Upload BART-Base-CE (Context Enhanced) for sarcasm rewriting

README.md

@@ -14,7 +14,7 @@ pipeline_tag: summarization
 
 # BART-Base-CE (Context Enhanced)
 
-BART-base fine-tuned with **Context Enhancement**: during training, the article body is prepended to the sarcastic headline so the model can ground its rewrite in factual context.
+BART-base fine-tuned with **Context Enhancement**: the non-sarcastic targets were re-generated with the full article body as additional context for the LLM annotator, producing deeper rewrites.
 
 Part of the **Project LLMao** sarcasm style transfer suite (CS4248 Team 14, NUS AY2025/26 S2).
 This model rewrites sarcastic news headlines as neutral, factual equivalents while
@@ -32,11 +32,12 @@ Example:
 ## Training
 
 - **Base model**: [`facebook/bart-base`](https://huggingface.co/facebook/bart-base) (139M params)
-- **Method**: Input format: `rewrite to non-sarcastic: <article_body> [SEP] <sarcastic_headline>`. The context provides disambiguation for headlines whose sarcasm relies on world knowledge.
-- **Dataset**: 71,730 sarcastic / non-sarcastic headline pairs derived from NHDSD
-  (News Headlines Dataset for Sarcasm Detection), augmented with 6 sarcasm strategy
-  variants (sarcasm, irony, satire, overstatement, understatement, rhetorical question).
-- **Input prefix**: `rewrite to non-sarcastic: ` is prepended to every input at inference time.
+- **Method**: Standard cross-entropy on (sarcastic_headline, non_sarcastic_target) pairs where the targets were authored by an LLM with access to the article body. The model itself takes only the sarcastic headline as input at inference time.
+- **Dataset**: 8,258 sarcastic->non-sarcastic headline pairs derived from NHDSD
+  (News Headlines Dataset for Sarcasm Detection). Non-sarcastic targets were generated
+  by an LLM annotator (StepFun Step-3.5 Flash) with cross-validation by Nemotron.
+  Split: `sar_to_non_context_enhanced`.
+- **Input format**: Raw sarcastic headline (no task prefix — BART is not pretrained with prefixes).
 - **Generation**: beam search with `num_beams=4`, `max_length=128`.
 
 ## Usage
@@ -49,9 +50,7 @@ tokenizer = AutoTokenizer.from_pretrained(model_id)
 model = AutoModelForSeq2SeqLM.from_pretrained(model_id)
 
 headline = "Area Man Passionate Defender Of What He Imagines Constitution To Be"
-prompt = "rewrite to non-sarcastic: " + headline
-
-inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=128)
+inputs = tokenizer(headline, return_tensors="pt", truncation=True, max_length=128)
 outputs = model.generate(**inputs, max_length=128, num_beams=4)
 print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```