Benign huggingface Download

hanseungwook/Llama-3.2-1B-Compressed-Joint-GSM8K-TrimAnswer main

Python repeats repository clone calls

SHA-256cca30c5eab1b6da7f5eeea98f46482fefd7e11004a241b3a7ba89a448843b78e

URLhttps://huggingface.co/hanseungwook/Llama-3.2-1B-Compressed-Joint-GSM8K-TrimAnswer

MaleculeH(Po)

Evidence

English function-word token "and" lines 1–11

1"""Thin model-family views of one PyX selection and token stream."""
2from __future__ import annotations
3
⋯8 lines

English function-word token "for" lines 22–27

30… ne
      self.max_sequence_tokens = 0
      for row, fields in zip(self.rows, encodings):
          q, c, a = (fields[name] for name in ("question_ids", "cot_ids", "answer_ids"))
          def tensor(values):
              retur …

English function-word token "the" lines 43–47

70… _id] + a)
              # Position predicting the first literal token, even when a token
              # merges the marker's trailing whitespace with the literal.
              ref_anchor = len(q) + len(c) + fields["answer_start"] - 1
       …

Python repeats repository clone calls lines 51–53

21…     "ref_input_ids": reference, "ref_labels": labels, "labels": decoder.clone(),
                      "ref_answer_position": ref_anchor,
                      "model_answer_position": fields["answer_s …