- 综合排序
- 最热优先
- 最新优先
- 来自专栏AI小白入门
【智能司法】可解释的Rationale增强罪名预测系统
论文标题：Interpretable Rationale Augmented Charge Prediction System 论文来源：COLING 2018 论文链接：https://www.aclweb.org 模型 Interpretable Rationale Augmented Charge Prediction System ? Phrase-level Rationale Extraction ? 考虑到词片段作为rationales在语义上应该更完整，因此本文用短语（而不是单词）来表示事实描述。 整个Phrase-level Rationale Extraction的过程如图所示，首先短语通过Embedding Layer之后过一个Bi-RNN Layer，然后到Selection Layer， Rationale Augmented Charge Prediction 本部分将利用Extractor生成的基本信息来训练Classifier。
1.4K10发布于 2020-03-31 - 来自专栏DrugOne
ICML | 有了RationaleRL，多目标分子设计不是难题！
，但现实中缺乏这样的实例，所以作者首先对单个性质的rationale进行提取，之后通过合并多个单性质的rationale得到多属性的rationale vocabulary。 下图为单个属性的rationale提取过程： ? 微调： 在特定rationale上进行微调。 3 结果 （1）分子评估 作者在四个属性GNK3β，JNK3（阿兹海默疾病的靶点），类药性和分子可及性上对分子进行属性得分评估。 该模型缓解了这个问题，因为生成的化合物是基于从用于训练特性预测因子的真实正向化合物中提取的rationale构建的。因此，本文生成的化合物比从头生成的化合物更接近真实的化合物。 4 总结 作者提出了基于rationale的分子设计生成模型，可分两个阶段生成分子：1）识别其存在表明每种特性的rationale；2）使用图生成模型将子结构图扩展为分子，并将其微调至所需的特性组合并展示了在各种任务中比以前的强化学习方法有很大的改进
1.3K50发布于 2021-02-02 - 来自专栏大模型成长之路
【大模型学习 | Prompt工程基础学习】
= rationale[clustered_idx[i][min_idx]].strip() c_pred_ans = pred_ans[clustered_idx[i][min_idx ("\n\n", "\n").split("\n")) <= max_ra_len and c_rationale[-1] == "." and c_pred_ans ! (".")[-2] or c_pred_ans.strip() in c_rationale.split()[-10:]): continue c_question = question[clustered_idx[i][min_idx]] c_rationale = c_rationale.replace("\n\n", "\n").replace("\n", " ").strip() c_rationale = " ".join(c_rationale.split()) if
1.2K21编辑于 2025-07-15 - 来自专栏CSDN社区搬运
逐步蒸馏论文复现
LLM 可以产生自然语言的理由（rationale），这些理由可以用来解释和支持模型所预测的标签。 例如，当被问及“一位先生携带着打高尔夫球的设备，他可能有什么？ 通过在输入中添加任务前缀（如“[label]”和“[rationale]”），指导模型在不同场景下生成标签或理由。 Inference, NLI） e-SNLI (Explainable SNLI):基于SNLI（Stanford Natural Language Inference）的扩展版本，增加了每个推理对的解释（rationale -model_type standard --label_type llm --batch_size 64 结合标签与推理的逐步蒸馏 使用PaLM生成的标签（PaLM label）和推理（PaLM rationale __init__(**kwargs) self.alpha = alpha self.output_rationale = output_rationale
53410编辑于 2025-01-02 - 来自专栏韩曙亮的移动开发专栏
【Android 应用开发】Google 官方 EasyPermissions 权限申请库 ( 权限申请原理对话框操作回调接口 | 永久拒绝权限后引导设用户置权限 )
永久拒绝权限后引导设用户置权限 一、权限申请原理对话框操作回调接口 RationaleCallbacks ---- 权限申请 对话框操作回调接口 RationaleCallbacks: ① 权限申请原理对话框 ( Rationale 再次申请会自动弹出该对话框 ; ② 对话框显示内容设置 : 如果用户之前拒绝过所申请的权限 , 再次申请就会自动弹出该对话框 , 对话框内容是如下 requestPermissions 方法的第三个参数 String rationale * @param rationale 解释为什么申请改组权限的原因的信息; * 这些信息会在用户第一次拒绝权限申请后显示在 权限申请原理对话框 */ public static void requestPermissions( @NonNull Activity host, @NonNull String rationale new PermissionRequest.Builder(host, requestCode, perms) .setRationale(rationale
1.4K00编辑于 2023-03-28 - 来自专栏数据和云
聊聊SQL优化的基础思路
Rationale The SQL spent 99% of its database time on CPU, I/O and Cluster waits. Rationale Database time for this SQL was divided as follows: 100% for SQL execution, 0% for Rationale SQL statement with SQL_ID "XXXXXXXXXXX" was executed 1094801 times and had an average Rationale I/O and Cluster wait for INDEX "XXXXXX.XXXXXXXX" with object ID 2133671 consumed create index XXXXX.IDX$$_5191F0001 on XxXX.XXXXXXxx(SUBSTR("ESN",-1),"STAT"); Rationale ----
85830发布于 2021-05-31 - 来自专栏数据分析与挖掘
CAIL2021-阅读理解任务-top3-模型模块
# [2, 512] rationale_logits = rationale_logits.view(batch_size, seq_length) # [2, 512] # 这里计算的是问题和文本之间的一个注意力 rationale_logits = F.softmax(rationale_logits start_logits = start_logits*rationale_logits end_logits = end_logits*rationale_logits rationale_loss = -alpha * ((1 - rationale_logits) ** gamma) * rationale_positions * torch.log( 1 - rationale_positions) * torch.log(1 - rationale_logits + 1e-8) rationale_loss = (rationale_loss
54910编辑于 2022-06-10 - 来自专栏青蛙要fly的专栏
项目需求讨论 - 动态权限申请分析及相关第三方库源码分析
requestCode, @NonNull String... perms) { requestPermissions(PermissionHelper.newInstance(host), rationale requestCode, perms); return; } // Request permissions helper.requestPermissions(rationale 然后没有该权限，就调用helper的方法 helper.requestPermissions(rationale, positiveButton, negativeButton, requestCode 我们查看到调用到的PermissionHelper中的requestPermission方法： public void requestPermissions(@NonNull String rationale 我们可以仔细看下原理，我们再回到Permission的requestPermission方法中： public void requestPermissions(@NonNull String rationale
2.3K40发布于 2018-08-29 - 来自专栏乐沙弥的世界
Oracle ADDM性能诊断利器及报告解读
Rationale Waiting for event "buffer busy waits" in wait class "Concurrency" accounted for Rationale Top level calls to execute the PL/SQL statement with SQL_ID "0w2qpuc6u2zsp" are Rationale The value of parameter "db_writer_processes" was "1" during the analysis period Rationale The value of parameter "disk_asynch_io" was "TRUE" during the analysis period. Rationale The INSERT statement with SQL_ID "gzhkw1qu6fwxm" was significantly affected by
2.6K20发布于 2018-08-06 - 来自专栏CSDN社区搬运
语言模型中的多模态链式推理
测试运行代码（epoch小、batch_size大方便加速调试）： # rationale generation 生成理由 python main.py \ --data_root data/ScienceQA caption_file data/instruct_captions.json \ --model declare-lab/flan-alpaca-base \ --user_msg rationale -05_bs16_op512_ep1/predictions_ans_eval.json \ --test_le experiments/rationale_declare-lab-flan-alpaca-base_resnet_QCM-E_lr5e # rationale generation 生成理由 python main.py \ --data_root data/ScienceQA/data \ --caption_file data/instruct_captions.json \ --model declare-lab/flan-alpaca-base \ --user_msg rationale --
76010编辑于 2024-12-01 - 来自专栏用户4822892的专栏
What is API Testing, Types & Benefits?
basic for computerizing testing since APIs currently fill in as the essential interface to application rationale It fundamentally focuses on the business rationale layer of product engineering. Programming interface comprises a lot of classes/capacities/methodology which speak to the business rationale
54000发布于 2020-03-07 - 来自专栏Android&Java技术
Android EasyPermission优雅地实现动态权限申请
request them now EasyPermissions.requestPermissions(this, getString(R.string.camera_and_location_rationale PermissionRequest.Builder(this, RC_CAMERA_AND_LOCATION, perms) .setRationale(R.string.camera_and_location_rationale ) .setPositiveButtonText(R.string.rationale_ask_ok) .setNegativeButtonText (R.string.rationale_ask_cancel) .setTheme(R.style.my_fancy_style) .build
2.6K30发布于 2020-02-27 - 来自专栏杨建荣的学习笔记
记一次数据库的分析和优化建议(r6笔记第24天)
RATIONALE: During the analysis period, the average data files' I/O throughput was 52 M persecond RELEVANT OBJECT: database object with id 51864 RATIONALE: The I/O usage statistics for the object RATIONALE: The SQL statement with SQL_ID "acbdxf552ud62" spent significant time waiting for User with SQL_ID acbdxf552ud62 update TEST_USER_BILLING set LOGIN_STATUS = 1 where UIN = :1 RATIONALE RATIONALE: During the analysis period, the average data files' I/O throughput was 1.8 M persecond
1.5K80发布于 2018-03-16 - 来自专栏乐沙弥的世界
检查及设置合理的undo表空间
Recommendation ------------------------- Minimum Recommendation : Size undo tablespace to 48 MB Rationale NUMBER; v_problem VARCHAR2(1000); v_recommendation VARCHAR2(1000); v_rationale dbms_undo_adv.undo_health(v_analyse_start_time, v_analyse_end_time, v_problem, v_recommendation, v_rationale DBMS_OUTPUT.PUT_LINE('Minimum Recommendation : ' || v_recommendation); DBMS_OUTPUT.PUT_LINE('Rationale : ' || v_rationale); DBMS_OUTPUT.PUT_LINE('Recommended Undo Tablespace
1K20发布于 2018-08-13 - 来自专栏用户4822892的专栏
What is API Testing, Types & Benefits?
basic for computerizing testing since APIs currently fill in as the essential interface to application rationale It fundamentally focuses on the business rationale layer of product engineering. Programming interface comprises a lot of classes/capacities/methodology which speak to the business rationale
58300发布于 2020-03-11 - 来自专栏编码前线
Android6.0运行时权限处理
onDenied(permissions -> { // Storage permission are not allowed. }) .start(); 权限被拒绝，说明权限用途 private Rationale mRationale = new Rationale() { @Override public void showRationale(Context context, List<String }; AndPermission.with(this) .runtime() .permission(Permission.WRITE_EXTERNAL_STORAGE) .rationale ..) .start(); 权限总是被拒绝，前往设置页授权 AndPermission.with(this) .runtime() .permission(...) .rationale
1.5K30发布于 2018-12-17 - 来自专栏登神长阶
【论文复现】语言模型中的多模态链式推理
测试运行代码（epoch小、batch_size大方便加速调试）： # rationale generation 生成理由 python main.py \ --data_root data/ScienceQA caption_file data/instruct_captions.json \ --model declare-lab/flan-alpaca-base \ --user_msg rationale -05_bs16_op512_ep1/predictions_ans_eval.json \ --test_le experiments/rationale_declare-lab-flan-alpaca-base_resnet_QCM-E_lr5e # rationale generation 生成理由 python main.py \ --data_root data/ScienceQA/data \ --caption_file data/instruct_captions.json \ --model declare-lab/flan-alpaca-base \ --user_msg rationale --
57310编辑于 2024-11-23 - 来自专栏新智元
7.7亿参数，超越5400亿PaLM！UW谷歌提出「分步蒸馏」，只需80%训练数据｜ACL 2023
论文链接：https://arxiv.org/abs/2305.02301 分布蒸馏机制把LLM中抽取出的预测理由（rationale）作为在多任务框架内训练小模型的额外监督信息。 从LLM中抽取原理（rationale） 研究人员利用少样本思维链（CoT）提示从LLM中提取预测中间步骤。 通过在输入提示中加入任务前缀「label」和「rationale」来区分标签预测和理由生成任务。
61720编辑于 2023-10-08 - 来自专栏产品优化
开发工具Tools·Nginx 配置
:beginner: Always keep Nginx up-to-date Rationale Nginx is a very secure and stable but vulnerabilities :beginner: Disable unnecessary modules Rationale It is recommended to disable any modules which are :beginner: Hide upstream proxy headers Rationale Securing a server goes far beyond not showing what’ Create a self-signed ECC certificate :beginner: Keep only TLS 1.3 and TLS 1.2 Rationale It is recommended :beginner: Reduce XSS risks (Content-Security-Policy) Rationale CSP reduce the risk and impact of XSS
1.7K10编辑于 2022-12-01 - 来自专栏机器之心
我是如何赢得GPT-4提示工程大赛冠军的
RATIONALE: Explain why [MARKETING_IDEAS] is relevant and effective for this customer group. ######### : [RATIONALE] <Annex> Give a table of the list of row numbers belonging to each cluster, in order to RATIONALE: Explain why [MARKETING_IDEAS] is relevant and effective for this customer group. ######### : [RATIONALE] <Annex> Give a table of the list of row numbers belonging to each cluster, in order to RATIONALE: Explain why [MARKETING_IDEAS] is relevant and effective for this customer group.
55510编辑于 2024-05-14
腾讯云开发者
Copyright © 2013 - 2026 Tencent Cloud. All Rights Reserved. 腾讯云 版权所有
深圳市腾讯计算机系统有限公司 ICP备案/许可证号：粤B2-20090059 
粤公网安备44030502008569号
腾讯云计算（北京）有限责任公司 京ICP证150476号 | 京ICP备11018762号