metadata
license: apache-2.0
license_link: https://huggingface.co/Qwen/QwQ-32B-Preview/blob/main/LICENSE
language:
- en
base_model:
- Qwen/QwQ-32B-Preview
pipeline_tag: text-generation
tags:
- gptqmodel
- modelcloud
- chat
- qwen2
- qwq
- instruct
- int4
- gptq
- 4bit
๐ Update: We have discovered a regression in the QwQ quant that may output extra strings such as "Edited Text" that was result of invalid calibration data auto-injected by our vortex pipeline. v2 of the quant is undergoing benchmark evaluations and will be released soon.
This model has been quantized using GPTQModel.
- bits: 4
- dynamic: null
- group_size: 32
- desc_act: true
- static_groups: false
- sym: true
- lm_head: false
- true_sequential: true
- quant_method: "gptq"
- checkpoint_format: "gptq"
- meta๏ผ
- quantizer: gptqmodel:1.2.2
- uri: https://github.com/modelcloud/gptqmodel
- damp_percent: 0.1
- damp_auto_increment: 0.0015
Example:
from transformers import AutoTokenizer
from gptqmodel import GPTQModel
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = GPTQModel.load("ModelCloud/QwQ-32B-Preview-GPTQ-4bit")
messages = [
{"role": "system", "content": "You are Qwen, created by Alibaba Cloud. You are a helpful assistant."},
{"role": "user", "content": "How can I design a data structure in C++ to store the top 5 largest integer numbers?"},
]
input_tensor = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
outputs = model.generate(input_ids=input_tensor.to(model.device), max_new_tokens=512)
result = tokenizer.decode(outputs[0][input_tensor.shape[1]:], skip_special_tokens=True)
print(result)