Changes

              
    

          
          
        
        

            
f
1
{
f
1
{
            
2
  "access_rights": "",
2
  "access_rights": "",
            
3
  "author": "Haojun Xia",
3
  "author": "Haojun Xia",
            
4
  "author_email": "",
4
  "author_email": "",
            
5
  "citation": [],
5
  "citation": [],
            
6
  "creator_user_id": "17755db4-395a-4b3b-ac09-e8e3484ca700",
6
  "creator_user_id": "17755db4-395a-4b3b-ac09-e8e3484ca700",
            
7
  "defined_in": "https://doi.org/10.48550/arXiv.2401.14112",
7
  "defined_in": "https://doi.org/10.48550/arXiv.2401.14112",
            
8
  "doi": "10.57702/z4ohr5qr",
8
  "doi": "10.57702/z4ohr5qr",
            
n
9
  "doi_date_published": null,
n
9
  "doi_date_published": "2025-01-03",
            
10
  "doi_publisher": "TIB",
10
  "doi_publisher": "TIB",
            
n
11
  "doi_status": false,
n
11
  "doi_status": true,
            
12
  "domain": "https://service.tib.eu/ldmservice",
12
  "domain": "https://service.tib.eu/ldmservice",
            
13
  "extra_authors": [
13
  "extra_authors": [
            
14
    {
14
    {
            
15
      "extra_author": "Zhen Zheng",
15
      "extra_author": "Zhen Zheng",
            
16
      "orcid": ""
16
      "orcid": ""
            
17
    },
17
    },
            
18
    {
18
    {
            
19
      "extra_author": "Xiaoxia Wu",
19
      "extra_author": "Xiaoxia Wu",
            
20
      "orcid": ""
20
      "orcid": ""
            
21
    },
21
    },
            
22
    {
22
    {
            
23
      "extra_author": "Shiyang Chen",
23
      "extra_author": "Shiyang Chen",
            
24
      "orcid": ""
24
      "orcid": ""
            
25
    },
25
    },
            
26
    {
26
    {
            
27
      "extra_author": "Zhewei Yao",
27
      "extra_author": "Zhewei Yao",
            
28
      "orcid": ""
28
      "orcid": ""
            
29
    },
29
    },
            
30
    {
30
    {
            
31
      "extra_author": "Stephen Youn",
31
      "extra_author": "Stephen Youn",
            
32
      "orcid": ""
32
      "orcid": ""
            
33
    },
33
    },
            
34
    {
34
    {
            
35
      "extra_author": "Arash Bakhtiari",
35
      "extra_author": "Arash Bakhtiari",
            
36
      "orcid": ""
36
      "orcid": ""
            
37
    },
37
    },
            
38
    {
38
    {
            
39
      "extra_author": "Michael Wyatt",
39
      "extra_author": "Michael Wyatt",
            
40
      "orcid": ""
40
      "orcid": ""
            
41
    },
41
    },
            
42
    {
42
    {
            
43
      "extra_author": "Yuxiong He",
43
      "extra_author": "Yuxiong He",
            
44
      "orcid": ""
44
      "orcid": ""
            
45
    },
45
    },
            
46
    {
46
    {
            
47
      "extra_author": "Olatunji Ruwase",
47
      "extra_author": "Olatunji Ruwase",
            
48
      "orcid": ""
48
      "orcid": ""
            
49
    },
49
    },
            
50
    {
50
    {
            
51
      "extra_author": "Shuaiwen Leon Song",
51
      "extra_author": "Shuaiwen Leon Song",
            
52
      "orcid": ""
52
      "orcid": ""
            
53
    }
53
    }
            
54
  ],
54
  ],
            
55
  "groups": [
55
  "groups": [
            
56
    {
56
    {
            
57
      "description": "",
57
      "description": "",
            
58
      "display_name": "Deep Learning",
58
      "display_name": "Deep Learning",
            
59
      "id": "d2734132-7098-4cc5-9f4c-5f9b6e1d7922",
59
      "id": "d2734132-7098-4cc5-9f4c-5f9b6e1d7922",
            
60
      "image_display_url": "",
60
      "image_display_url": "",
            
61
      "name": "deep-learning",
61
      "name": "deep-learning",
            
62
      "title": "Deep Learning"
62
      "title": "Deep Learning"
            
63
    },
63
    },
            
64
    {
64
    {
            
65
      "description": "",
65
      "description": "",
            
66
      "display_name": "Machine Learning",
66
      "display_name": "Machine Learning",
            
67
      "id": "1d8623c9-adbd-4f91-be1e-53847c4ac32a",
67
      "id": "1d8623c9-adbd-4f91-be1e-53847c4ac32a",
            
68
      "image_display_url": "",
68
      "image_display_url": "",
            
69
      "name": "machine-learning",
69
      "name": "machine-learning",
            
70
      "title": "Machine Learning"
70
      "title": "Machine Learning"
            
71
    },
71
    },
            
72
    {
72
    {
            
73
      "description": "",
73
      "description": "",
            
74
      "display_name": "Natural Language Processing",
74
      "display_name": "Natural Language Processing",
            
75
      "id": "5b974bcc-8f79-40fc-a05d-75b861d55ed0",
75
      "id": "5b974bcc-8f79-40fc-a05d-75b861d55ed0",
            
76
      "image_display_url": "",
76
      "image_display_url": "",
            
77
      "name": "natural-language-processing",
77
      "name": "natural-language-processing",
            
78
      "title": "Natural Language Processing"
78
      "title": "Natural Language Processing"
            
79
    }
79
    }
            
80
  ],
80
  ],
            
81
  "id": "1fc4f86e-19be-49e1-8a09-b9fffd004254",
81
  "id": "1fc4f86e-19be-49e1-8a09-b9fffd004254",
            
82
  "isopen": false,
82
  "isopen": false,
            
83
  "landing_page": "https://github.com/usyd-fsalab/fp6_llm",
83
  "landing_page": "https://github.com/usyd-fsalab/fp6_llm",
            
84
  "license_title": null,
84
  "license_title": null,
            
85
  "link_orkg": "",
85
  "link_orkg": "",
            
86
  "metadata_created": "2025-01-03T00:14:28.468137",
86
  "metadata_created": "2025-01-03T00:14:28.468137",
            
n
87
  "metadata_modified": "2025-01-03T00:14:28.468144",
n
87
  "metadata_modified": "2025-01-03T00:14:29.094100",
            
88
  "name": 
88
  "name": 
            
89
large-language-models-through-fp6-centric-algorithm-system-co-design",
89
large-language-models-through-fp6-centric-algorithm-system-co-design",
            
90
  "notes": "Six-bit quantization can effectively reduce the size of 
90
  "notes": "Six-bit quantization can effectively reduce the size of 
            
91
large language models and preserve the model quality consistently 
91
large language models and preserve the model quality consistently 
            
92
across varied applications.",
92
across varied applications.",
            
n
93
  "num_resources": 0,
n
93
  "num_resources": 1,
            
94
  "num_tags": 5,
94
  "num_tags": 5,
            
95
  "organization": {
95
  "organization": {
            
96
    "approval_status": "approved",
96
    "approval_status": "approved",
            
97
    "created": "2024-11-25T12:11:38.292601",
97
    "created": "2024-11-25T12:11:38.292601",
            
98
    "description": "",
98
    "description": "",
            
99
    "id": "079d46db-32df-4b48-91f3-0a8bc8f69559",
99
    "id": "079d46db-32df-4b48-91f3-0a8bc8f69559",
            
100
    "image_url": "",
100
    "image_url": "",
            
101
    "is_organization": true,
101
    "is_organization": true,
            
102
    "name": "no-organization",
102
    "name": "no-organization",
            
103
    "state": "active",
103
    "state": "active",
            
104
    "title": "No Organization",
104
    "title": "No Organization",
            
105
    "type": "organization"
105
    "type": "organization"
            
106
  },
106
  },
            
107
  "owner_org": "079d46db-32df-4b48-91f3-0a8bc8f69559",
107
  "owner_org": "079d46db-32df-4b48-91f3-0a8bc8f69559",
            
108
  "private": false,
108
  "private": false,
            
109
  "relationships_as_object": [],
109
  "relationships_as_object": [],
            
110
  "relationships_as_subject": [],
110
  "relationships_as_subject": [],
            
t
111
  "resources": [],
t
111
  "resources": [
            
112
    {
            
113
      "cache_last_updated": null,
            
114
      "cache_url": null,
            
115
      "created": "2025-01-03T00:16:32",
            
116
      "data": [
            
117
        "dcterms:title",
            
118
        "dcterms:accessRights",
            
119
        "dcterms:creator",
            
120
        "dcterms:description",
            
121
        "dcterms:issued",
            
122
        "dcterms:language",
            
123
        "dcterms:identifier",
            
124
        "dcat:theme",
            
125
        "dcterms:type",
            
126
        "dcat:keyword",
            
127
        "dcat:landingPage",
            
128
        "dcterms:hasVersion",
            
129
        "dcterms:format",
            
130
        "mls:task",
            
131
        "datacite:isDescribedBy"
            
132
      ],
            
133
      "description": "The json representation of the dataset with its 
            
134
distributions based on DCAT.",
            
135
      "format": "JSON",
            
136
      "hash": "",
            
137
      "id": "3d656b1f-27d1-446e-9732-74b0608b9eaa",
            
138
      "last_modified": "2025-01-03T00:14:29.085730",
            
139
      "metadata_modified": "2025-01-03T00:14:29.097008",
            
140
      "mimetype": "application/json",
            
141
      "mimetype_inner": null,
            
142
      "name": "Original Metadata",
            
143
      "package_id": "1fc4f86e-19be-49e1-8a09-b9fffd004254",
            
144
      "position": 0,
            
145
      "resource_type": null,
            
146
      "size": 1058,
            
147
      "state": "active",
            
148
      "url": 
            
149
resource/3d656b1f-27d1-446e-9732-74b0608b9eaa/download/metadata.json",
            
150
      "url_type": "upload"
            
151
    }
            
152
  ],
            
112
  "services_used_list": "",
153
  "services_used_list": "",
            
113
  "state": "active",
154
  "state": "active",
            
114
  "tags": [
155
  "tags": [
            
115
    {
156
    {
            
116
      "display_name": "6-bit quantization",
157
      "display_name": "6-bit quantization",
            
117
      "id": "e910c710-c042-4a44-9192-52a89fc5ef6e",
158
      "id": "e910c710-c042-4a44-9192-52a89fc5ef6e",
            
118
      "name": "6-bit quantization",
159
      "name": "6-bit quantization",
            
119
      "state": "active",
160
      "state": "active",
            
120
      "vocabulary_id": null
161
      "vocabulary_id": null
            
121
    },
162
    },
            
122
    {
163
    {
            
123
      "display_name": "FP6",
164
      "display_name": "FP6",
            
124
      "id": "7820fae3-55a6-4b64-b6dc-7f3fcc495668",
165
      "id": "7820fae3-55a6-4b64-b6dc-7f3fcc495668",
            
125
      "name": "FP6",
166
      "name": "FP6",
            
126
      "state": "active",
167
      "state": "active",
            
127
      "vocabulary_id": null
168
      "vocabulary_id": null
            
128
    },
169
    },
            
129
    {
170
    {
            
130
      "display_name": "Large Language Models",
171
      "display_name": "Large Language Models",
            
131
      "id": "6e366b4a-402d-4c0e-b818-6fec429fc72f",
172
      "id": "6e366b4a-402d-4c0e-b818-6fec429fc72f",
            
132
      "name": "Large Language Models",
173
      "name": "Large Language Models",
            
133
      "state": "active",
174
      "state": "active",
            
134
      "vocabulary_id": null
175
      "vocabulary_id": null
            
135
    },
176
    },
            
136
    {
177
    {
            
137
      "display_name": "SIMT Cores",
178
      "display_name": "SIMT Cores",
            
138
      "id": "f7e24ad7-cd4d-4882-a395-2923fe8d0ea8",
179
      "id": "f7e24ad7-cd4d-4882-a395-2923fe8d0ea8",
            
139
      "name": "SIMT Cores",
180
      "name": "SIMT Cores",
            
140
      "state": "active",
181
      "state": "active",
            
141
      "vocabulary_id": null
182
      "vocabulary_id": null
            
142
    },
183
    },
            
143
    {
184
    {
            
144
      "display_name": "Tensor Cores",
185
      "display_name": "Tensor Cores",
            
145
      "id": "d841aedb-9d68-40ee-a1cc-b8f023b43841",
186
      "id": "d841aedb-9d68-40ee-a1cc-b8f023b43841",
            
146
      "name": "Tensor Cores",
187
      "name": "Tensor Cores",
            
147
      "state": "active",
188
      "state": "active",
            
148
      "vocabulary_id": null
189
      "vocabulary_id": null
            
149
    }
190
    }
            
150
  ],
191
  ],
            
151
  "title": "FP6-LLM: Efficiently Serving Large Language Models Through 
192
  "title": "FP6-LLM: Efficiently Serving Large Language Models Through 
            
152
FP6-Centric Algorithm-System Co-Design",
193
FP6-Centric Algorithm-System Co-Design",
            
153
  "type": "dataset",
194
  "type": "dataset",
            
154
  "version": ""
195
  "version": ""
            
155
}
196
}