Top-Trending LLMs Over the Last Week. Week #12.

In this post, we'd like to share the list of top trending models that caught people's attention in the AI world over the last week. We ranked them by how many times they were downloaded and liked, based on information from Hugging Face and LLM Explorer.

1. C4ai Command-R V01, developed by Cohere For AI, leads the pack with its 35 billion parameters, excelling in reasoning, summarization, and question answering across multiple languages. It's designed for devices with limited processing power, making advanced AI tasks more accessible.

Community Insights

AI experts like the C4AI Command-R model for how well it does different tasks like reasoning and answering questions in many languages. They're especially impressed with its ability to work with multiple languages and generate answers based on document information. Although it's not for commercial use without permission, some believe smaller groups might still be able to use it. People also talk about how this model could lead to new advancements in AI.

2. Hermes 2 Pro on Mistral 7B (by NousResearch) is a new, upgraded version of Nous Hermes 2, optimized with an updated OpenHermes 2.5 Dataset and a new Function Calling and JSON Mode dataset. It excels in general tasks, conversations, function calls, and JSON outputs, all while being efficient on basic devices. It also features a special prompt and structure for easy function calls.

User-Driven Feedback

Users appreciate that the developer has quantized the model and made both versions available. They find its special skills in handling tasks and making JSON outputs useful, but some say it's not as good as the older ones. Even with mixed opinions on how well it works overall, there's excitement about what it can do differently.

3. StarChat2 15B V0.1 from HuggingFace H4 is a GPT-like coding assistant that supports English and over 600 programming languages. While it's great for chat and coding, users should be cautious of potential biases and the accuracy of its code outputs.

4. GemMoE Beta 1, developed by Crystalcareai (also known as Lucas Atkins, an active contributor to the world of AI), is a text generation model that stands out for its mixture of experts approach. This method combines different components to improve text generation, showing continuous development and refinement.

5. The 4-bit quantized version of C4AI Command-R, developed by Cohere and Cohere For AI, distinguishes itself by optimizing for devices with limited processing capability without compromising on its wide-ranging functionalities including reasoning, summarization, and multilingual question answering. Despite retaining the original model's robust 35 billion parameters and extensive language support, this quantized variant ensures greater accessibility and efficiency in use, making advanced AI tasks more approachable for a broader audience.

6. Cerebrum 1.0 7B from Aether AI excels in reasoning tasks with a native chain of thought approach, making it ideal for logical and scientific thinking, as well as general use as a language model.

7. Yi 9B 200K from 01-ai. The Yi series models by 01.AI are next-gen bilingual LLMs trained on a 3T multilingual corpus. The Yi-34B-Chat model ranks second on the AlpacaEval Leaderboard, outperforming giants like GPT-4 and Claude, while the Yi-34B leads in English and Chinese on various benchmarks, thanks to the collaborative efforts within the AI community. These models are designed for comprehensive language understanding and reasoning tasks.

8. Mistral Orpo Beta from KAIST AI. Mistral-ORPO-β (7B) improves the original Mistral-7B model with odds ratio preference optimization (ORPO), eliminating the need for a supervised fine-tuning phase. It excels in AlpacaEval and MT-Bench, outperforming models like Llama-2-Chat, and is fine-tuned on a clean UltraFeedback dataset. This model showcases enhanced alignment and response generation, making it efficient for various AI tasks.

9. Trendyol LLM 7B Chat V1.0, based on Mistral 7B and optimized with LoRA, is designed for conversational text in English and Turkish. It emphasizes the need for ethical use and human oversight. The developer is Trendyol Group.

10. Phi 2 Layla V1 Chatml (by l3utterfly) offers enhanced performance in multi-turn conversations and character impersonation, showing the versatility of models in embodying characters for various applications.

That's all for now. Want to know more? Check the LLM Explorer website anytime for the newest popular models. The list is always there on the homepage.

Stay tuned!

Next Week's Top LLMs →