Sign in Subscribe
Large Language Model

The Expanding Universe of Large Language Models

Large language models(LLM) have become quite a phenomenon in the last year! In this post, we will cover some of the popular LLMs. I will continue to update this LLM list as more LLMs(both closed and open source) are released over the year.

We will mainly categorize LLMs into closed-source and open-source.

Closed Source LLMs

  1. GPT-4: GPT-4 is the largest and most advanced large language model developed by OpenAI. GPT-4 has almost ~15% improved performance across different benchmarks as released by OpenAI. This model was released in 2023.
  2. GPT-3: Another version of the model created by OpenAI, with less complexity than GPT-4. GPT-3 was released by OpenAI in 2020. GPT-3 is a decoder-only model which uses attention-based techniques for modeling.
  3. Claude: An LLM developed by Anothropic, Claude is known for its sophisticated design.
  4. Gopher: This is a product of DeepMind's research and development efforts.
  5. Chinchilla: Google's team is responsible for this model's development.
  6. LaMDA: Also developed by Google, LaMDA is unique for its conversational capabilities.
  7. AlexaTM: This is Amazon's proprietary language model used in the Alexa virtual assistant.
  8. Cohere: Developed by Cohere, this model stands out for its specific applications.
  9. Bloomberg GPT: Bloomberg's custom-tailored LLM, further details can be found in their research paper here.
Evolution of GPT Models from OpenAI

Open Source LLMs

  1. falcon-40b-instruct
  2. guanaco-65b-merged
  3. 30B-Lazarus
  4. LLaMA Model
  5. LLaMA 2: https://ai.meta.com/llama/
  6. Free Willy 2: https://stability.ai/blog/freewilly-large-instruction-fine-tuned-models
  7. Persimmon-8B: Persimmon-8B is an open-source model from Adept AI
  8. Wizard LM: Family of LM trained for different tasks. Some of the LLMs include WizardCoder-Python-34B-V1.0 and WizardMath-70B-V1.0.
  9. Phi-1: This is a smaller LM with 1B parameters from Microsoft.
  10. Mistral-7b: This is 7B parameter model from Mistral
  11. TigerBot 70B Base: This is a 70B parameter model from TigerResearch.
  12. Mixtral MOE Model: Mixtral 8x7B, a high-quality sparse mixture of experts model (SMoE) with open weights.

There are many other open-source model listed here at HuggingFace leaderboard.

Read next