EBFT: Matching Features, Not Tokens: Energy-Based Fine-Tuning of Language Models (#3527) [skip ci]

* EBFT wip

* fixes

* more fixeS

* add missing strided module

* ebft fixes for multi-turn

* make ebft work with async

* add example for ebft w qwen3.5

* fix for split thinking and update yaml for lora over linear attention only

* enforce_eager for vllm arg in schema

* fix sync weights

* fix multi-gpu

* handle updated sig for mm

* ddp fixes

* improve multi-gpu handling, don't calculate logits, adaptive completion length

* chore: lint

* chore: lint

* support completion_mean

* Address corereview feedback

* clamp min IS ratio

* Address PR code review

* more fixes identified

* address code review

* Fix property from rebase conflict

examples/ebft/ebft_pretrain.py

@@ -0,0 +1,31 @@
+"""
+Dataset transform for unstructured text data with strided EBFT.
+
+Tokenizes raw text into fixed-length input_ids for the strided trainer.
+Sequences are padded to sequence_len for uniform batching.
+"""
+
+
+def transform(cfg, *args, **kwargs):
+    seq_len = cfg.sequence_len
+
+    def transform_fn(example, tokenizer=None):
+        text = example.get("question", example.get("text", ""))
+        if tokenizer is None:
+            return {"prompt": text}
+
+        encoded = tokenizer(
+            text,
+            truncation=True,
+            max_length=seq_len,
+            padding="max_length",
+            add_special_tokens=True,
+            return_tensors=None,
+        )
+        return {
+            "input_ids": encoded["input_ids"],
+            "attention_mask": encoded["attention_mask"],
+            "labels": list(encoded["input_ids"]),
+        }
+
+    return transform_fn, {"remove_columns": ["question", "answer"]}