PaddleRec的deepfm模型不明原因出core #61629

[xiuechen]

请提出你的问题 Please ask your question
grep: warning: GREP_OPTIONS is deprecated; please use an alias or script
2024-02-05 10:25:05,138 - INFO - ****common.configs
2024-02-05 10:25:05,138 - INFO - use_gpu: False, use_xpu: False, use_npu: False, use_visual: False, train_batch_size: 2, train_data_dir: data/, epochs: 3, print_interval: 2, model_save_path: output_model_deepfm
2024-02-05 10:25:05,138 - INFO - ***common.configs
2024-02-05 10:25:05,157 - INFO - read data
2024-02-05 10:25:05,157 - INFO - reader path:criteo_reader
2024-02-05 10:25:05,276 - INFO - epoch: 0, batch_id: 0, auc:0.000000, loss:0.8438953, avg_reader_cost: 0.00402 sec, avg_batch_cost: 0.05622 sec, avg_samples: 1.00000, ips: 17.77118 ins/s
C++ Traceback (most recent call last):
0 embedding_ad_func(paddle::Tensor const&, paddle::Tensor const&, long, bool)
1 paddle::experimental::embedding(paddle::Tensor const&, paddle::Tensor const&, long, bool)
2 void phi::EmbeddingKernel<float, phi::CPUContext>(phi::CPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, long, phi::DenseTensor)
3 GOMP_parallel

Error Message Summary:
FatalError: Process abort signal is detected by the operating system.
[TimeInfo: *** Aborted at 1707128705 (unix time) try "date -d @1707128705" if you are using GNU date ***]
[SignalInfo: *** SIGABRT (@0x8df) received by PID 2271 (TID 0x7f89f4145080) from PID 2271 ***]

做了一些data的改动和代码微调后，刚训练一个batch就出core,帮忙看一下

改动如下：
1）reader的修改：
sparse_slots = "label all_rank city_line last_click_sem_time_days_ago first_reg_date_days_ago potential_corporate_user delete_pv_last_active_day_days_ago career is_aiqicha_user is_old_enterprise last_active_client all_revrank active_pcmac_cuidnum_in7days upload_doc_last_date_days_ago download_doc_last_date_days_ago rank is_history_pay categoryconsume income_level age married consumption_amount_last_year consumer_doc_last_date_days_ago edu tc ustage sexual product_name company_cert_page_last_day_days_ago enterprise_pc_pop_up_show_lastday_days_ago svip_overdue_date_days_ago doc_backup_last_date_days_ago province constellation vip_overdue_date_days_ago carlevel is_xinyun_user company_offical_web_last_day_days_ago consume asset enterprise_pc_pop_up_clk_lastday_days_ago vip_type accountnum cuid_num used_quota"
self.sparse_slots = sparse_slots.strip().split(" ")
self.dense_slots = ["feat_value"]
self.dense_slots_shape = [1]

2）data文件修改为：
all_rank:81 city_line:169 last_click_sem_time_days_ago:266 first_reg_date_days_ago:157 potential_corporate_user:234 delete_pv_last_active_day_days_ago:225 career:233 is_aiqicha_user:36 is_old_enterprise:212 last_active_client:60 all_revrank:149 active_pcmac_cuidnum_in7days:27 upload_doc_last_date_days_ago:141 download_doc_last_date_days_ago:106 rank:267 is_history_pay:221 categoryconsume:131 income_level:195 age:184 married:250 consumption_amount_last_year:66 consumer_doc_last_date_days_ago:50 edu:150 tc:55 ustage:214 sexual:84 product_name:112 company_cert_page_last_day_days_ago:39 enterprise_pc_pop_up_show_lastday_days_ago:253 svip_overdue_date_days_ago:259 doc_backup_last_date_days_ago:105 province:48 constellation:21 vip_overdue_date_days_ago:12 carlevel:87 is_xinyun_user:193 company_offical_web_last_day_days_ago:73 consume:192 asset:38 enterprise_pc_pop_up_clk_lastday_days_ago:92 vip_type:227 accountnum:197 cuid_num:239 used_quota:76 feat_value:0.215693 label:0 uid:XXX
配置中的相应变量也做了修改