Models & Labs
Fine-Tuning NVIDIA Cosmos Predict 2.5 with LoRA/DoRA
Hugging Face Bloghigh confidence
Why it matters
- →LoRA and DoRA enable efficient fine-tuning on a single GPU, reducing costs and memory usage.
- →Synthetic trajectory generation offers a scalable alternative to expensive real-robot data collection.
- →This approach prevents catastrophic forgetting, maintaining the model's general knowledge.
NVIDIA Cosmos Predict 2.5, a large-scale world model, is being fine-tuned using LoRA and DoRA techniques to generate synthetic robot trajectories. This method allows for efficient fine-tuning on a single GPU, reducing memory requirements and preventing the loss of general knowledge. By using small trainable adapter modules, developers can adapt the model to specific domains like robot manipulation. This advancement offers a scalable alternative to collecting real-robot data, potentially accelerating progress in robot learning tasks.
Read originalMore from Hugging Face Blog
© Hugging Face BlogModels & Labsmodels
Nemotron-Labs Introduces Diffusion Language Models
Nemotron-Labs has unveiled a new family of diffusion language models that promise to revolutionize text generation by allowing multiple tokens to be generated in parallel. This approach contrasts with traditional autoregressive models that generate text one token at a time, potentially improving performance and accuracy. The models, available in various scales, offer a flexible design that supports three generation modes, including a novel self-speculation mode that combines diffusion drafting with autoregressive verification. This innovation could significantly enhance the efficiency of text generation tasks, making it a compelling option for developers seeking faster and more accurate AI solutions.
Hugging Face Blog
© Hugging Face BlogResearchresearch
Specialized AI Models Outperform Larger Counterparts
In a surprising turn for AI procurement strategies, a specialized 3-billion-parameter model has outperformed larger commercial models in a specific enterprise domain, demonstrating that specialization can trump scale. This model excelled in Brazilian Portuguese OCR tasks, achieving higher quality at a fraction of the cost compared to leading frontier APIs. The findings challenge the prevailing assumption that larger models are inherently superior, highlighting the importance of aligning a model's training history with its deployment task. This shift suggests that enterprises might benefit from focusing on specialized models tailored to their specific needs rather than defaulting to larger, more generalized models.
Hugging Face Blog
More in Models & Labs
Models & Labsmodels
llama.cpp b9297 release enhances tensor support
The b9297 release of llama.cpp brings a notable enhancement with the introduction of NVFP4 MTP scale tensors, boosting its tensor processing capabilities. This update also integrates Qwen3.5 MTP tensors, which improves performance across a spectrum of hardware configurations, including Apple Silicon, Vulkan, and ROCm on Ubuntu, as well as CUDA on Windows. The release supports a wide array of architectures, from macOS to Linux and Windows, ensuring compatibility with both CPU and GPU setups. While there are no new model architectures, the inclusion of KleidiAI on Apple Silicon and ROCm 7.2 on Ubuntu highlights llama.cpp's commitment to optimizing for diverse environments. This update reinforces llama.cpp's role as a flexible inference runtime, catering to a broad range of hardware setups.
llama.cpp Releases
Models & Labsmodels
llama.cpp b9309 release fixes integer overflows
The b9309 release of llama.cpp tackles significant integer overflow issues in its perplexity calculations, co-authored by Stanisław Szymczyk. This update is vital for enhancing the accuracy and reliability of the model's performance metrics, which are crucial for developers. By resolving these overflows, the release ensures that users can depend on precise data outputs. This fix is a testament to the ongoing efforts to improve the tool's robustness, allowing developers to trust the integrity of their AI computations. While it might seem like a minor adjustment, it plays a critical role in maintaining the tool's reliability.
llama.cpp Releases
© The AI Daily BriefModels & Labsmodels
OpenAI Achieves Math Breakthrough
OpenAI has made a significant advancement in mathematical capabilities within its AI models.
The AI Daily Brief