File size: 6,213 Bytes
0c0f086
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b63ff12
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0c0f086
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
07482ee
 
633e287
07482ee
 
 
 
 
 
 
633e287
07482ee
 
 
 
 
 
 
633e287
07482ee
 
 
 
 
0c0f086
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
"""Module containing performance results for the Arcee-Nova model."""

results_arcee_nova = {
    "name": "Arcee-Nova",
    "modelType": "Qwen2 72B",
    "notes": "",
    "configurations": [
        {
            "instanceType": "g4dn.12xlarge",
            "quantization": "bitsandbytes-nf4",
            "container": "TGI 2.2.0",
            "status": "KO",
            "tokensPerSecond": "-",
            "notes": "Flash Attention requires Ampere GPUs or newer",
        },
        {
            "instanceType": "g5.12xlarge",
            "configurations": [
                {
                    "quantization": "bitsandbytes-nf4",
                    "container": "TGI 2.2.0",
                    "status": "OK",
                    "tokensPerSecond": "12",
                },
                {
                    "quantization": "bitsandbytes-fp4",
                    "container": "TGI 2.2.0",
                    "status": "OK",
                    "tokensPerSecond": "12",
                },
                {
                    "quantization": "bitsandbytes (int8)",
                    "container": "TGI 2.2.0",
                    "status": "KO",
                    "tokensPerSecond": "-",
                    "notes": "CUDA OOM",
                },
                {
                    "quantization": "eetq (int8)",
                    "container": "TGI 2.2.0",
                    "status": "KO",
                    "tokensPerSecond": "-",
                    "notes": "[FT Error] Heurisitc failed to find a valid config.",
                },
            ],
        },
        {
            "instanceType": "g5.48xlarge",
            "configurations": [
                {
                    "quantization": "none",
                    "container": "TGI 2.2.0",
                    "status": "KO",
                    "tokensPerSecond": "-",
                    "notes": "CUDA OOM (but g6.48xlarge works!)",
                },
                {
                    "quantization": "bitsandbytes-nf4",
                    "container": "TGI 2.2.0",
                    "status": "OK",
                    "tokensPerSecond": "12.3",
                },
                {
                    "quantization": "bitsandbytes-fp4",
                    "container": "TGI 2.2.0",
                    "status": "OK",
                    "tokensPerSecond": "12.5",
                },
                {
                    "quantization": "bitsandbytes (int8)",
                    "container": "TGI 2.2.0",
                    "status": "KO",
                    "tokensPerSecond": "-",
                    "notes": "The model deploys, but inference times out.",
                },
            ],
        },
        {
            "instanceType": "g6.12xlarge",
            "configurations": [
                {
                    "quantization": "bitsandbytes-nf4",
                    "container": "TGI 2.2.0",
                    "status": "OK",
                    "tokensPerSecond": "1.5-2",
                    "notes": "Too slow, timeouts are likely",
                },
                {
                    "quantization": "bitsandbytes-fp4",
                    "container": "TGI 2.2.0",
                    "status": "OK",
                    "tokensPerSecond": "2",
                    "notes": "Too slow, timeouts are likely",
                },
                {
                    "quantization": "bitsandbytes (int8)",
                    "container": "TGI 2.2.0",
                    "status": "KO",
                    "tokensPerSecond": "-",
                    "notes": "CUDA OOM",
                },
            ],
        },
        {
            "instanceType": "g6.48xlarge",
            "quantization": "none",
            "container": "TGI 2.2.0",
            "status": "OK",
            "tokensPerSecond": "12",
        },
        {
            "instanceType": "g6e.12xlarge",
            "configurations": [
                {
                    "quantization": "none",
                    "container": "TGI 2.2.0",
                    "status": "OK",
                    "tokensPerSecond": "17",
                },
                {
                    "quantization": "none",
                    "container": "vLLM 0.5.5",
                    "status": "OK",
                    "tokensPerSecond": "17.8",
                },
                {
                    "quantization": "none",
                    "container": "SGLang 0.2.13",
                    "status": "OK",
                    "tokensPerSecond": "18.2",
                },
            ],
        },
        {
            "instanceType": "p4d.24xlarge",
            "quantization": "none",
            "container": "TGI 2.2.0",
            "status": "OK",
            "tokensPerSecond": "40",
            "notes": '"MAX_INPUT_LENGTH": "16384", "MAX_TOTAL_TOKENS": "32768",',
        },
        {
            "instanceType": "p4de.24xlarge",
            "quantization": "none",
            "container": "TGI 2.2.0",
            "status": "waiting for quota",
        },
        {
            "instanceType": "p5.48xlarge",
            "quantization": "none",
            "container": "TGI 2.2.0",
            "status": "OK",
            "tokensPerSecond": "58",
            "notes": '"MAX_INPUT_LENGTH": "16384", "MAX_TOTAL_TOKENS": "32768",',
        },
        {
            "instanceType": "p5.48xlarge",
            "quantization": "none",
            "container": "vLLM 0.6.4.post1",
            "status": "OK",
            "tokensPerSecond": "76",
            "notes": "--tensor-parallel-size 8",
        },
        {
            "instanceType": "p5.48xlarge (4 GPUs)",
            "quantization": "none",
            "container": "vLLM 0.6.4.post1",
            "status": "OK",
            "tokensPerSecond": "51",
            "notes": "--tensor-parallel-size 4",
        },
        {
            "instanceType": "p5.48xlarge (2 GPUs)",
            "quantization": "none",
            "container": "vLLM 0.6.4.post1",
            "status": "OK",
            "tokensPerSecond": "32",
            "notes": "--tensor-parallel-size 2 --max-model-len 16384 --gpu_memory-utilization 0.95",
        },
    ],
}