Has there been any replication work on less toy examples?

Let’s goooooo. Looking forward to this.

Is there any reason to use llava vision over llama vision at this point? Seems like all the metrics have switched over