Fix bugs in website

Naqu6 · Naqu6 · commit 103ccdebbbca · 2025-01-27T13:57:35.000-08:00
diff --git a/_blogs/cats.md b/_blogs/cats.md
@@ -16,7 +16,7 @@ date: 2024-06-01
 doi:
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
 teaser: The human brain operates with remarkable energy efficiency, despite its complexity. Usually, a human brain activates only a sparse array of neurons at any given moment. Do state-of-the-art large language models (LLMs) behave similarly? Until recently, the answer was no. They typically exhibit over 99% non-zero activations during inference. However, our recent research unveils a surprising observation, LLMs activations are intrinsically sparse.
 materials:
   - name: Blog Post
diff --git a/_blogs/kernelbench.md b/_blogs/kernelbench.md
@@ -9,7 +9,7 @@ authors:
 tags:
   - ml systems
   - gpu programming
-  - generative AI
+  - generative ai
 venue: none
 year: 2024
 date: 2024-12-03
diff --git a/_blogs/monkeys.md b/_blogs/monkeys.md
@@ -18,7 +18,7 @@ authors:
 tags:
   - natural language processing
   - scaling laws
-  - generative AI
+  - generative ai
 venue: none
 year: 2024
 date: 2024-09-01
diff --git a/_data/tags.yml b/_data/tags.yml
@@ -5,13 +5,9 @@
   - model
   - dataset
   - empirical study
-- # methods
-  - qualitative methods
-  - quantitative methods
-  - mixed methods
 - # domain
   - natural language processing
   - machine learning
-  - generative AI
-  - ML systems
+  - generative ai
+  - ml systems
   
diff --git a/_layouts/page.html b/_layouts/page.html
@@ -6,9 +6,13 @@
   {% if page.title == 'Publications' %}
     <script src="https://cdn.jsdelivr.net/gh/nextapps-de/flexsearch@0.7.2/dist/flexsearch.compact.js"></script>
     <script type="text/javascript">
-      // {% include blogsearch.js %}
       {% include pubsearch.js %}
     </script>
+  {% elsif page.title == 'Blogs' %}
+    <script src="https://cdn.jsdelivr.net/gh/nextapps-de/flexsearch@0.7.2/dist/flexsearch.compact.js"></script>
+    <script type="text/javascript">
+      {% include blogsearch.js %}
+    </script>
   {% endif %}
 </head>
 <body>
diff --git a/_pubs/CATSpaper.md b/_pubs/CATSpaper.md
@@ -13,12 +13,12 @@ authors:
   - key: azaliamirhoseini
 venue: preprint
 year: 2024
-day: 117
+date: 2024-04-26
 has_pdf: true
 doi: 10.48550/arXiv.2404.08763
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
 teaser: The human brain operates with remarkable energy efficiency, despite its complexity. Usually, a human brain activates only a sparse array of neurons at any given moment. Do state-of-the-art large language models (LLMs) behave similarly? Until recently, the answer was no. They typically exhibit over 99% non-zero activations during inference. However, our recent research unveils a surprising observation, LLMs activations are intrinsically sparse.
 materials:
   - name: CATS
diff --git a/_pubs/CHESSpaper.md b/_pubs/CHESSpaper.md
@@ -12,12 +12,12 @@ authors:
     affiliation: Stanford University
 venue: preprint
 year: 2024
-day: 178
+date: 2024-04-23
 has_pdf: true
 doi: 2405.16755​
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
 teaser: Translating natural language questions into database queries, or text-to-SQL, is a long-standing research problem. This issue has been exacerbated in recent years due to the growing complexity of databases, driven by the increasing sizes of schemas (sets of columns and tables), values (content), and catalogs (metadata describing schemas and values) stored within them. 
 materials:
   - name: CHESS
diff --git a/_pubs/archon.md b/_pubs/archon.md
@@ -22,12 +22,12 @@ authors:
   - key: azaliamirhoseini
 venue: preprint
 year: 2024
-day: 267
+date: 2024-09-23
 has_pdf: true
 doi: 10.48550/arXiv.2409.15254
 tags:
   - machine learning
-  - generative AI
+  - generative ai
   - inference-time techniques
 teaser: Archon, a modular framework for designing inference-time architectures, outperforms top language models like GPT-4 and Claude 3.5 on various benchmarks by optimally combining LLMs and inference techniques.
 materials:
diff --git a/_pubs/codemonkeys.md b/_pubs/codemonkeys.md
@@ -15,12 +15,12 @@ authors:
   - key: azaliamirhoseini
 venue: preprint
 year: 2025
-day: 23
+date: 2025-01-23
 has_pdf: true
 doi: 10.48550/arXiv.2501.14723
 tags:
   - machine learning
-  - generative AI
+  - generative ai
 teaser: CodeMonkeys, a system designed to solve software engineering problems by scaling test time compute.
 materials:
   - name: Paper
diff --git a/_pubs/constitutionalai.md b/_pubs/constitutionalai.md
@@ -55,12 +55,12 @@ authors:
 
 venue: preprint
 year: 2022
-day: 349
+date: 2022-12-15
 has_pdf: false
 doi: 2212.08073v1
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
   - highlight
 teaser: Imagine an AI that polices itself to ensure it remains helpful and safe. Our approach, called Constitutional AI, trains an AI assistant through a two-phase process. First, it learns from itself by generating and revising its own responses. Then, it refines its performance through reinforcement learning based on feedback from its own evaluations. This method not only improves the AI’s ability to handle sensitive queries but also requires minimal human oversight, making it a powerful tool for creating AI systems that are both effective and secure.
 materials:
diff --git a/_pubs/dla.md b/_pubs/dla.md
@@ -9,12 +9,12 @@ authors:
   - key: azaliamirhoseini
 venue: preprint
 year: 2022
-day: 139
+date: 2022-05-22
 has_pdf: false
 doi: 1558-1101
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
   - highlight
 teaser: Unlock the potential of power-efficient deep learning with AutoApprox, a groundbreaking framework that creates low-power accelerators without sacrificing accuracy. By leveraging a TPUv3-like systolic array and a learned router, AutoApprox tailors approximate circuits specifically for each neural network, achieving up to 3.2% power reduction and 5.2% area savings with zero loss in accuracy on large-scale datasets like ImageNet. Experience the future of energy-efficient deep learning with unparalleled accuracy and efficiency.
 materials:
diff --git a/_pubs/dpo.md b/_pubs/dpo.md
@@ -15,12 +15,12 @@ authors:
   - name: Jeff Dean
 venue: preprint
 year: 2017
-day: 164
+date: 2017-06-13
 has_pdf: false
 doi: NA
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
   - highlight
 teaser: In recent years, neural network training and inference have surged in complexity and resource demands. We introduce a novel approach that leverages a sequence-to-sequence model to optimize device placement for TensorFlow graphs. Our method outperforms traditional heuristics and algorithms by learning optimal device assignments, as demonstrated with Inception-V3 on ImageNet and RNN LSTM models for language tasks.
 materials:
diff --git a/_pubs/fast.md b/_pubs/fast.md
@@ -15,13 +15,13 @@ authors:
     affiliation: Google Brain
   - key: azaliamirhoseini
 venue: preprint
-year: 2022
-day: 53
+year: 2021
+date: 2021-05-26
 has_pdf: false
 doi: 3503222.3507767v
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
   - highlight
 teaser: Explore the cutting-edge Full-stack Accelerator Search Technique (FAST), a game-changing framework designed to optimize hardware accelerators for today's dynamic deep learning demands. This innovative approach fine-tunes every aspect of the hardware-software stack, from datapath design to software scheduling and compiler optimizations. By targeting bottlenecks in leading models like EfficientNet and BERT, FAST creates accelerators that deliver up to 3.7× better performance per watt compared to TPU-v3 for single workloads, and 2.4× better for a range of tasks. Discover how FAST can revolutionize datacenter efficiency and performance.
 materials:
diff --git a/_pubs/gpm.md b/_pubs/gpm.md
@@ -26,12 +26,12 @@ authors:
 
 venue: preprint
 year: 2021
-day: 103
+date: 2021-04-13
 has_pdf: false
 doi: 10.1038'
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
   - highlight
 teaser: Revolutionize chip design with our deep reinforcement learning approach to chip floorplanning. In under six hours, our method generates chip layouts that match or exceed human-designed standards in power, performance, and area. By using an edge-based graph convolutional neural network, we enable AI to learn from past designs and continuously improve. This technology has already shaped the next generation of Google’s AI accelerators and promises to drastically cut human design time while accelerating advances in both AI and hardware.
 materials:
diff --git a/_pubs/hmdp.md b/_pubs/hmdp.md
@@ -11,12 +11,12 @@ authors:
   - name: Jeff Dean
 venue: preprint
 year: 2018
-day: 1
+date: 2018-02-15
 has_pdf: false
 doi: NA
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
   - highlight
 teaser: We present a hierarchical model that optimizes the placement of computational graphs across heterogeneous hardware, including CPUs and GPUs. Our approach, trained with policy gradients and requiring no human input, significantly enhances placement efficiency and achieves up to 60.6% runtime reductions per training step for models like Neural Machine Translation.
 materials:
diff --git a/_pubs/hydragen.md b/_pubs/hydragen.md
@@ -15,12 +15,12 @@ authors:
   - key: azaliamirhoseini
 venue: preprint
 year: 2024
-day: 133
+date: 2024-02-07
 has_pdf: true
 doi: 10.48550/arXiv.2402.05099
 tags:
   - machine learning
-  - generative AI
+  - generative ai
 teaser: Hydragen also enables the use of very long shared contexts with a large batch size, increasing the prefix length from 1K to 16K tokens decreases Hydragen throughput by less than 15%, while the throughput of baselines drops by over 90%. Hydragen generalizes beyond simple prefix-suffix decomposition and can be applied to tree-based prompt sharing patterns, allowing us to further reduce inference time on competitive programming problems by 55%.
 materials:
   - name: Hydragen
diff --git a/_pubs/large_language_monkeys.md b/_pubs/large_language_monkeys.md
@@ -17,12 +17,12 @@ authors:
   - key: azaliamirhoseini
 venue: preprint
 year: 2024
-day: 212
+date: 2024-07-31
 has_pdf: true
 doi: 10.48550/arXiv.2407.21787
 tags:
   - machine learning
-  - generative AI
+  - generative ai
 teaser: Increasing the amount of inference compute through repeated sampling leads to large improvements in coverage - the fraction of problems solved by any attempt - across a variety tasks, models, and sample budgets. This makes it possible, and sometimes cost-effective, to amplify weaker models with many samples and outperform single attempts from more capable models.
 materials:
   - name: Paper
diff --git a/_pubs/olln.md b/_pubs/olln.md
@@ -13,12 +13,12 @@ authors:
   - name: Jeff Dean
 venue: preprint
 year: 2017
-day: 23
+date: 2017-01-23
 has_pdf: false
 doi: 1701.06538'
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
   - highlight
 teaser: This study enhances neural network capacity using conditional computation, specifically through a Sparsely-Gated Mixture-of-Experts (MoE) layer. By activating selective sub-networks per example, the MoE significantly boosts model capacity with minimal computational overhead, improving performance in language modeling and machine translation tasks.
 materials:
diff --git a/_pubs/rlrc.md b/_pubs/rlrc.md
@@ -9,12 +9,12 @@ authors:
   - name: Ion Stocia
 venue: preprint
 year: 2021
-day: 21
+date: 2021-01-21
 has_pdf: true
 doi: NA
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
   - highlight
 teaser: GraphTrans combines GNNs with Transformer-based self-attention to learn long-range relationships and uses a novel "readout" mechanism for global graph embeddings. It achieves state-of-the-art results in graph classification, suggesting that learning-based approaches can effectively capture long-range dependencies.
 materials:
diff --git a/_pubs/tgo.md b/_pubs/tgo.md
@@ -15,12 +15,12 @@ authors:
   - name: James Laudon
 venue: preprint
 year: 2020
-day: 51
+date: 2020-02-20
 has_pdf: false
 doi: 2010.12438'
 tags:
   - natural language processing
-  - generative AI
+  - generative ai
   - highlight
 teaser: Revolutionize graph optimization with GO, our deep reinforcement learning method that outperforms traditional heuristics. By optimizing entire graphs at once, GO delivers 33%-60% faster results and up to 21% better performance compared to previous methods, achieving 15× faster convergence on complex tasks.
 materials:
diff --git a/blogs/index.md b/blogs/index.md
@@ -6,7 +6,7 @@ layout: page
   <div id="content" class="pure-u-1 pure-u-md-3-4">
     <h1 class="title">Blogs</h1>
 
-    {% assign blogYears = site.blogs | group_by:"year" | sort: "name" | reverse %}
+    {% assign blogYears = site.blogs | group_by:"year" | sort: "date" | reverse %}
     {% for year in blogYears %}
       <div id="year-{{year.name}}" class="year pure-g">
         <div class="pure-u-1-3 pure-u-md-1-5"></div>
diff --git a/pubs/index.md b/pubs/index.md
