主流大模型测试程序-用于导出算子列表

2024-05-03 13:04
文章标签 用于 模型 导出 列表 算子 主流 测试程序

本文主要是介绍主流大模型测试程序-用于导出算子列表，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！

主流大模型测试程序-用于导出算子列表

  • 一.参考链接
  • 二.下载链接
  • 三.测试程序
  • 四.算子列表

需要多少算子才能覆盖主流大模型呢,于是 基于__torch_dispatch__机制的dump方法 dump出算子及参数列表,考虑到设备内存容量,设置为一层

一.参考链接

  • 基于__torch_dispatch__机制的dump方法
  • python序列化、反序列化函数的参数,用于问题复现

二.下载链接

下载链接
https://huggingface.co/google-bert/bert-base-chinese
https://modelscope.cn/models/baichuan-inc/baichuan-7B/summary
https://modelscope.cn/models/baichuan-inc/Baichuan2-13B-Chat/files
https://modelscope.cn/models/ZhipuAI/ChatGLM-6B/files
https://modelscope.cn/models/ZhipuAI/chatglm2-6b/files
https://modelscope.cn/models/ZhipuAI/chatglm3-6b/files
https://modelscope.cn/models/deepseek-ai/deepseek-moe-16b-chat/files
https://modelscope.cn/models/deepseek-ai/deepseek-coder-33b-base/files
https://modelscope.cn/models/AI-ModelScope/falcon-7b-instruct/files
https://modelscope.cn/models/AI-ModelScope/gpt2/files
https://modelscope.cn/models/AI-ModelScope/gemma-7b/files
https://www.modelscope.cn/models/colossalai/grok-1-pytorch/files
https://modelscope.cn/models/CHUPer/internLM/files
https://huggingface.co/internlm/internlm2-20b/tree/main
https://modelscope.cn/models/skyline2006/llama-13b/files
https://modelscope.cn/models/Cookize/Llama-2-13B-chat/files
https://modelscope.cn/models/LLM-Research/Llama3-8B-Chinese-Chat/files
https://huggingface.co/mistral-community/Mixtral-8x22B-v0.1/tree/main
https://huggingface.co/allenai/OLMo-7B/tree/main
https://huggingface.co/apple/OpenELM-3B/tree/main
https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/tree/main
https://modelscope.cn/models/qwen/Qwen-14B-Chat/files
https://modelscope.cn/models/qwen/Qwen1.5-7B/files
https://huggingface.co/google-t5/t5-base/tree/main
https://modelscope.cn/models/xverse/XVERSE-7B/files
https://modelscope.cn/models/01ai/Yi-34B/files
https://huggingface.co/IEITYuan/Yuan2-51B-hf/tree/main

三.测试程序

import warnings 
warnings.filterwarnings("ignore")
import copy
import sys
import torch
import multiprocessing as mp
from tqdm import tqdmop_mapping={}
class llm_forward:def __init__(self,func):global op_mapping  op_mapping[func.__name__]=funcself.func=funcdef __call__(self,*args,**kwargs):return self.func(*args,**kwargs)try:from torch_hook import TorchDumper,TorchDumpDispatchMode
except:class TorchDumpDispatchMode:passclass TorchDumper:def __init__(self,*args,**kwargs):        passdef __enter__(self):passdef __exit__(self, exc_type, exc_val, exc_tb):pass@llm_forward
def bert_base_chinese(use_half,device):from transformers import AutoModelForMaskedLM,BertConfigconfig=BertConfig.from_pretrained("bert_base_chinese/config.json")config.num_hidden_layers=1model = AutoModelForMaskedLM.from_config(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,config.max_position_embeddings))with TorchDumper(TorchDumpDispatchMode,op_log_path="bert_base_chinesee.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def Baichuan2_13B_Chat(use_half,device):import syssys.path.insert(0,"./Baichuan2_13B_Chat")from configuration_baichuan2 import BaichuanConfigfrom modeling_baichuan2 import BaichuanForCausalLMconfig=BaichuanConfig.from_pretrained("Baichuan2_13B_Chat/config.json")config.num_hidden_layers=1model = BaichuanForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,config.model_max_length//4))with TorchDumper(TorchDumpDispatchMode,op_log_path="Baichuan2_13B_Chat.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def baichuan_7B(use_half,device):import sysimport ossys.path.insert(0,os.path.join(os.getcwd(),"baichuan_7B"))from configuration_baichuan import BaiChuanConfigfrom modeling_baichuan import BaiChuanForCausalLMconfig=BaiChuanConfig.from_pretrained("baichuan_7B/config.json")config.num_hidden_layers=1model = BaiChuanForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,config.max_position_embeddings//4))with TorchDumper(TorchDumpDispatchMode,op_log_path="baichuan_7B.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def ChatGLM_6B(use_half,device):import syssys.path.append("./ChatGLM_6B")from configuration_chatglm import ChatGLMConfigfrom modeling_chatglm import ChatGLMModelconfig=ChatGLMConfig.from_pretrained("ChatGLM_6B/config.json")config.num_layers=1model = ChatGLMModel(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,config.max_sequence_length))input_tokens[:,0]=config.bos_token_idinput_tokens[:,2]=config.mask_token_id  input_tokens[:,-1]=config.eos_token_idwith TorchDumper(TorchDumpDispatchMode,op_log_path="ChatGLM_6B.pkl"):output=model(input_tokens.to(device))logits=output.last_hidden_stateloss=logits.mean()-1.0loss.backward()@llm_forward
def ChatGLM2_6B(use_half,device):import syssys.path.append("./ChatGLM2_6B")from configuration_chatglm import ChatGLMConfigfrom modeling_chatglm import ChatGLMModelconfig=ChatGLMConfig.from_pretrained("ChatGLM2_6B/config.json")config.num_layers=1model = ChatGLMModel(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.padded_vocab_size,(1,config.seq_length//10))with TorchDumper(TorchDumpDispatchMode,op_log_path="ChatGLM2_6B.pkl"):output=model(input_tokens.to(device))logits=output.last_hidden_stateloss=logits.mean()-1.0loss.backward()@llm_forward
def ChatGLM3_6B(use_half,device):import syssys.path.append("./ChatGLM3_6B")from configuration_chatglm import ChatGLMConfigfrom modeling_chatglm import ChatGLMModelconfig=ChatGLMConfig.from_pretrained("ChatGLM3_6B/config.json")config.num_layers=1model = ChatGLMModel(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.padded_vocab_size,(1,config.seq_length//4))with TorchDumper(TorchDumpDispatchMode,op_log_path="ChatGLM3_6B.pkl"):output=model(input_tokens.to(device))logits=output.last_hidden_stateloss=logits.mean()-1.0loss.backward()@llm_forward
def deepseek_moe_16b_chat(use_half,device):import syssys.path.append("./deepseek_moe_16b_chat")from configuration_deepseek import DeepseekConfigfrom modeling_deepseek import DeepseekForCausalLMconfig=DeepseekConfig.from_pretrained("deepseek_moe_16b_chat/config.json")config.num_hidden_layers=1model = DeepseekForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,config.max_position_embeddings))with TorchDumper(TorchDumpDispatchMode,op_log_path="deepseek_moe_16b_chat.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def deepseek_coder_33b_base(use_half,device):from transformers.models.llama import LlamaForCausalLM, LlamaConfigconfig=LlamaConfig.from_pretrained("deepseek_coder_33b_base/config.json")config.num_hidden_layers=1model = LlamaForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,config.max_position_embeddings//10))with TorchDumper(TorchDumpDispatchMode,op_log_path="deepseek_coder_33b_base.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def falcon_7b_instruct(use_half,device):import syssys.path.append("./falcon_7b_instruct")from configuration_RW import RWConfigfrom modelling_RW import RWForCausalLMconfig=RWConfig.from_pretrained("falcon_7b_instruct/config.json")config.n_layer=1model = RWForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="falcon_7b_instruct.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def GPT2(use_half,device):from transformers import GPT2LMHeadModel, GPT2Configconfig=GPT2Config.from_pretrained("GPT2/config.json")config.n_layer=1model = GPT2LMHeadModel(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="GPT2.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def gemma_7b(use_half,device):import syssys.path.append("./gemma_7b")from config import GemmaConfigfrom model import GemmaForCausalLMconfig=GemmaConfig.from_pretrained("gemma_7b/config.json")config.num_hidden_layers=1model = GemmaForCausalLM(config)if use_half:model=model.half()model.train().to(device)max_seq_len=512batch_size=1prompt_tokens=torch.randint(0,config.vocab_size,(batch_size,max_seq_len)).to(device)temperature= 0.95top_p  = 1.0top_k = 100# build KV cacheskv_caches = []for _ in range(config.num_hidden_layers):size = (batch_size, max_seq_len, config.num_key_value_heads,config.head_dim)dtype = config.get_dtype()k_cache = torch.zeros(size=size, dtype=dtype).to(device)v_cache = torch.zeros(size=size, dtype=dtype).to(device)kv_caches.append((k_cache, v_cache))# prepare inputsinput_token_ids_tensor = torch.full((batch_size, max_seq_len),0,dtype=torch.int64)input_token_ids_tensor = input_token_ids_tensor.to(device)input_positions_tensor = torch.arange(0, max_seq_len,dtype=torch.int64).to(device)mask_tensor = torch.full((1, 1, max_seq_len, max_seq_len),-2.3819763e38).to(torch.float)mask_tensor = torch.triu(mask_tensor, diagonal=1).to(device)output_positions_tensor = torch.LongTensor([max_seq_len - 1]).to(device)temperatures_tensor = None if not temperature else torch.FloatTensor([temperature] * batch_size).to(device)top_ps_tensor = torch.FloatTensor([top_p] * batch_size).to(device)top_ks_tensor = torch.LongTensor([top_k] * batch_size).to(device)with TorchDumper(TorchDumpDispatchMode,op_log_path="gemma_7b.pkl"):output=model(prompt_tokens,input_positions_tensor,None,kv_caches,mask_tensor,output_positions_tensor,temperatures_tensor,top_ps_tensor,top_ks_tensor)_,logits=outputloss=logits.mean()-1.0loss.backward()@llm_forward
def grok1_pytorch(use_half,device):import syssys.path.append("./grok1_pytorch")from configuration_grok1 import Grok1Configfrom modeling_grok1 import Grok1ModelForCausalLMconfig=Grok1Config.from_pretrained("grok1_pytorch/config.json")config.num_hidden_layers=1config.num_experts=1config.num_experts_per_tok=1model = Grok1ModelForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="grok1_pytorch.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def internLM(use_half,device):import syssys.path.append("./internLM")from configuration_internlm import InternLMConfigfrom modeling_internlm import InternLMForCausalLMconfig=InternLMConfig.from_pretrained("internLM/config.json")config.num_hidden_layers=1model = InternLMForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="internLM.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def internlm2_20b(use_half,device):import syssys.path.append("./internlm2_20b")from configuration_internlm2 import InternLM2Configfrom modeling_internlm2 import InternLM2ForCausalLMconfig=InternLM2Config.from_pretrained("internlm2_20b/config.json")config.num_hidden_layers=1model = InternLM2ForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="internlm2_20b.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def llama_13b(use_half,device):from transformers.models.llama import LlamaForCausalLM, LlamaConfigconfig=LlamaConfig.from_pretrained("llama_13b/config.json")config.num_hidden_layers=1model = LlamaForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,config.max_sequence_length))with TorchDumper(TorchDumpDispatchMode,op_log_path="llama_13b.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def Llama2_13B_chat(use_half,device):from transformers.models.llama import LlamaForCausalLM, LlamaConfigconfig=LlamaConfig.from_pretrained("Llama2_13B_chat/config.json")config.num_hidden_layers=1model = LlamaForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,128))with TorchDumper(TorchDumpDispatchMode,op_log_path="Llama2_13B_chat.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def Llama3_8B_Chinese_Chat(use_half,device):from transformers.models.llama import LlamaForCausalLM, LlamaConfigconfig=LlamaConfig.from_pretrained("Llama3_8B_Chinese_Chat/config.json")config.num_hidden_layers=1model = LlamaForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,128))with TorchDumper(TorchDumpDispatchMode,op_log_path="Llama3_8B_Chinese_Chat.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def Mixtral_8x22B(use_half,device):import syssys.path.append("./Mixtral_8x22B")from configuration_mixtral import MixtralConfigfrom modeling_mixtral import MixtralForCausalLMconfig=MixtralConfig.from_pretrained("Mixtral_8x22B/config.json")config.num_hidden_layers=1model = MixtralForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="Mixtral_8x22B.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def OLMo_7B(use_half,device):import syssys.path.append("./OLMo_7B")from configuration_olmo import OLMoConfigfrom modeling_olmo import OLMoForCausalLMconfig=OLMoConfig.from_pretrained("OLMo_7B/config.json")config.n_layers=1model = OLMoForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="OLMo_7B.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def Phi3_mini_4k_instruct(use_half,device):import syssys.path.append("./Phi3_mini_4k_instruct")from configuration_phi3 import Phi3Configfrom modeling_phi3 import Phi3ForCausalLMconfig=Phi3Config.from_pretrained("Phi3_mini_4k_instruct/config.json")config.num_hidden_layers=1model = Phi3ForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="Phi3_mini_4k_instruct.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def OpenELM_3B(use_half,device):import syssys.path.append("./OpenELM_3B")from configuration_openelm import OpenELMConfigfrom modeling_openelm import OpenELMForCausalLMconfig=OpenELMConfig.from_pretrained("OpenELM_3B/config.json")config.num_transformer_layers=1model = OpenELMForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="OpenELM_3B.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def Qwen_14B_Chat(use_half,device):import syssys.path.append("./Qwen_14B_Chat")from configuration_qwen import QWenConfigfrom modeling_qwen import QWenLMHeadModelconfig=QWenConfig.from_pretrained("Qwen_14B_Chat/config.json")config.num_hidden_layers=1model = QWenLMHeadModel(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="Qwen_14B_Chat.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def Qwen1_5_7B(use_half,device):import syssys.path.append("./Qwen1_5_7B")from configuration_qwen2 import Qwen2Configfrom modeling_qwen2 import Qwen2ForCausalLMconfig=Qwen2Config.from_pretrained("Qwen1_5_7B/config.json")config.num_hidden_layers=1model = Qwen2ForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="Qwen1_5_7B.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def t5_base(use_half,device):import syssys.path.append("./t5_base")from transformers import T5Config, T5ForConditionalGenerationconfig=T5Config.from_pretrained("t5_base/config.json")config.num_layers=1config.max_new_tokens=512model = T5ForConditionalGeneration(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,config.max_new_tokens))with TorchDumper(TorchDumpDispatchMode,op_log_path="t5_base.pkl"):output=model.generate(input_tokens.to(device))#logits=output#loss=logits.mean()-1.0#loss.backward()@llm_forward
def XVERSE_7B(use_half,device):import syssys.path.append("./XVERSE_7B")from configuration_xverse import XverseConfigfrom modeling_xverse import XverseForCausalLMconfig=XverseConfig.from_pretrained("XVERSE_7B/config.json")config.num_hidden_layers=1model = XverseForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,512))with TorchDumper(TorchDumpDispatchMode,op_log_path="XVERSE_7B.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def Yi_34B(use_half,device):from transformers.models.llama import LlamaForCausalLM, LlamaConfigconfig=LlamaConfig.from_pretrained("Yi_34B/config.json")config.num_hidden_layers=1model = LlamaForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,config.max_position_embeddings//10))with TorchDumper(TorchDumpDispatchMode,op_log_path="Yi_34B.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()@llm_forward
def Yuan2_51B_hf(use_half,device):import syssys.path.append("./Yuan2_51B_hf")from configuration_yuan import YuanConfigfrom yuan_hf_model import YuanForCausalLMconfig=YuanConfig.from_pretrained("Yuan2_51B_hf/config.json")config.num_hidden_layers=1config.intermediate_size=2048config.model_max_length=config.max_position_embeddings=2model = YuanForCausalLM(config)if use_half:model=model.half()model.train().to(device)input_tokens=torch.randint(0,config.vocab_size,(1,2))with TorchDumper(TorchDumpDispatchMode,op_log_path="Yuan2_51B_hf.pkl"):output=model(input_tokens.to(device))logits=output.logitsloss=logits.mean()-1.0loss.backward()def main():global op_mappingdevice="cuda"use_half=Truepbar=tqdm(list(op_mapping.keys()))for name in pbar:        torch.manual_seed(1)p = mp.Process(target=op_mapping[name],args=(use_half,device))p.start()p.join()torch.cuda.empty_cache()pbar.set_description("%s" % (name))if __name__=='__main__':main()

四.算子列表

算子列表
aten.abs.default
aten.addmm.default
aten.add.Tensor
aten.add_.Tensor
aten.alias.default
aten.all.default
aten.any.default
aten.arange.default
aten.arange.start
aten.arange.start_step
aten.argmax.default
aten.as_strided.default
aten.baddbmm.default
aten.bitwise_not.default
aten.bitwise_or.Tensor
aten.bmm.default
aten.cat.default
aten.clamp_min.default
aten.clamp.Tensor
aten.clone.default
aten._conj.default
aten.convolution_backward.default
aten.convolution.default
aten.copy_.default
aten.cos.default
aten.cumsum.default
aten.diagonal_copy.default
aten.div.Scalar
aten.div.Tensor
aten.div_.Tensor
aten.embedding.default
aten.embedding_dense_backward.default
aten.empty.memory_format
aten.empty.names
aten.eq.Scalar
aten.eq.Tensor
aten.expand.default
aten.fill_.Scalar
aten.fill_.Tensor
aten.full.default
aten.full_like.default
aten.gather.default
aten.gelu_backward.default
aten.gelu.default
aten.ge.Scalar
aten.ge.Tensor
aten.gt.Scalar
aten.gt.Tensor
aten.index_add_.default
aten.index_copy_.default
aten.index_put.default
aten.index_select.default
aten.index.Tensor
aten.isinf.default
aten.is_same_size.default
aten.le.Tensor
aten.lift_fresh.default
aten.linalg_vector_norm.default
aten._local_scalar_dense.default
aten.log.default
aten.logical_not.default
aten.lt.Scalar
aten.lt.Tensor
aten.masked_fill.Scalar
aten.masked_fill_.Scalar
aten.max.default
aten.maximum.default
aten.mean.default
aten.mean.dim
aten.minimum.default
aten.mm.default
aten.mul.Scalar
aten.mul.Tensor
aten.multinomial.default
aten.native_dropout_backward.default
aten.native_dropout.default
aten.native_layer_norm_backward.default
aten.native_layer_norm.default
aten.neg.default
aten.ne.Tensor
aten.new_empty.default
aten.new_empty_strided.default
aten.new_zeros.default
aten.nonzero.default
aten.ones.default
aten.ones_like.default
aten.permute.default
aten.pow.Scalar
aten.pow.Tensor_Scalar
aten.prod.dim_int
aten.reciprocal.default
aten.relu.default
aten.repeat.default
aten.rsqrt.default
aten.rsub.Scalar
aten.scalar_tensor.default
aten._scaled_dot_product_efficient_attention_backward.default
aten._scaled_dot_product_efficient_attention.default
aten.scatter.src
aten.scatter_.value
aten.select_backward.default
aten.select.int
aten.set_.source_Storage
aten.set_.source_Storage_storage_offset
aten.silu_backward.default
aten.silu.default
aten.sin.default
aten.slice_backward.default
aten.slice.Tensor
aten._softmax_backward_data.default
aten._softmax.default
aten.sort.default
aten.split.Tensor
aten.split_with_sizes.default
aten.squeeze.dim
aten.stack.default
aten.sub.Tensor
aten.sum.dim_IntList
aten.tanh_backward.default
aten.tanh.default
aten.t.default
aten._to_copy.default
aten.topk.default
aten.transpose.int
aten.tril.default
aten.tril_.default
aten.triu.default
aten.unbind.int
aten._unsafe_view.default
aten.unsqueeze.default
aten.unsqueeze_.default
aten.view_as_complex.default
aten.view_as_real.default
aten.view.default
aten.where.self
aten.zeros.default
aten.zeros_like.default

这篇关于主流大模型测试程序-用于导出算子列表的文章就介绍到这儿，希望我们推荐的文章对编程师们有所帮助！


http://www.chinasem.cn/article/956812。

相关文章

Python如何计算两个不同类型列表的相似度

Python如何计算两个不同类型列表的相似度

《Python如何计算两个不同类型列表的相似度》在编程中,经常需要比较两个列表的相似度,尤其是当这两个列表包含不同类型的元素时,下面小编就来讲讲如何使用Python计算两个不同类型列表的相似度吧... 目录摘要引言数字类型相似度欧几里得距离曼哈顿距离字符串类型相似度Levenshtein距离Jaccard相
阅读更多...
0基础租个硬件玩deepseek,蓝耘元生代智算云|本地部署DeepSeek R1模型的操作流程

0基础租个硬件玩deepseek,蓝耘元生代智算云|本地部署DeepSeek R1模型的操作流程

《0基础租个硬件玩deepseek,蓝耘元生代智算云|本地部署DeepSeekR1模型的操作流程》DeepSeekR1模型凭借其强大的自然语言处理能力,在未来具有广阔的应用前景,有望在多个领域发... 目录0基础租个硬件玩deepseek,蓝耘元生代智算云|本地部署DeepSeek R1模型，3步搞定一个应
阅读更多...
Deepseek R1模型本地化部署+API接口调用详细教程(释放AI生产力)

Deepseek R1模型本地化部署+API接口调用详细教程(释放AI生产力)

《DeepseekR1模型本地化部署+API接口调用详细教程(释放AI生产力)》本文介绍了本地部署DeepSeekR1模型和通过API调用将其集成到VSCode中的过程,作者详细步骤展示了如何下载和... 目录前言一、deepseek R1模型与chatGPT o1系列模型对比二、本地部署步骤1.安装oll
阅读更多...
Spring AI Alibaba接入大模型时的依赖问题小结

Spring AI Alibaba接入大模型时的依赖问题小结

《SpringAIAlibaba接入大模型时的依赖问题小结》文章介绍了如何在pom.xml文件中配置SpringAIAlibaba依赖,并提供了一个示例pom.xml文件,同时,建议将Maven仓... 目录（一）pom.XML文件：（二）application.yml配置文件（一）pom.xml文件：首
阅读更多...
如何在本地部署 DeepSeek Janus Pro 文生图大模型

如何在本地部署 DeepSeek Janus Pro 文生图大模型

《如何在本地部署DeepSeekJanusPro文生图大模型》DeepSeekJanusPro模型在本地成功部署,支持图片理解和文生图功能,通过Gradio界面进行交互,展示了其强大的多模态处... 目录什么是 Janus Pro1. 安装 conda2. 创建 python 虚拟环境3. 克隆 janus
阅读更多...
Redis存储的列表分页和检索的实现方法

Redis存储的列表分页和检索的实现方法

《Redis存储的列表分页和检索的实现方法》在Redis中,列表（List）是一种有序的数据结构,通常用于存储一系列元素,由于列表是有序的,可以通过索引来访问元素,因此可以很方便地实现分页和检索功能,... 目录一、Redis 列表的基本操作二、分页实现三、检索实现3.1 方法 1：客户端过滤3.2 方法
阅读更多...
本地私有化部署DeepSeek模型的详细教程

本地私有化部署DeepSeek模型的详细教程

《本地私有化部署DeepSeek模型的详细教程》DeepSeek模型是一种强大的语言模型,本地私有化部署可以让用户在自己的环境中安全、高效地使用该模型,避免数据传输到外部带来的安全风险,同时也能根据自... 目录一、引言二、环境准备（一）硬件要求（二）软件要求（三）创建虚拟环境三、安装依赖库四、获取 Dee
阅读更多...
Java导出Excel动态表头的示例详解

Java导出Excel动态表头的示例详解

《Java导出Excel动态表头的示例详解》这篇文章主要为大家详细介绍了Java导出Excel动态表头的相关知识,文中的示例代码简洁易懂,具有一定的借鉴价值,有需要的小伙伴可以了解下... 目录前言一、效果展示二、代码实现1.固定头实体类2.动态头实现3.导出动态头前言本文只记录大致思路以及做法，代码不进
阅读更多...
DeepSeek模型本地部署的详细教程

DeepSeek模型本地部署的详细教程

《DeepSeek模型本地部署的详细教程》DeepSeek作为一款开源且性能强大的大语言模型,提供了灵活的本地部署方案,让用户能够在本地环境中高效运行模型,同时保护数据隐私,在本地成功部署DeepSe... 目录一、环境准备（一）硬件需求（二）软件依赖二、安装Ollama三、下载并部署DeepSeek模型选
阅读更多...
详解Vue如何使用xlsx库导出Excel文件

详解Vue如何使用xlsx库导出Excel文件

《详解Vue如何使用xlsx库导出Excel文件》第三方库xlsx提供了强大的功能来处理Excel文件,它可以简化导出Excel文件这个过程,本文将为大家详细介绍一下它的具体使用,需要的小伙伴可以了解... 目录1. 安装依赖2. 创建vue组件3. 解释代码在Vue.js项目中导出Excel文件，使用第三
阅读更多...

Copyright China编程 23002807@qq.com

沪ICP备2023033072号-2