Topic

#Qwen

15 articles on Qwen — news, releases, guides and analysis from the SourceFeed engine.

Google's 4.7x Qwen 3.5 Speedup Is a Sharding Story

Two KV heads and 512 experts pushed Ironwood onto DP+EP, the same topology vLLM now recommends for GPUs.

The 2.4T flagship got the headlines, but the Apache-licensed multimodal 27B is the release developers can actually use.

Alibaba's Apache 2.0 dense release targets single-GPU agentic work — and its license split signals where open weights are heading.

The #1 slot flipped on reload — the model's 64-turn work ethic and imminent open-weights drop matter more.

Wire Qwen2.5-Coder through Ollama and Continue for private, offline autocomplete and chat in VS Code.

Swiftlet streams sparse Qwen experts off disk, fitting an 80B model into 4.3 GB of Mac RAM.

Swiftlet streams MoE experts from SSD, fitting an 80B Qwen in 4.3GB of RAM — and a 35B on an iPhone.

The public git trace matters more than the model — and the orchestration harness is the part worth stealing.

A free, OpenAI-compatible endpoint from Europe's cheapest host says more about inference margins than any benchmark.

Open weights, real Runpod jobs, and held-out transfer make meta-RL a forkable template, not theater.

Why brute-forcing frontier models onto local hardware fails, and how clever pipeline decomposition offers a sustainable path forward.

The dense 27B model delivers frontier-class intelligence on local hardware without the compromises of lightweight mixtures of experts.

Qwen-AgentWorld replaces heavy external simulators with a native language world model to train and evaluate AI agents.

Developers are swapping frontier models for local setups, but managing context caching and agent loops requires careful tuning.

A GitHub issue claims Rio de Janeiro's municipal LLM is just a 60/40 blend of Nex-N2 and Qwen.