Brunobkr/llama.cpp_AlgMor24_github

1from __future__ import annotations2 3from .base import ModelBase, TextModel, gguf4 5 6@ModelBase.register("OrionForCausalLM")7class OrionModel(TextModel):8    model_arch = gguf.MODEL_ARCH.ORION9 10    def set_vocab(self):11        self._set_vocab_sentencepiece()12 13    def set_gguf_parameters(self):14        head_count = self.hparams["num_attention_heads"]15        head_count_kv = self.hparams.get("num_key_value_heads", head_count)16 17        ctx_length = 018        if "max_sequence_length" in self.hparams:19            ctx_length = self.hparams["max_sequence_length"]20        elif "max_position_embeddings" in self.hparams:21            ctx_length = self.hparams["max_position_embeddings"]22        elif "model_max_length" in self.hparams:23            ctx_length = self.hparams["model_max_length"]24        else:25            raise ValueError("gguf: can not find ctx length parameter.")26 27        self.gguf_writer.add_file_type(self.ftype)28        self.gguf_writer.add_tensor_data_layout("Meta AI original pth")29        self.gguf_writer.add_context_length(ctx_length)30        self.gguf_writer.add_embedding_length(self.hparams["hidden_size"])31        self.gguf_writer.add_block_count(self.block_count)32        self.gguf_writer.add_feed_forward_length(self.hparams["intermediate_size"])33        self.gguf_writer.add_head_count(head_count)34        self.gguf_writer.add_head_count_kv(head_count_kv)35        # note: config provides rms norm but it is actually layer norm36        # ref:  https://huggingface.co/OrionStarAI/Orion-14B-Chat/blob/276a17221ce42beb45f66fac657a41540e71f4f5/modeling_orion.py#L570-L57137        self.gguf_writer.add_layer_norm_eps(self.hparams["rms_norm_eps"])38