SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

MiMo V2.6 Flash Run Locally on Two Mac Studios Trails Cloud Version in Benchmark

0
·9 views

A developer ran Xiaomi's MiMo V2.6 Flash, a 309-billion-parameter model, locally by splitting its layers across two Mac Studios linked by a Thunderbolt 5 cable. On a payments-app build benchmark, the local setup scored 0.6605, compared with 0.7909 for the same model on Xiaomi's cloud endpoint. Most of the gap came from a benchmark ceiling triggered by failed 3D checks, since the two runs earned nearly identical scores before that cap was applied. Adding vision support removed several failing checks, while the cloud run was stronger on 3D pixel work. The author presents this as a single-task field report rather than a broad comparison.

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)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Developer Shares Method for Fair, Verifiable Giveaway Draws Using Python and Node

A marketing team needed to pick ten winners from a webinar registration list of over 8,000 people, with the draw kept unpredictable until it happened. The author resolved the tension between secrecy and reproducibility by generating a random seed from the operating system's entropy pool at draw time and publishing it only after winners were announced. The Python script removes duplicate registrations by email before sampling without replacement. The article also covers a Node.js version that uses a cryptographic random integer function to avoid skewed results. Readers are shown how anyone can later replay the draw with the recorded seed to confirm the winners.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Weaviate's anonymous-access default raises authorization concerns for production deployments

Weaviate, a vector database with REST and GraphQL interfaces, enables anonymous access by default in Docker, a setting its documentation describes as meant for development and evaluation. Deployments that keep this default while holding production data create an authorization gap, since unauthenticated requests run as an anonymous user. A count-only ZoomEye search on October 29, 2026 returned 59 title matches for Weaviate and none for the application field, but these are matching observations rather than confirmed exposed instances. The analysis recommends enabling API key or OIDC authentication alongside role-based access control scoped to specific collections and tenants. It also advises testing anonymous and low-privilege requests across read, write, delete, backup, and schema operations to verify access restrictions.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Routing website contact form enquiries with an LLM restricted to a fixed category list

The author built an n8n workflow that uses an LLM to sort contact form submissions into predefined categories such as sales, support, partnership, job application, spam, or other. The model may only reply with one of the fixed category ids or null, since free-form answers would break spreadsheet filtering. Any reply that is not an exact match, as well as API errors, is flagged as needs_human with the reason recorded, and every valid enquiry is still logged so none are lost. Form text containing curly braces is stored as sent, avoiding template-syntax problems in the LangChain nodes. The workflow passed 17 local checks against stand-in APIs, but the author has not tested it with a real model or sheet, and it lacks authentication and deduplication for public use.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Android Studio Adds Support for Outside AI Coding Agents; Android CLI Plugins Arrive

At the Android Dev Summit on October 28, Google announced that Android Studio now hosts over 40 ACP-compatible coding agents through its Bring Your Own Agent marketplace. The feature is stable in the Rabbit 2 release, and named agents include Claude Agent, Codex from OpenAI, Junie, and Google Antigravity. Developers can install several agents and switch between them, with the IDE supplying project structure, build, and emulator context. Separately, Google's Android CLI is now available as an official plugin for Claude Code, Codex, and Antigravity, allowing agents to build and debug Android apps without the IDE running.

0 comments Read more at DEV Community