OLMo 2 32B is the most capable and largest model in the OLMo 2 family. High-performance, fully open LLM. Outperforms GPT-3.5 Turbo & GPT-4o mini. Trained efficiently on 6T tokens.
Sharing OLMo 2 32B, a new language model from the Allen Institute for AI (Ai2). This model is notable for being fully open, with the data, code, and weights all publicly available.
Key features:
🎛️ 32B parameters, placing it in a mid-size range for current LLMs. 🧠 Trained on 6 trillion tokens, using a two-stage pretraining process. 📊 AI2 reports performance comparable to GPT-3.5 Turbo and GPT-4o mini on several benchmarks. ⚙️ The training process incorporated techniques like Group Relative Policy Optimization (GRPO) and RLVR. 🔓 The model, training data, and code are available under an Apache 2.0 license.
This release provides researchers and developers with full access to a capable LLM and its training pipeline. You have to admire Ai2's commitment to transparency here.
Replies
Hi everyone!
Sharing OLMo 2 32B, a new language model from the Allen Institute for AI (Ai2). This model is notable for being fully open, with the data, code, and weights all publicly available.
Key features:
🎛️ 32B parameters, placing it in a mid-size range for current LLMs.
🧠 Trained on 6 trillion tokens, using a two-stage pretraining process.
📊 AI2 reports performance comparable to GPT-3.5 Turbo and GPT-4o mini on several benchmarks.
⚙️ The training process incorporated techniques like Group Relative Policy Optimization (GRPO) and RLVR.
🔓 The model, training data, and code are available under an Apache 2.0 license.
This release provides researchers and developers with full access to a capable LLM and its training pipeline. You have to admire Ai2's commitment to transparency here.
Try it by yourself in Ai2's Playground.