vectara/hallucination_evaluation_model
LoRA-style (fewer layers) vs google/flan-t5-base. vectara/hallucination_evaluation_model matches part of google/flan-t5-base (111 of its 282 tensors; the rest of google/flan-t5-base has no counterpart). On the shared part: Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of q, v changed, in 12 of 12 layers; norms, biases and embeddings are untouched.
Tensors changed: Attention 24/49, Norms 0/25, other 0/37
1 other repos hold byte-for-byte identical weight files.