llama.cpp/conversion/orion.py

from __future__ import annotations

from .base import ModelBase, TextModel, gguf


@ModelBase.register("OrionForCausalLM")
class OrionModel(TextModel):
    model_arch = gguf.MODEL_ARCH.ORION

    def set_vocab(self):
        self._set_vocab_sentencepiece()

    def set_gguf_parameters(self):
        head_count = self.hparams["num_attention_heads"]
        head_count_kv = self.hparams.get("num_key_value_heads", head_count)

        ctx_length = 0
        if "max_sequence_length" in self.hparams:
            ctx_length = self.hparams["max_sequence_length"]
        elif "max_position_embeddings" in self.hparams:
            ctx_length = self.hparams["max_position_embeddings"]
        elif "model_max_length" in self.hparams:
            ctx_length = self.hparams["model_max_length"]
        else:
            raise ValueError("gguf: can not find ctx length parameter.")

        self.gguf_writer.add_file_type(self.ftype)
        self.gguf_writer.add_tensor_data_layout("Meta AI original pth")
        self.gguf_writer.add_context_length(ctx_length)
        self.gguf_writer.add_embedding_length(self.hparams["hidden_size"])
        self.gguf_writer.add_block_count(self.block_count)
        self.gguf_writer.add_feed_forward_length(self.hparams["intermediate_size"])
        self.gguf_writer.add_head_count(head_count)
        self.gguf_writer.add_head_count_kv(head_count_kv)
        # note: config provides rms norm but it is actually layer norm
        # ref:  https://huggingface.co/OrionStarAI/Orion-14B-Chat/blob/276a17221ce42beb45f66fac657a41540e71f4f5/modeling_orion.py#L570-L571
        self.gguf_writer.add_layer_norm_eps(self.hparams["rms_norm_eps"])