Method write_all

convert-dense.py:916–948

(fname_out: Path, ftype: GGMLFileType, params: Params, model: LazyModel, vocab: Vocab, svocab: gguf.SpecialVocab, concurrency: int = DEFAULT_CONCURRENCY, endianess: gguf.GGUFEndian = gguf.GGUFEndian.LITTLE)

    @staticmethod
    def write_all(fname_out: Path, ftype: GGMLFileType, params: Params, model: LazyModel, vocab: Vocab, svocab: gguf.SpecialVocab, concurrency: int = DEFAULT_CONCURRENCY, endianess: gguf.GGUFEndian = gguf.GGUFEndian.LITTLE) -> None:
        check_vocab_size(params, vocab)

        of = OutputFile(fname_out, params.arch, endianess=endianess)

        # meta data
        of.add_meta_arch(params)
        of.add_meta_vocab(vocab)
        of.add_meta_special_vocab(svocab)

        # tensor info
        for name, lazy_tensor in model.items():
            of.add_tensor_info(name, lazy_tensor)

        of.write_meta()
        of.write_tensor_info()

        # tensor data
        ndarrays_inner = bounded_parallel_map(OutputFile.do_item, model.items(), concurrency = concurrency)
        if ftype == GGMLFileType.MostlyQ8_0:
            ndarrays = bounded_parallel_map(OutputFile.maybe_do_quantize, ndarrays_inner, concurrency = concurrency, max_workers = concurrency, use_processpool_executor = True)
        else:
            ndarrays = map(OutputFile.maybe_do_quantize, ndarrays_inner)

        start = time.time()
        for i, ((name, lazy_tensor), ndarray) in enumerate(zip(model.items(), ndarrays)):
            elapsed = time.time() - start
            size = ' x '.join(f"{dim:6d}" for dim in lazy_tensor.shape)
            padi = len(str(len(model)))
            print(f"[{i+1:{padi}d}/{len(model)}] Writing tensor {name:38s} | size {size:16} | type {lazy_tensor.data_type.name:4} | T+{int(elapsed):4}")
            of.gguf.write_tensor_data(ndarray)

        of.close()

def pick_output_type(model: LazyModel, output_type_str: str | None) -> GGMLFileType:
    wq_type = model[gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ATTN_Q].format(bid=0)+".weight"].data_type