How do you know that cosmo-xl outperforms Blender-Bot?
"outperforming previous best-performing agents (e.g., GODEL, BlenderBot, DialoGPT) on both in- and out-of-domain datasets (e.g., DailyDialog, BlendedSkillTalk)"
Hello, we compare COSMO and BlenderBot by running human evaluations on DailyDialog (i.e., an unseen dataset for both models) and BlendedSkillTalk (i.e., the training dataset of BlenderBot). Evaluation results show that human judges prefer the responses from COSMO over the responses from BlenderBot on both datasets. For more details, we gently guide you to our recent paper (Section 5).
Do you have any info on how much GPU ram is required to run the cosmo-xl? I might setup cosmo-xl up on www.justlearn.com to power the chatbots
Can I run fine-tuning on the model using my own data?
cosmo-xl can run on 24GB GPU ram. You can freely fine-tune cosmo on other datasets :)
"human judges prefer the responses from COSMO over the responses from BlenderBot on both datasets". Do you use BlenderBot3 with search-server?
Do you have any plans to add search-server function similar to BlenderBot3? It works great, if I ask "What happens in year 1011?". Then AI will do a search similar to "Bing Web Search".
Do you have any plans to add cosmo-xxl?
How can I add chat history?
Chat history
Philip: Hi
Jakob: Hi
Philip: How are you?
Jakob: I am good.
Philip: I like Python.
Jakob: Cool.
Like this?
dialogue = [
"Philip: Hi",
"Jakob: Hi",
"Philip: How are you?",
"Jakob: I am good.",
"Philip: I like Python."
]
Can the cosmo-xl send emojis or smiley like :-) ? I can't get this to return any emoji or smiley.
dialogue = [
"Can you send emojis? 🍵🥂☕😋👌✨"
]