`accelerate test` works with the default config… But the script above hangs for 10 minutes likely due to the NCCL timeout. Somewhere between these two log lines: INFO [core(.)py:74] Initializing a V1 LLM engine (v0.10.1.1) with config: model='unsloth/Qwen3-4B-Base'
INFO
LLMS
-
NCCL Timeout Issue with Qwen3 LLM Engine Initialization
By
–
-
Training Qwen3 Multi-GPU Setup with Accelerate Library
By
–
Anyone manage to get Qwen3 to train on multiple GPUs using `accelerate`? With a fresh install, I can't get even SFT to work with two 3090s.@UnslothAI The demo notebooks are nice, but a reproducible script known to work would be great, ideally with version dependencies pinned! https://t.co/8FkFpC3pdO
— Alex J. Champandard 🌱 (@alexjc) 10 septembre 2025Anyone manage to get Qwen3 to train on multiple GPUs using `accelerate`? With a fresh install, I can't get even SFT to work with two 3090s. @UnslothAI The demo notebooks are nice, but a reproducible script known to work would be great, ideally with version dependencies pinned!
-
Deep dive into understanding AI models explained
By
–
Amazing episode for anyone who wants to understand AI models at a deeper level. https://t.co/3mDrBA3fER
— Paul Roetzer (@paulroetzer) 10 septembre 2025Amazing episode for anyone who wants to understand AI models at a deeper level.
-
The Hidden Leverage in Better Prompt Structure
By
–
Wild to think how much leverage is hiding in better prompt structure.
-

Salesforce Leverages Open Source Models for Advanced AI Research Agent
By
–
Interesting example of how open weights models provide opportunities for innovation. Salesforce builds a strong deep research agent from OpenAI’s small open source model. Though open models development is dependent on the good will of OpenAI, Mistral & a few Chinese firms.
-

AI Wearables, Microsoft-Claude Partnership, and Claude Updates
By
–
Top stories in AI today: – Alterego’s “near-telepathic” AI wearable
– Microsoft eyes Claude for Office 365
– Create mini product photos with Nano Banana
– Claude gains file creation capabilities
– 4 new AI tools, community workflows, and more Read more: https://
therundown.ai/p/alterego-deb
uts-near-telepathic-ai-wearable
… -
GPT-4.5 Reasoning Capabilities and Model Provider Constraints
By
–
You're misunderstanding… GPT-4.5 is NOT a reasoning model, so we're not bound by any model provider limits. Also just a helpful tip, when asking an AI for an answer, try not to say "that's a fact" or similar, as it'll bias the model to agree blindly
-
AI needs better instructions, not better models
By
–
“AI doesn’t need better models—it needs better instructions.”
That’s the whole game right there. -
Native Format vs JSON Prompting for AI Model Training
By
–
Yep, when it’s trained in like that, it’s usually better to use the native format. I’m talking more about freeform JSON prompting here