Spaces:
Running
Running
Update index.html
Browse files- index.html +11 -97
index.html
CHANGED
@@ -61,81 +61,6 @@
|
|
61 |
<div class="columns is-centered">
|
62 |
<div class="column is-10">
|
63 |
|
64 |
-
<!-- Native Sparse Attention -->
|
65 |
-
<div class="card paper-card">
|
66 |
-
<div class="card-content">
|
67 |
-
<h3 class="title is-4">
|
68 |
-
<a href="https://arxiv.org/abs/2502.11089">Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention</a>
|
69 |
-
<span class="coming-soon-badge">Deep Dive Coming Soon</span>
|
70 |
-
</h3>
|
71 |
-
<p class="release-date">Released: February 2025</p>
|
72 |
-
<p class="paper-description">
|
73 |
-
Introduces a new approach to sparse attention that is both hardware-efficient and natively trainable,
|
74 |
-
improving the performance of large language models.
|
75 |
-
</p>
|
76 |
-
</div>
|
77 |
-
</div>
|
78 |
-
|
79 |
-
<!-- DeepSeek-R1 -->
|
80 |
-
<div class="card paper-card">
|
81 |
-
<div class="card-content">
|
82 |
-
<h3 class="title is-4">
|
83 |
-
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
|
84 |
-
<span class="coming-soon-badge">Deep Dive Coming Soon</span>
|
85 |
-
</h3>
|
86 |
-
<p class="release-date">Released: January 20, 2025</p>
|
87 |
-
<p class="paper-description">
|
88 |
-
The R1 model builds on previous work to enhance reasoning capabilities through large-scale
|
89 |
-
reinforcement learning, competing directly with leading models like OpenAI's o1.
|
90 |
-
</p>
|
91 |
-
</div>
|
92 |
-
</div>
|
93 |
-
|
94 |
-
<!-- DeepSeek-V3 -->
|
95 |
-
<div class="card paper-card">
|
96 |
-
<div class="card-content">
|
97 |
-
<h3 class="title is-4">
|
98 |
-
DeepSeek-V3 Technical Report
|
99 |
-
<span class="coming-soon-badge">Deep Dive Coming Soon</span>
|
100 |
-
</h3>
|
101 |
-
<p class="release-date">Released: December 2024</p>
|
102 |
-
<p class="paper-description">
|
103 |
-
Discusses the scaling of sparse MoE networks to 671 billion parameters, utilizing mixed precision
|
104 |
-
training and high-performance computing (HPC) co-design strategies.
|
105 |
-
</p>
|
106 |
-
</div>
|
107 |
-
</div>
|
108 |
-
|
109 |
-
<!-- DeepSeek-V2 -->
|
110 |
-
<div class="card paper-card">
|
111 |
-
<div class="card-content">
|
112 |
-
<h3 class="title is-4">
|
113 |
-
DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
|
114 |
-
<span class="coming-soon-badge">Deep Dive Coming Soon</span>
|
115 |
-
</h3>
|
116 |
-
<p class="release-date">Released: May 2024</p>
|
117 |
-
<p class="paper-description">
|
118 |
-
Introduces a Mixture-of-Experts (MoE) architecture, enhancing performance while reducing
|
119 |
-
training costs by 42%. Emphasizes strong performance characteristics and efficiency improvements.
|
120 |
-
</p>
|
121 |
-
</div>
|
122 |
-
</div>
|
123 |
-
|
124 |
-
<!-- DeepSeekMath -->
|
125 |
-
<div class="card paper-card">
|
126 |
-
<div class="card-content">
|
127 |
-
<h3 class="title is-4">
|
128 |
-
DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
|
129 |
-
<span class="coming-soon-badge">Deep Dive Coming Soon</span>
|
130 |
-
</h3>
|
131 |
-
<p class="release-date">Released: April 2024</p>
|
132 |
-
<p class="paper-description">
|
133 |
-
This paper presents methods to improve mathematical reasoning in LLMs, introducing the
|
134 |
-
Group Relative Policy Optimization (GRPO) algorithm during reinforcement learning stages.
|
135 |
-
</p>
|
136 |
-
</div>
|
137 |
-
</div>
|
138 |
-
|
139 |
<!-- DeepSeekLLM -->
|
140 |
<div class="card paper-card">
|
141 |
<div class="card-content">
|
@@ -151,48 +76,37 @@
|
|
151 |
</div>
|
152 |
</div>
|
153 |
|
154 |
-
<!--
|
155 |
-
<!-- DeepSeek-Prover -->
|
156 |
<div class="card paper-card">
|
157 |
<div class="card-content">
|
158 |
<h3 class="title is-4">
|
159 |
-
DeepSeek-
|
160 |
<span class="coming-soon-badge">Deep Dive Coming Soon</span>
|
161 |
</h3>
|
|
|
162 |
<p class="paper-description">
|
163 |
-
|
164 |
-
|
165 |
</p>
|
166 |
</div>
|
167 |
</div>
|
168 |
|
169 |
-
<!--
|
170 |
<div class="card paper-card">
|
171 |
<div class="card-content">
|
172 |
<h3 class="title is-4">
|
173 |
-
DeepSeek-
|
174 |
<span class="coming-soon-badge">Deep Dive Coming Soon</span>
|
175 |
</h3>
|
|
|
176 |
<p class="paper-description">
|
177 |
-
|
178 |
-
|
179 |
</p>
|
180 |
</div>
|
181 |
</div>
|
182 |
|
183 |
-
<!--
|
184 |
-
<div class="card paper-card">
|
185 |
-
<div class="card-content">
|
186 |
-
<h3 class="title is-4">
|
187 |
-
DeepSeekMoE: Advancing Mixture-of-Experts Architecture
|
188 |
-
<span class="coming-soon-badge">Deep Dive Coming Soon</span>
|
189 |
-
</h3>
|
190 |
-
<p class="paper-description">
|
191 |
-
Discusses the integration and benefits of the Mixture-of-Experts approach within the
|
192 |
-
DeepSeek framework, focusing on scalability and efficiency improvements.
|
193 |
-
</p>
|
194 |
-
</div>
|
195 |
-
</div>
|
196 |
|
197 |
</div>
|
198 |
</div>
|
|
|
61 |
<div class="columns is-centered">
|
62 |
<div class="column is-10">
|
63 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
<!-- DeepSeekLLM -->
|
65 |
<div class="card paper-card">
|
66 |
<div class="card-content">
|
|
|
76 |
</div>
|
77 |
</div>
|
78 |
|
79 |
+
<!-- DeepSeek-V2 -->
|
|
|
80 |
<div class="card paper-card">
|
81 |
<div class="card-content">
|
82 |
<h3 class="title is-4">
|
83 |
+
DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
|
84 |
<span class="coming-soon-badge">Deep Dive Coming Soon</span>
|
85 |
</h3>
|
86 |
+
<p class="release-date">Released: May 2024</p>
|
87 |
<p class="paper-description">
|
88 |
+
Introduces a Mixture-of-Experts (MoE) architecture, enhancing performance while reducing
|
89 |
+
training costs by 42%. Emphasizes strong performance characteristics and efficiency improvements.
|
90 |
</p>
|
91 |
</div>
|
92 |
</div>
|
93 |
|
94 |
+
<!-- Continue with other papers... -->
|
95 |
<div class="card paper-card">
|
96 |
<div class="card-content">
|
97 |
<h3 class="title is-4">
|
98 |
+
DeepSeek-V3 Technical Report
|
99 |
<span class="coming-soon-badge">Deep Dive Coming Soon</span>
|
100 |
</h3>
|
101 |
+
<p class="release-date">Released: December 2024</p>
|
102 |
<p class="paper-description">
|
103 |
+
Discusses the scaling of sparse MoE networks to 671 billion parameters, utilizing mixed precision
|
104 |
+
training and high-performance computing (HPC) co-design strategies.
|
105 |
</p>
|
106 |
</div>
|
107 |
</div>
|
108 |
|
109 |
+
<!-- Add remaining papers following the same pattern -->
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
110 |
|
111 |
</div>
|
112 |
</div>
|