uses a minimal n1-standard-2 VM with 2 vCPUs, 7.8 GB RAM, and a T4 GPU running on Debian 13 with no pre-installed NVIDIA drivers. The software stack includes vLLM 0.29.0 and PyTorch 2.13.0, and takes approximately 362 seconds from script launch to a healthy inference endpoint. A companion shell script called vllm-t4 manages host-level state, while a suite of Python MCP tools handles deployment management. The guide also covers GPU-specific cloud configuration requirements, such as setting the maintenance policy to TERMINATE since live migration is unsupported for GPU-attached instances.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

JavaScript Operators Explained: Assignment, Arithmetic, and Beyond

JavaScript operators are symbols or keywords that instruct the computer to perform specific operations on values or variables, known as operands. They are broadly classified by functionality — including assignment, arithmetic, comparison, and logical operators — and by the number of operands they take. Assignment operators store values in variables, with shorthand forms like +=, -=, and *= combining assignment with arithmetic. Arithmetic operators handle mathematical tasks such as addition, subtraction, and multiplication, with the + operator also capable of concatenating strings. Notably, most arithmetic operators auto-convert string operands to numbers, except when the values are non-numeric, in which case the result is NaN.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How a Missing readOnly Flag Let JPA Silently Corrupt Production Data

A junior developer built a read-only internal API that unexpectedly corrupted production data due to a misunderstood JPA behavior. The developer used the @Transactional annotation while retrieving entities and modifying their fields temporarily to store intermediate processing state, without calling any explicit save or update method. Because the entities remained in a managed state within an active transaction, Hibernate's dirty-checking mechanism detected the field mutations and automatically issued UPDATE statements against the database at commit time. The fix was straightforward: replacing @Transactional with @Transactional(readOnly = true) instructs Hibernate to bypass dirty checking and flushing for that transaction context. The incident highlights how Hibernate's abstraction layer can produce serious side effects when developers lack a deep understanding of its lifecycle and persistence mechanisms.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

WebForms Core Lets TypeScript Run the Same Code on Server and Browser

WebForms Core is a framework that enables TypeScript to function as an isomorphic WebForms Commander, meaning the same class can run on both the server side and in the browser. The system is split into two parts: the Commander, which generates commands using WebForms classes, and the Executor, which is the front-end library WebFormsJS that carries out those commands in the browser. Developers interact with WebForms class functions rather than manipulating the DOM directly, keeping the programming model consistent across environments. The TypeScript package is available via npm under the name webformscore-ts and supports front-end, server-side, and WebAssembly contexts. As a practical example, the framework can be used to build a button that continuously changes a webpage's background to a randomly generated RGB color at 500-millisecond intervals.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Study Finds Nine Major AI Services Hide Free Plan Message Limits From Users

A review of nine popular AI platforms — including ChatGPT, Gemini, Claude, and Grok — found that none publicly discloses a numeric cap on free-tier chat messages. Researchers preserved and timestamped 70 official pages from these services as of September 22, 2026, tying every documented value to its source. While companies maintain pricing pages and help centers, the relevant usage details are scattered, often undated, and subject to change without clear notice — one value changed during the research itself. Limits are typically described in vague terms such as "generous limits" or "practically unlimited," leaving users unable to anticipate when access will be cut off. The findings make no service recommendations but aim to document what each company discloses and, crucially, what it omits.

0 comments Read more at DEV Community