import mindspore
from mindnlp.transformers import AutoModelForCausalLM, AutoTokenizer
device = "npu" # the device to load the model onto
model = AutoModelForCausalLM.from_pretrained("./models", ms_dtype=mindspore.float16)
tokenizer = AutoTokenizer.from_pretrained("./models", ms_dtype=mindspore.float16)
prompt = "Give me a short introduction to large language model."
messages = [{"role": "user", "content": prompt}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
model_inputs = tokenizer([text], return_tensors="ms")
generated_ids = model.generate(model_inputs.input_ids, max_new_tokens=512, do_sample=True)
generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)
点赞
收藏
回复
分享
举报
已解决
整体部署参考 微信公众平台
区别是，没有用DeepSeek-R1-Distill-Qwen-1.5B.py脚本，而是自己用mindnlp写了个小脚本运行。
import mindspore from mindnlp.transformers import AutoModelForCausalLM, AutoTokenizer device = "npu" # the device to load the model onto model = AutoModelForCausalLM.from_pretrained("./models", ms_dtype=mindspore.float16) tokenizer = AutoTokenizer.from_pretrained("./models", ms_dtype=mindspore.float16) prompt = "Give me a short introduction to large language model." messages = [{"role": "user", "content": prompt}] text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) model_inputs = tokenizer([text], return_tensors="ms") generated_ids = model.generate(model_inputs.input_ids, max_new_tokens=512, do_sample=True) generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)] response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] print(response)搞了一天环境，千辛万苦跑起来，报aclnnCumSum算子不支持
具体环境信息如下：
求助，是否是驱动/固件/CANN版本不匹配，还是当前就是不支持。
若不支持，合适能升级适配。