Hey, | Welcome to this week's edition of AlphaSignal.
Whether you are a researcher, engineer, developer, or data scientist, our summaries ensure you're always up-to-date with the latest breakthroughs in AI.
Let's get into it!
Lior |
|
|
|
IN TODAY'S SIGNAL | Top News: Databricks releases Top Announcements: Claude Opus beats GPT-4, $50,000 Gemini Hackathon, and Sora Pytorch Tip: num_workers and pin_memory
Read Time: 4 min 17 sec |
|
|
|
TOP NEWS | Databricks Releases an Open LLM that Beats Grok-1, LLama2 70B, and Mixtral |  | What's New? | Databricks just announced DBRX, a large language model (LLM) with advanced capabilities in language and code understanding. Designed with a mixture-of-experts architecture, this model aims to redefine performance standards in AI.
Performance Benchmarks: DBRX surpasses traditional open LLMs and specialized code models, setting new highs in language comprehension and coding tasks. - Leads on Hugging Face Open LLM Leaderboard and Databricks Gauntlet.
- Scores 70.1% on HumanEval (programming benchmark).
- Achieves 73.7% in MMLU (language understanding benchmark).
Model Specs: - Total parameters: 132 billion
- Active parameters: 32 billion
- Context length: Up to 32,000 tokens
- Training data: 12 trillion tokens
Efficiency: The MoE design significantly enhances DBRX's training and inference efficiency. - Nearly 4x more compute-efficient than prior models.
- Inference speed up to 2x faster than comparable models.
Availability: DBRX is accessible for both Databricks customers and the open community.
- Available under a commercial-friendly license.
- Model weights are hosted on Hugging Face.
| | Try dbrx |
|
|
|
 | Speech is the world’s most untapped resource. Let’s change that. | Speech is the purest form of communication. It’s what differentiates humans from other species. And gives us the ability to interact with and truly understand one another.
But speech is unstructured and often underutilized.
Speech Intelligence by Speechmatics changes that. It makes speech the biggest driver of value for your business.
Combining highly accurate automatic speech recognition (ASR) with the latest breakthroughs in AI and Large Language Models, Speechmatics interprets, analyzes, and understands over half the world’s population, so you can:
| | Try now (free) | partner with us |
|
|
|
TOP REPOS | --------------------- Anthropic's Claude 3 Opus has overtaken OpenAI's GPT-4 in performance, leading the Elo ratings on Chatbot Arena. This is the first model to surpass GPT-4 since its launch a year ago. However, Claude 3 Opus is more expensive, costing $15 for input and $75 for output per million tokens, compared to GPT-4’s $10 and $30. | | --------------------- Google is hosting a virtual hackathon focused on building with their generative AI, Gemini, offering a $50,000 prize. Participants can work remotely, utilizing Gemini's strengths in text summarization, image information extraction, and email drafting. The hackathon aims to apply Gemini in real-world scenarios like daily commutes and annual events. Participants can learn coding through interaction with Gemini, with a deadline set for the coming weeks. | | --------------------- OpenAI's Sora model, announced a month ago, has been refined based on feedback from the creative community. This collaboration has improved Sora's surreal and realistic visual generation capabilities, aiding artists, designers, and filmmakers. Specifically, the model contributed to the creation of "Air Head," the first short film based on the model. | |
| |
|
|
Maximize Your Productivity Through Strategic LLM Prompting | The industry leader in executive AI and Generative AI training, AI Product Institute, released the secrets this week to maximize people's productivity at work through strategic LLM prompting.
They shared a preview of the 15-hour advanced prompting training program for free.
Watch Here ↗️ |
|
|
|
PYTORCH TIP | num_workers and pin_memory | To speed up your data processing in PyTorch, use the torch.utils.data.DataLoader with num_workers and pin_memory parameters.
Setting num_workers to a value greater than 0 allows data loading and augmentation to happen in parallel with model training, using separate worker subprocesses.
Adjust num_workers based on your system's specifications and data location for optimal performance. For GPU-based training, set pin_memory=True. This makes DataLoader allocate data in pinned (page-locked) memory, facilitating faster data transfer to the GPU. | from torch.utils.data import DataLoader, Dataset
# Example dataset class MyDataset(Dataset): def __len__(self): # Example size return 1000
def __getitem__(self, index): # Returning data and a label as an example return (index, index % 2)
# Creating a DataLoader instance dataset = MyDataset()
data_loader = DataLoader(dataset, batch_size=32, num_workers=4, pin_memory=True)
|
|
|
|
TOP NEWS | Stability Releases Code Instruct 3B |  | What's New? | Stability AI just unveiled Stable Code Instruct 3B, the instruction fine-tuned version of its Stable Code 3B model. The model can now be prompted with natural language to generate code, or solve math problems, along with other software development related tasks.
Performance: - Outperforming comparable models : According to Stability AI's analysis, Stable Code Instruct 3b performs better than Codellama 7B Instruct and DeepSeek-Coder Instruct 1.3B in various code related tasks.
- Languages supported: Stable Code Instruct 3B was mainly trained on the most popular programming languages such as Python, Javascript, Java, C, C++, and Go, with some training exposure to SQL, PHP and Rust. It is still capable of generating code in other languages, showcasing strong generalization capabilities.
- Capabilities: The model can perform code generation, Fill in the Middle tasks, database queries, code translation, explanation and creation, thanks to its instruction fine-tuning.
Use it now: Stable Code Instruct 3B is now accessible with a Stability AI Membership for commercial use. The weights and code are also available for download on the Hugging Face hub. | | Try now |
|
|
|
| | Ghita is an editor at AlphaSignal and AI Engineer at Clear Ventures. Experienced in Computer Vision and Language models, she holds 2 masters in Applied Mathematics. | | Thank You. |
| | Want to promote your company, product, job, or event to 175,000+ AI researchers and engineers? You can reach out here. | |
|
|
|
|