Benign huggingface Download

SlayerLab/gollem-v4-250m-pl main

Python json.load callAI PyTorch or SafeTensors extension

SHA-2562c68e6fcb694e4aafd52de58c6ba2c5df7d47c920f5ed2c7c742e1397dd0102e

URLhttps://huggingface.co/SlayerLab/gollem-v4-250m-pl

Evidence

AI PyTorch or SafeTensors extension lines 1–18

⋯6 lines
7    from modeling_gollem import GollemGPT
8    m = GollemGPT.from_pretrained("./")                     # najnowsza migawka
9    m = GollemGPT.from_pretrained("./", ckpt="ckpt_6000.safetensors")  # konkretny etap
10    print(m.generate("Polska to kraj", max_new_tokens=60))
11"""
⋯7 lines

English function-word token "for" lines 40–43

70… )
      s.blocks = nn.ModuleList([Block(d, nh) for _ in range(nl)])
      s.lnf = nn.LayerNorm(d); s.head = nn.Linear(d, vocab, bias=False)
      s. …

Calls Python glob for file enumeration lines 53–59

26… ss
54
  @staticmethod
  def _latest_ckpt(path):
      cks = glob.glob(os.path.join(path, "ckpt_*.safetensors"))
      assert cks, f"brak ckpt_*.safetensors w {path}"
      return os.path.basename(max(cks, key=lambda p: int(re.search(r"ckpt_(\d+)", p).group(1) …

Python json.load call lines 62–64

12… m_pretrained(cls, path, ckpt=None, device="cpu"):
      cfg = json.load(open(os.path.join(path, "config.json")))
      m = cls(cfg["vocab_size"], cfg["d_model"], cfg["n_layer"], cfg["n_head"], cfg["block_ …

English function-word token "and" lines 68–74

27… d = m.load_state_dict(sd, strict=False)
      assert not unexpected and missing in ([], ["head.weight"]), f"load mismatch: {missing}/{unexpected}"
      m.eval().to(device)
      tokp = os.path.join(path, "tokenizer.json")
      if os.path.exists(tokp):
          from tokenizers import Tokenizer
          m._tok = Tokenizer.from_file(to …