Alibaba Tongyi
7 articles tagged with this topic
Community Qwen3.8 Quantization Saves 30GB — Local LLM Bar Drops Again
Community dev agentionai ships a custom quantized Qwen3.8-Flash-Next, 20–30GB smaller than mainstream versions at comparable quality. Local LLM bar dr
Qwen 27B Runs 80-Step Agent on a Single GPU — Local Models Can Now Do Real Work
A Reddit test caught our eye: Qwen 27B on a consumer GPU made 80 autonomous tool calls from one prompt. The "cloud-only Agent" default is crumbling.
Qwen 27B Runs 10 Hours Solo on RTX 3090 — DeepSeek's Engine Decouples from Model
Developer ran Qwen 27B with DeepSeek's open harness on an RTX 3090 for 10 hours — no crash. The story isn't the benchmark. It's modularity.
Netizens strip Alibaba Qwen's refusals — Community fork 3.8 quietly updates
An unofficial HF account quietly dropped an 'abliterated' Qwen fork that strips refusal behavior. Mainstream Chinese media didn't cover it.
Qwen 3.8 Spotted on GitHub — Alibaba's Open-Source Cadence Outpaces Rivals
Traces of Qwen 3.8 35B-A3B surfaced in Alibaba Tongyi's ms-swift framework on GitHub, hinting at the next open-source drop from a leading model family
Open-Source 27B Renders One Piece in 7 Min — Local AI Coding Works, Still Slow
Reddit user runs Alibaba's Qwen 27B locally, generates a One Piece ship SVG after 7 minutes of AI thinking — task needing GPT-4 or Claude a year ago.
16GB Consumer GPUs Run Qwen 14B at 44 Tokens/Second — Local AI Gets Practical
Reddit user benchmarks Alibaba's Qwen2.5-14B on a Nvidia 5060Ti 16GB at 44 chars/sec — local AI just crossed into consumer hardware territory.