char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

Xuan-Son Nguyenconvert: add @ModelBase.example (#27208)fa88ae936

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
1.1 KiB35 linesraw
1from __future__ import annotations
2
3from typing import Callable, TYPE_CHECKING
4
5if TYPE_CHECKING:
6    from torch import Tensor
7
8from .base import MmprojModel, ModelBase, gguf
9
10from .llama import LlamaModel
11
12
13@ModelBase.register("CogVLMForCausalLM")
14@ModelBase.example("THUDM/cogvlm2-llama3-chat-19B", "THUDM/cogvlm-chat-hf")
15class CogVLMVisionModel(MmprojModel):
16
17    def set_gguf_parameters(self):
18        super().set_gguf_parameters()
19        self.gguf_writer.add_vision_attention_layernorm_eps(self.hparams.get("layer_norm_eps", 1e-6))
20        self.gguf_writer.add_clip_projector_type(gguf.VisionProjectorType.COGVLM)
21
22    @classmethod
23    def filter_tensors(cls, item: tuple[str, Callable[[], Tensor]]) -> tuple[str, Callable[[], Tensor]] | None:
24        name, gen = item
25
26        if not name.startswith("model.vision."):
27            return None
28
29        return super().filter_tensors(item)
30
31
32@ModelBase.register("CogVLMForCausalLM")
33@ModelBase.example("THUDM/cogvlm2-llama3-chat-19B", "THUDM/cogvlm-chat-hf")
34class CogVLMModel(LlamaModel):
35    model_arch = gguf.MODEL_ARCH.COGVLM