Datasets:

Modalities:
Text
Formats:
parquet
Size:
< 1K
ArXiv:
Tags:
code
Libraries:
Datasets
pandas
License:
deniskokosss commited on
Commit
9d9f5da
1 Parent(s): dfe2dea

Added reproduction script

Browse files
Files changed (1) hide show
  1. README.md +94 -7
README.md CHANGED
@@ -88,11 +88,98 @@ accelerate launch main.py \
88
  | CodeLlama-13b-Python-hf | 0.4380 | 0.5796 | 0.4301 | 0.6226 |
89
 
90
  <details>
91
- <summary> Generation parameters to reproduce results in the table </summary>
92
- 'do_sample': True,
93
- 'temperature': 0.2,
94
- 'top_k': 0,
95
- 'top_p': 0.95,
96
- 'n_samples': 20,
97
- 'max_length_generation': 512,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
98
  </details>
 
88
  | CodeLlama-13b-Python-hf | 0.4380 | 0.5796 | 0.4301 | 0.6226 |
89
 
90
  <details>
91
+ <summary> Generation scripts in the table </summary>
92
+ ```console
93
+ #!/bin/bash
94
+ # use with https://github.com/NLP-Core-Team/bigcode-evaluation-harness
95
+
96
+ # RU
97
+ mkdir -p ./outs/humaneval_ru
98
+ mkdir -p ./results/humaneval_ru
99
+ MODELS_PATH="bigcode"
100
+ echo $MODELS_PATH
101
+ declare -A bs=( ["starcoderbase-1b"]=16 ["starcoderbase-3b"]=8 ["starcoderbase-7b"]=4 ["starcoderbase"]=1 ["starcoder"]=1)
102
+ for model_name in starcoder # starcoderbase-1b starcoderbase-3b starcoderbase-7b starcoderbase codellama_2_13b
103
+ do
104
+ echo $MODELS_PATH/$model_name
105
+ conda run -p /home/jovyan/kokosinskiy/envs/codegen310 --no-capture-output accelerate launch --mixed_precision="fp16" main.py \
106
+ --model $MODELS_PATH/$model_name \
107
+ --max_length_generation 512 \
108
+ --tasks humaneval_ru \
109
+ --use_auth_token \
110
+ --temperature 0.2 \
111
+ --n_samples 20 \
112
+ --precision fp16 \
113
+ --batch_size ${bs[$model_name]} \
114
+ --allow_code_execution \
115
+ --save_generations_path ./outs/humaneval_ru/$model_name.json \
116
+ --metric_output_path ./results/humaneval_ru/$model_name.metrics
117
+ done
118
+
119
+ MODELS_PATH="codellama"
120
+ echo $MODELS_PATH
121
+ declare -A bs=( ["CodeLlama-7b-Python-hf"]=8 ["CodeLlama-7b-hf"]=16 ["CodeLlama-13b-Python-hf"]=4 ["CodeLlama-13b-hf"]=4 )
122
+ for model_name in CodeLlama-7b-hf CodeLlama-7b-Python-hf CodeLlama-13b-hf CodeLlama-13b-Python-hf # codellama_2_13b
123
+ do
124
+ echo $MODELS_PATH/$model_name
125
+ conda run -p /home/jovyan/kokosinskiy/envs/codegen310 --no-capture-output accelerate launch --mixed_precision="fp16" main.py \
126
+ --model $MODELS_PATH/$model_name \
127
+ --max_length_generation 512 \
128
+ --tasks humaneval_ru \
129
+ --use_auth_token \
130
+ --temperature 0.2 \
131
+ --n_samples 20 \
132
+ --precision fp16 \
133
+ --batch_size ${bs[$model_name]} \
134
+ --allow_code_execution \
135
+ --save_generations_path ./outs/humaneval_ru/$model_name.json \
136
+ --metric_output_path ./results/humaneval_ru/$model_name.metrics
137
+ done
138
+
139
+ # EN
140
+
141
+ mkdir -p ./outs/humaneval
142
+ mkdir -p ./results/humaneval
143
+ MODELS_PATH="bigcode"
144
+ echo $MODELS_PATH
145
+ declare -A bs=( ["starcoderbase-1b"]=16 ["starcoderbase-3b"]=8 ["starcoderbase-7b"]=4 ["starcoderbase"]=1 ["starcoder"]=1)
146
+ for model_name in starcoder starcoderbase-1b starcoderbase-3b starcoderbase-7b starcoderbase
147
+ do
148
+ echo $MODELS_PATH/$model_name
149
+ conda run -p /home/jovyan/kokosinskiy/envs/codegen310 --no-capture-output accelerate launch --mixed_precision="fp16" main.py \
150
+ --model $MODELS_PATH/$model_name \
151
+ --max_length_generation 512 \
152
+ --tasks humaneval \
153
+ --use_auth_token \
154
+ --temperature 0.2 \
155
+ --n_samples 20 \
156
+ --precision fp16 \
157
+ --batch_size ${bs[$model_name]} \
158
+ --allow_code_execution \
159
+ --save_generations_path ./outs/humaneval/$model_name.json \
160
+ --metric_output_path ./results/humaneval/$model_name.metrics
161
+ done
162
+
163
+ MODELS_PATH="codellama"
164
+ echo $MODELS_PATH
165
+ declare -A bs=( ["CodeLlama-7b-Python-hf"]=8 ["CodeLlama-7b-hf"]=16 ["CodeLlama-13b-Python-hf"]=4 ["CodeLlama-13b-hf"]=4 )
166
+ for model_name in CodeLlama-7b-hf CodeLlama-7b-Python-hf CodeLlama-13b-hf CodeLlama-13b-Python-hf
167
+ do
168
+ echo $MODELS_PATH/$model_name
169
+ conda run -p /home/jovyan/kokosinskiy/envs/codegen310 --no-capture-output accelerate launch --mixed_precision="fp16" main.py \
170
+ --model $MODELS_PATH/$model_name \
171
+ --max_length_generation 512 \
172
+ --tasks humaneval \
173
+ --use_auth_token \
174
+ --temperature 0.2 \
175
+ --n_samples 20 \
176
+ --precision fp16 \
177
+ --batch_size ${bs[$model_name]} \
178
+ --allow_code_execution \
179
+ --save_generations_path ./outs/humaneval/$model_name.json \
180
+ --metric_output_path ./results/humaneval/$model_name.metrics
181
+ done
182
+
183
+
184
+ ```
185
  </details>