Ivy-Fake checkpoint missing lm_head.weight → unusable inference output

[a-poptsov]

Summary

The uploaded AI-Safeguard/Ivy-Fake checkpoint appears to be incomplete.
When loading the model with the standard Transformers Qwen2.5-VL pipeline, the checkpoint is missing lm_head.weight https://huggingface.co/AI-Safeguard/Ivy-Fake/tree/main, which causes the model to initialize a random LM head and produce nonsensical output.

As a result, the model cannot produce meaningful predictions via .generate().

Please upload a checkpoint that includes:

lm_head.weight