原始模型:https://huggingface.co/openbmb/MiniCPM5-2B

GPTQ Int4 量化参数:

quantize_config = QuantizeConfig(
        method="gptq",
        bits=4,
        group_size=128,
        desc_act=True,
        static_groups=True,
        sym=True,
        act_group_aware=False,
    )

以下几个参数是AX650的硬性要求,AX650只支持这种参数的GPTQ模型

bits=4,
static_groups=True,
sym=True,

desc_act=True 打开它有助于提升量化精度。 static_groupsact_group_aware是互斥项,不能同时为True。这是GPTQModel 中的量化逻辑决定的。

Downloads last month
10
Safetensors
Model size
3B params
Tensor type
BF16
·
I32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for AXERA-TECH/MiniCPM5-2B-GPTQ-Int4

Quantized
(74)
this model

Collection including AXERA-TECH/MiniCPM5-2B-GPTQ-Int4