|
5 | 5 |
|
6 | 6 | # The "args" values in TEST_CONFIGS are not optimized for speed but only check that the inference is working for the different models architectures. |
7 | 7 | TEST_CONFIGS = { |
8 | | - "meta-llama/Llama-3.1-8B-Instruct-shared": { |
9 | | - "model_id": "meta-llama/Llama-3.1-8B-Instruct", |
10 | | - "input": "What is Deep Learning?", |
11 | | - "expected_greedy_output": " A Beginner’s Guide\nDeep learning is a subset of machine learning that involves the use of artificial neural networks to analyze and interpret data. It is a type of", |
12 | | - "expected_batch_output": " A Beginner’s Guide\nDeep learning is a subset of machine learning that involves the use of artificial neural networks to analyze and interpret data. It is a type of", |
13 | | - "args": [ |
14 | | - "--sharded", |
15 | | - "true", |
16 | | - "--num-shard", |
17 | | - "8", |
18 | | - "--max-input-tokens", |
19 | | - "512", |
20 | | - "--max-total-tokens", |
21 | | - "1024", |
22 | | - "--max-batch-size", |
23 | | - "8", |
24 | | - "--max-batch-prefill-tokens", |
25 | | - "2048", |
26 | | - ], |
27 | | - }, |
| 8 | + # "meta-llama/Llama-3.1-8B-Instruct-shared": { |
| 9 | + # "model_id": "meta-llama/Llama-3.1-8B-Instruct", |
| 10 | + # "input": "What is Deep Learning?", |
| 11 | + # "expected_greedy_output": " A Beginner’s Guide\nDeep learning is a subset of machine learning that involves the use of artificial neural networks to analyze and interpret data. It is a type of", |
| 12 | + # "expected_batch_output": " A Beginner’s Guide\nDeep learning is a subset of machine learning that involves the use of artificial neural networks to analyze and interpret data. It is a type of", |
| 13 | + # "args": [ |
| 14 | + # "--sharded", |
| 15 | + # "true", |
| 16 | + # "--num-shard", |
| 17 | + # "8", |
| 18 | + # "--max-input-tokens", |
| 19 | + # "512", |
| 20 | + # "--max-total-tokens", |
| 21 | + # "1024", |
| 22 | + # "--max-batch-size", |
| 23 | + # "8", |
| 24 | + # "--max-batch-prefill-tokens", |
| 25 | + # "2048", |
| 26 | + # ], |
| 27 | + # }, |
28 | 28 | "meta-llama/Llama-3.1-8B-Instruct": { |
29 | 29 | "model_id": "meta-llama/Llama-3.1-8B-Instruct", |
30 | 30 | "input": "What is Deep Learning?", |
|
42 | 42 | "2048", |
43 | 43 | ], |
44 | 44 | }, |
45 | | - "meta-llama/Llama-2-7b-chat-hf": { |
46 | | - "model_id": "meta-llama/Llama-2-7b-chat-hf", |
47 | | - "input": "What is Deep Learning?", |
48 | | - "expected_greedy_output": "\n\nDeep learning (also known as deep structured learning) is part of a broader family of machine learning techniques based on artificial neural networks\u2014specific", |
49 | | - "expected_batch_output": "\n\nDeep learning (also known as deep structured learning) is part of a broader family of machine learning techniques based on artificial neural networks\u2014specific", |
50 | | - "args": [ |
51 | | - "--max-input-tokens", |
52 | | - "512", |
53 | | - "--max-total-tokens", |
54 | | - "1024", |
55 | | - "--max-batch-size", |
56 | | - "4", |
57 | | - "--max-batch-prefill-tokens", |
58 | | - "2048", |
59 | | - ], |
60 | | - }, |
61 | | - "mistralai/Mistral-7B-Instruct-v0.3": { |
62 | | - "model_id": "mistralai/Mistral-7B-Instruct-v0.3", |
63 | | - "input": "What is Deep Learning?", |
64 | | - "expected_greedy_output": "\n\nDeep learning is a subset of machine learning in artificial intelligence (AI) that has networks capable of learning unsupervised from data that is unstructured", |
65 | | - "expected_batch_output": "\n\nDeep learning is a subset of machine learning in artificial intelligence (AI) that has networks capable of learning unsupervised from data that is unstructured", |
66 | | - "args": [ |
67 | | - "--max-input-tokens", |
68 | | - "512", |
69 | | - "--max-total-tokens", |
70 | | - "1024", |
71 | | - "--max-batch-size", |
72 | | - "4", |
73 | | - "--max-batch-prefill-tokens", |
74 | | - "2048", |
75 | | - ], |
76 | | - }, |
77 | | - "bigcode/starcoder2-3b": { |
78 | | - "model_id": "bigcode/starcoder2-3b", |
79 | | - "input": "What is Deep Learning?", |
80 | | - "expected_greedy_output": "\n\nDeep learning is a subset of machine learning that uses artificial neural networks to perform tasks.\n\nNeural networks are a type of machine learning algorithm that", |
81 | | - "expected_batch_output": "\n\nDeep learning is a subset of machine learning that uses artificial neural networks to perform tasks.\n\nNeural networks are a type of machine learning algorithm that", |
82 | | - "args": [ |
83 | | - "--max-input-tokens", |
84 | | - "512", |
85 | | - "--max-total-tokens", |
86 | | - "1024", |
87 | | - "--max-batch-size", |
88 | | - "4", |
89 | | - "--max-batch-prefill-tokens", |
90 | | - "2048", |
91 | | - ], |
92 | | - }, |
93 | | - "google/gemma-7b-it": { |
94 | | - "model_id": "google/gemma-7b-it", |
95 | | - "input": "What is Deep Learning?", |
96 | | - "expected_greedy_output": "\n\nDeep learning is a subset of machine learning that uses artificial neural networks to learn from large amounts of data. Neural networks are inspired by the structure and function of", |
97 | | - "expected_batch_output": "\n\nDeep learning is a subset of machine learning that uses artificial neural networks to learn from large amounts of data. Neural networks are inspired by the structure and function of", |
98 | | - "args": [ |
99 | | - "--max-input-tokens", |
100 | | - "512", |
101 | | - "--max-total-tokens", |
102 | | - "1024", |
103 | | - "--max-batch-size", |
104 | | - "4", |
105 | | - "--max-batch-prefill-tokens", |
106 | | - "2048", |
107 | | - ], |
108 | | - }, |
109 | | - "Qwen/Qwen2-0.5B-Instruct": { |
110 | | - "model_id": "Qwen/Qwen2-0.5B-Instruct", |
111 | | - "input": "What is Deep Learning?", |
112 | | - "expected_greedy_output": " Deep Learning is a type of machine learning that is based on the principles of artificial neural networks. It is a type of machine learning that is used to train models", |
113 | | - "expected_batch_output": " Deep Learning is a type of machine learning that is based on the principles of artificial neural networks. It is a type of machine learning that is used to train models", |
114 | | - "args": [ |
115 | | - "--max-input-tokens", |
116 | | - "512", |
117 | | - "--max-total-tokens", |
118 | | - "1024", |
119 | | - "--max-batch-size", |
120 | | - "4", |
121 | | - "--max-batch-prefill-tokens", |
122 | | - "2048", |
123 | | - ], |
124 | | - }, |
125 | | - "tiiuae/falcon-7b-instruct": { |
126 | | - "model_id": "tiiuae/falcon-7b-instruct", |
127 | | - "input": "What is Deep Learning?", |
128 | | - "expected_greedy_output": "\nDeep learning is a branch of machine learning that uses artificial neural networks to learn and make decisions. It is based on the concept of hierarchical learning, where a", |
129 | | - "expected_batch_output": "\nDeep learning is a branch of machine learning that uses artificial neural networks to learn and make decisions. It is based on the concept of hierarchical learning, where a", |
130 | | - "args": [ |
131 | | - "--max-input-tokens", |
132 | | - "512", |
133 | | - "--max-total-tokens", |
134 | | - "1024", |
135 | | - "--max-batch-size", |
136 | | - "4", |
137 | | - ], |
138 | | - }, |
139 | | - "microsoft/phi-1_5": { |
140 | | - "model_id": "microsoft/phi-1_5", |
141 | | - "input": "What is Deep Learning?", |
142 | | - "expected_greedy_output": "\n\nDeep Learning is a subfield of Machine Learning that focuses on building neural networks with multiple layers of interconnected nodes. These networks are designed to learn from large", |
143 | | - "expected_batch_output": "\n\nDeep Learning is a subfield of Machine Learning that focuses on building neural networks with multiple layers of interconnected nodes. These networks are designed to learn from large", |
144 | | - "args": [ |
145 | | - "--max-input-tokens", |
146 | | - "512", |
147 | | - "--max-total-tokens", |
148 | | - "1024", |
149 | | - "--max-batch-size", |
150 | | - "4", |
151 | | - ], |
152 | | - }, |
153 | | - "openai-community/gpt2": { |
154 | | - "model_id": "openai-community/gpt2", |
155 | | - "input": "What is Deep Learning?", |
156 | | - "expected_greedy_output": "\n\nDeep learning is a new field of research that has been around for a long time. It is a new field of research that has been around for a", |
157 | | - "expected_batch_output": "\n\nDeep learning is a new field of research that has been around for a long time. It is a new field of research that has been around for a", |
158 | | - "args": [ |
159 | | - "--max-input-tokens", |
160 | | - "512", |
161 | | - "--max-total-tokens", |
162 | | - "1024", |
163 | | - "--max-batch-size", |
164 | | - "4", |
165 | | - ], |
166 | | - }, |
167 | | - "facebook/opt-125m": { |
168 | | - "model_id": "facebook/opt-125m", |
169 | | - "input": "What is Deep Learning?", |
170 | | - "expected_greedy_output": "\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout", |
171 | | - "expected_batch_output": "\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout", |
172 | | - "args": [ |
173 | | - "--max-input-tokens", |
174 | | - "512", |
175 | | - "--max-total-tokens", |
176 | | - "1024", |
177 | | - "--max-batch-size", |
178 | | - "4", |
179 | | - ], |
180 | | - }, |
181 | | - "EleutherAI/gpt-j-6b": { |
182 | | - "model_id": "EleutherAI/gpt-j-6b", |
183 | | - "input": "What is Deep Learning?", |
184 | | - "expected_greedy_output": "\n\nDeep learning is a subset of machine learning that is based on the idea of neural networks. Neural networks are a type of artificial intelligence that is inspired by", |
185 | | - "expected_batch_output": "\n\nDeep learning is a subset of machine learning that is based on the idea of neural networks. Neural networks are a type of artificial intelligence that is inspired by", |
186 | | - "args": [ |
187 | | - "--max-input-tokens", |
188 | | - "512", |
189 | | - "--max-total-tokens", |
190 | | - "1024", |
191 | | - "--max-batch-size", |
192 | | - "4", |
193 | | - ], |
194 | | - }, |
| 45 | + # "meta-llama/Llama-2-7b-chat-hf": { |
| 46 | + # "model_id": "meta-llama/Llama-2-7b-chat-hf", |
| 47 | + # "input": "What is Deep Learning?", |
| 48 | + # "expected_greedy_output": "\n\nDeep learning (also known as deep structured learning) is part of a broader family of machine learning techniques based on artificial neural networks\u2014specific", |
| 49 | + # "expected_batch_output": "\n\nDeep learning (also known as deep structured learning) is part of a broader family of machine learning techniques based on artificial neural networks\u2014specific", |
| 50 | + # "args": [ |
| 51 | + # "--max-input-tokens", |
| 52 | + # "512", |
| 53 | + # "--max-total-tokens", |
| 54 | + # "1024", |
| 55 | + # "--max-batch-size", |
| 56 | + # "4", |
| 57 | + # "--max-batch-prefill-tokens", |
| 58 | + # "2048", |
| 59 | + # ], |
| 60 | + # }, |
| 61 | + # "mistralai/Mistral-7B-Instruct-v0.3": { |
| 62 | + # "model_id": "mistralai/Mistral-7B-Instruct-v0.3", |
| 63 | + # "input": "What is Deep Learning?", |
| 64 | + # "expected_greedy_output": "\n\nDeep learning is a subset of machine learning in artificial intelligence (AI) that has networks capable of learning unsupervised from data that is unstructured", |
| 65 | + # "expected_batch_output": "\n\nDeep learning is a subset of machine learning in artificial intelligence (AI) that has networks capable of learning unsupervised from data that is unstructured", |
| 66 | + # "args": [ |
| 67 | + # "--max-input-tokens", |
| 68 | + # "512", |
| 69 | + # "--max-total-tokens", |
| 70 | + # "1024", |
| 71 | + # "--max-batch-size", |
| 72 | + # "4", |
| 73 | + # "--max-batch-prefill-tokens", |
| 74 | + # "2048", |
| 75 | + # ], |
| 76 | + # }, |
| 77 | + # "bigcode/starcoder2-3b": { |
| 78 | + # "model_id": "bigcode/starcoder2-3b", |
| 79 | + # "input": "What is Deep Learning?", |
| 80 | + # "expected_greedy_output": "\n\nDeep learning is a subset of machine learning that uses artificial neural networks to perform tasks.\n\nNeural networks are a type of machine learning algorithm that", |
| 81 | + # "expected_batch_output": "\n\nDeep learning is a subset of machine learning that uses artificial neural networks to perform tasks.\n\nNeural networks are a type of machine learning algorithm that", |
| 82 | + # "args": [ |
| 83 | + # "--max-input-tokens", |
| 84 | + # "512", |
| 85 | + # "--max-total-tokens", |
| 86 | + # "1024", |
| 87 | + # "--max-batch-size", |
| 88 | + # "4", |
| 89 | + # "--max-batch-prefill-tokens", |
| 90 | + # "2048", |
| 91 | + # ], |
| 92 | + # }, |
| 93 | + # "google/gemma-7b-it": { |
| 94 | + # "model_id": "google/gemma-7b-it", |
| 95 | + # "input": "What is Deep Learning?", |
| 96 | + # "expected_greedy_output": "\n\nDeep learning is a subset of machine learning that uses artificial neural networks to learn from large amounts of data. Neural networks are inspired by the structure and function of", |
| 97 | + # "expected_batch_output": "\n\nDeep learning is a subset of machine learning that uses artificial neural networks to learn from large amounts of data. Neural networks are inspired by the structure and function of", |
| 98 | + # "args": [ |
| 99 | + # "--max-input-tokens", |
| 100 | + # "512", |
| 101 | + # "--max-total-tokens", |
| 102 | + # "1024", |
| 103 | + # "--max-batch-size", |
| 104 | + # "4", |
| 105 | + # "--max-batch-prefill-tokens", |
| 106 | + # "2048", |
| 107 | + # ], |
| 108 | + # }, |
| 109 | + # "Qwen/Qwen2-0.5B-Instruct": { |
| 110 | + # "model_id": "Qwen/Qwen2-0.5B-Instruct", |
| 111 | + # "input": "What is Deep Learning?", |
| 112 | + # "expected_greedy_output": " Deep Learning is a type of machine learning that is based on the principles of artificial neural networks. It is a type of machine learning that is used to train models", |
| 113 | + # "expected_batch_output": " Deep Learning is a type of machine learning that is based on the principles of artificial neural networks. It is a type of machine learning that is used to train models", |
| 114 | + # "args": [ |
| 115 | + # "--max-input-tokens", |
| 116 | + # "512", |
| 117 | + # "--max-total-tokens", |
| 118 | + # "1024", |
| 119 | + # "--max-batch-size", |
| 120 | + # "4", |
| 121 | + # "--max-batch-prefill-tokens", |
| 122 | + # "2048", |
| 123 | + # ], |
| 124 | + # }, |
| 125 | + # "tiiuae/falcon-7b-instruct": { |
| 126 | + # "model_id": "tiiuae/falcon-7b-instruct", |
| 127 | + # "input": "What is Deep Learning?", |
| 128 | + # "expected_greedy_output": "\nDeep learning is a branch of machine learning that uses artificial neural networks to learn and make decisions. It is based on the concept of hierarchical learning, where a", |
| 129 | + # "expected_batch_output": "\nDeep learning is a branch of machine learning that uses artificial neural networks to learn and make decisions. It is based on the concept of hierarchical learning, where a", |
| 130 | + # "args": [ |
| 131 | + # "--max-input-tokens", |
| 132 | + # "512", |
| 133 | + # "--max-total-tokens", |
| 134 | + # "1024", |
| 135 | + # "--max-batch-size", |
| 136 | + # "4", |
| 137 | + # ], |
| 138 | + # }, |
| 139 | + # "microsoft/phi-1_5": { |
| 140 | + # "model_id": "microsoft/phi-1_5", |
| 141 | + # "input": "What is Deep Learning?", |
| 142 | + # "expected_greedy_output": "\n\nDeep Learning is a subfield of Machine Learning that focuses on building neural networks with multiple layers of interconnected nodes. These networks are designed to learn from large", |
| 143 | + # "expected_batch_output": "\n\nDeep Learning is a subfield of Machine Learning that focuses on building neural networks with multiple layers of interconnected nodes. These networks are designed to learn from large", |
| 144 | + # "args": [ |
| 145 | + # "--max-input-tokens", |
| 146 | + # "512", |
| 147 | + # "--max-total-tokens", |
| 148 | + # "1024", |
| 149 | + # "--max-batch-size", |
| 150 | + # "4", |
| 151 | + # ], |
| 152 | + # }, |
| 153 | + # "openai-community/gpt2": { |
| 154 | + # "model_id": "openai-community/gpt2", |
| 155 | + # "input": "What is Deep Learning?", |
| 156 | + # "expected_greedy_output": "\n\nDeep learning is a new field of research that has been around for a long time. It is a new field of research that has been around for a", |
| 157 | + # "expected_batch_output": "\n\nDeep learning is a new field of research that has been around for a long time. It is a new field of research that has been around for a", |
| 158 | + # "args": [ |
| 159 | + # "--max-input-tokens", |
| 160 | + # "512", |
| 161 | + # "--max-total-tokens", |
| 162 | + # "1024", |
| 163 | + # "--max-batch-size", |
| 164 | + # "4", |
| 165 | + # ], |
| 166 | + # }, |
| 167 | + # "facebook/opt-125m": { |
| 168 | + # "model_id": "facebook/opt-125m", |
| 169 | + # "input": "What is Deep Learning?", |
| 170 | + # "expected_greedy_output": "\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout", |
| 171 | + # "expected_batch_output": "\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout the Author\n\nAbout", |
| 172 | + # "args": [ |
| 173 | + # "--max-input-tokens", |
| 174 | + # "512", |
| 175 | + # "--max-total-tokens", |
| 176 | + # "1024", |
| 177 | + # "--max-batch-size", |
| 178 | + # "4", |
| 179 | + # ], |
| 180 | + # }, |
| 181 | + # "EleutherAI/gpt-j-6b": { |
| 182 | + # "model_id": "EleutherAI/gpt-j-6b", |
| 183 | + # "input": "What is Deep Learning?", |
| 184 | + # "expected_greedy_output": "\n\nDeep learning is a subset of machine learning that is based on the idea of neural networks. Neural networks are a type of artificial intelligence that is inspired by", |
| 185 | + # "expected_batch_output": "\n\nDeep learning is a subset of machine learning that is based on the idea of neural networks. Neural networks are a type of artificial intelligence that is inspired by", |
| 186 | + # "args": [ |
| 187 | + # "--max-input-tokens", |
| 188 | + # "512", |
| 189 | + # "--max-total-tokens", |
| 190 | + # "1024", |
| 191 | + # "--max-batch-size", |
| 192 | + # "4", |
| 193 | + # ], |
| 194 | + # }, |
195 | 195 | } |
196 | 196 |
|
197 | 197 | print(f"Testing {len(TEST_CONFIGS)} models") |
|
0 commit comments