@QingQ77: Training a 0.1B end-to-end omnimodal model from scratch. A single set of weights handles text, speech, and image inputs, while outputting text and streaming speech. https://github.com/jingyaogong/minimind-o… MiniMind-O is an omnimodal model with only 0.1B parameters…
Summary
MiniMind-O has released an end-to-end omnimodal model with only 0.1B parameters, supporting text, speech, and image inputs as well as streaming speech output. The project opensources the code, weights, training data, and technical report, emphasizing that both training and inference can be performed quickly on standard GPUs.
View Cached Full Text
Cached at: 05/09/26, 04:10 PM
“Great truths are simple”
Similar Articles
@vintcessun: Pretraining can be this cost-effective? Train a usable 1B base model from scratch for ~$1000, slashing compute and data by hundreds of times. The key isn't brute-force compute, but hierarchical recursive architecture plus latent space reasoning, combined with PrefixLM packing and FA3 to maximize efficiency. Sounds insane, but the paper and code are open-sourced.
HRM-Text released a 1B-parameter base model, claiming it can be pretrained from scratch for only ~$1000, reducing compute and data volume by hundreds of times. It employs efficient techniques such as hierarchical recursive architecture, latent space reasoning, and PrefixLM packing. The paper and code are open-sourced.
@seclink: This 12-billion-parameter model uses a unified Transformer architecture to efficiently handle raw multimodal inputs. It requires only 16GB of RAM to run, making it a perfect fit for devices like the MacBook Pro. It excels in various benchmarks, such as achieving 78.8% on GPQA Diamond and...
A 12-billion-parameter multimodal model has been released as open source. It features a unified Transformer architecture and requires only 16GB of RAM to run. It performs exceptionally well across multiple benchmarks, supports a 256K context window, and works with over 140 languages.
@indigox: Finally, the US also has a fully open-weight large model, after the failure of Llama... Thinking Machines' first OpenWeight Model - Inkling, trained entirely from scratch on GB300s, with 975 billion parameters, 41 billion active...
Thinking Machines has released Inkling, a fully open-weight multimodal large model with 975 billion parameters, supporting text, images, and audio. It has 41 billion active parameters and a 1 million token context window, and can be fine-tuned on Tinker.
@FeitengLi: A 99M parameter TTS runs on CPU, faster than a 2B model on A100. Supertone's newly open-sourced supertonic-3 with ONNX Runtime, fully local, can run in browser, on phone, and even on Raspberry Pi.
Supertone released Supertonic 3, an open-source TTS model with 99M parameters that runs faster on CPU than a 2B model on A100, supporting 31 languages and ONNX Runtime for fully local inference.
@Phoenixyin13: This is the brand new best open-source model from the US. Thinking Machines Lab directly released a 975B parameter MoE multimodal giant, fully open-source under Apache-2.0! On standard benchmarks, it even outperforms NVIDIA's Nemotron strong model. Its core...
Thinking Machines Lab released a new 975B parameter MoE multimodal open-source model under Apache-2.0 license, surpassing NVIDIA's Nemotron on standard benchmarks, supporting text, image, and audio modalities with only 41B activated parameters, efficient and deployment-friendly.