CALM is a result of a collaboration between @convai-uiuc.bsky.social and #Oumi.

Special thanks for the great team work, it would not be possible without Jeremiah Greer, Akul Datta, Ze Yang, William Zeng, Oussama Elachqar, Manos Koukoumidis, @dilekh.bsky.social, and @gokhantur.bsky.social.

We are making everything open-source with open models, open data, open checkpoints!

📄Arxiv: arxiv.org/abs/2502.08820
💻 Code: github.com/oumi-ai/oumi...
🤗 Models: huggingface.co/collections/...
🤗 Dataset: huggingface.co/datasets/uiu...

#ConversationalAgents #LLMs #Agents #OpenSourceAI #NLProc

How does the CALM model family perform?

✅ Outperforms GPT-4o & other top domain-specific models on:
📌 MultiWOZ 2.4 (TOD)
📌 BFCL V3 (Function Calling)
📌 API-Bank (Function Calling)
Achieving top zero-shot scores not in one but across all benchmarks!

🔥 Trained on CALM-IT, our unified dataset blending multi-turn ReAct style TOD & complex API use, trained using the Oumi AI platform in partnership with #Oumi and #TogetherAI.

📊 Models: CALM 8B, CALM 70B, CALM-405B trained from Llama model series

Most models struggle with either long-term conversations and dialogue state tracking (TOD) or function-calling (LA).

CALM (Conversational Agentic Language Model) bridges this gap! 💡

🦍Spoiler: CALM 405B is the largest open model in BFCL V3 Leaderboard ranking #7, surpassing many proprietary models.