add the support of the openbmb/minicpm

[LDLINGLINGLING]

Wǒ shì zhōngguó openbmb shíyàn shì de gōngzuò rényuán, zhège pull request zēngjiāle auto_gptq duì openbmb/minicpm de zhīchí, rúguǒ néng dádào nín de yāoqiú, qǐng hébìng
79 / 5,000
I am a staff member of the openbmb laboratory in China. This pull request adds auto_gptq support for openbmb/minicpm. If it meets your requirements, please merge it.

[Qubitium]

@LDLINGLINGLING Please submit a quantized test model and have it on hf if possible that has been tested with this code.

[LDLINGLINGLING]

Okay, the model URL for Hugging Face is as follows:https://huggingface.co/openbmb/minicpm_2b_int4/tree/main ，if you want to test the model, please download all the file in the space, I give a script of test quantized model, the result follow:

Map: 100%|██████████████████████████████████████████████████████| 500/500 [00:00<00:00, 12027.51 examples/s]
eval result for base model: {'ppl': 6.5847134469057895}

eval result for quantized model: {'ppl': 6.603494453221514}

[LDLINGLINGLING]

Hello, may I ask if the model has been tested? We look forward to our model being supported by the main code of Autogptq.

[Qubitium]

@LDLINGLINGLING Ported, tested, and validated minicpm model to our fork + major refractor of AutoGPTQ that we will officially announce later today. Please check with your team.

ModelCloud/GPTQModel#18

[LDLINGLINGLING]

"I don't fully understand what you mean. What should I do?"，