Update README.md

README.md

@@ -2,12 +2,14 @@
 language:
 - en
 ---
-# entailment-verifier-xxl
+# nli-entailment-verifier-xxl
 
 ## Model description
-Entailment-verifier-xxl is based on [flan-t5-xxl model](https://huggingface.co/google/flan-t5-xxl) and finetuned with a ranking objective (rank the most supported hypothesis from a given pair of hypotheses for a given premise). Please refer to our paper [Are Machines Better at Complex Reasoning? Unveiling Human-Machine Inference Gaps in Entailment Verification](https://arxiv.org/abs/2402.03686) for more detals.
+**nli-entailment-verifier-xxl** is based on [flan-t5-xxl model](https://huggingface.co/google/flan-t5-xxl) and finetuned with a ranking objective (rank the most supported hypothesis from a given pair of hypotheses for a given premise). Please refer to our paper [Are Machines Better at Complex Reasoning? Unveiling Human-Machine Inference Gaps in Entailment Verification](https://arxiv.org/abs/2402.03686) for more detals.
 
-It is built to verify whether a given premise entails (or supports) a hypothesis or not. It works for both NLI-style of datasets and CoT rationales.
+It is built to verify whether a given premise supports a hypothesis or not. It works for both NLI-style datasets and CoT rationales. This model is specifically trained to handle multi-sentence premises (similar to what we expect in CoT rationales and other modern LLM use cases).
+
+**Note**: You can use 4-bit/8-bit [quantization](https://huggingface.co/docs/bitsandbytes/main/en/index) to reduce GPU memory usage.
 
 ## Usage
 
@@ -29,7 +31,7 @@ def get_score(model, tokenizer, input_ids):
     return scores
 
 tokenizer = AutoTokenizer.from_pretrained('google/flan-t5-xxl')
-model = AutoModelForSeq2SeqLM.from_pretrained('soumyasanyal/entailment-verifier-xxl')
+model = AutoModelForSeq2SeqLM.from_pretrained('soumyasanyal/nli-entailment-verifier-xxl')
 
 premise = "A fossil fuel is a kind of natural resource. Coal is a kind of fossil fuel."
 hypothesis = "Coal is a kind of natural resource."