mlfoundations-dev/train_fasttext_classifier_seed_code_best_codegolf
05
1operators:2- id: load_hf_python3 config:4 type: hf_source5 dataset: bigcode/commitpackft6 split: train7 subset: python8 trust_remote_code: True9- id: load_hf_c++10 config:11 type: hf_source12 dataset: bigcode/commitpackft13 split: train14 subset: c++15 trust_remote_code: True16- id: load_hf_java17 config:18 type: hf_source19 dataset: bigcode/commitpackft20 split: train21 subset: java22 trust_remote_code: True23- id: load_hf_c24 config:25 trust_remote_code: True26 type: hf_source27 dataset: bigcode/commitpackft28 split: train29 subset: c30- id: load_hf_c#31 config:32 trust_remote_code: True33 type: hf_source34 dataset: bigcode/commitpackft35 split: train36 subset: c#37- id: load_hf_css38 config:39 trust_remote_code: True40 type: hf_source41 dataset: bigcode/commitpackft42 split: train43 subset: css44- id: load_hf_javascript45 config:46 trust_remote_code: True47 type: hf_source48 dataset: bigcode/commitpackft49 split: train50 subset: javascript51- id: load_hf_ruby52 config:53 trust_remote_code: True54 type: hf_source55 dataset: bigcode/commitpackft56 split: train57 subset: ruby58- id: load_hf_shell59 config:60 trust_remote_code: True61 type: hf_source62 dataset: bigcode/commitpackft63 split: train64 subset: shell65- id: mix_operator66 config:67 type: mix68 input_ids:69 - load_hf_python70 - load_hf_shell71 - load_hf_c++72 - load_hf_java73 - load_hf_c74 - load_hf_c#75 - load_hf_css76 - load_hf_javascript77 - load_hf_ruby78- id: sample_dataset79 config:80 type: function81 function: data_strategies.commons.uniform_sample_limited82 function_config:83 num_samples: 50000 # 1.25% of 10000 target from stage 184 input_ids:85 - mix_operator86- id: generate_instructions87 config:88 type: completions89 map: chat90 map_config:91 system_message: "You are a helpful assistant."92 user_message: |93 You are to generate a question or task for a language model based on the following instruction and code pairs. 94 95 Instruction: {{message}}96 Code: {{old_contents}}97 98 Include only the new question and task. Do not include anything like "Here is the instruction". Include99 the code in your question and make the task sound like what a human would ask a language model. 100 output_column: instruction_seed101 model: gpt-4o-mini102 temperature: 1.0103 batch: True104 input_ids:105 - sample_dataset106- id: add_constant_column107 config:108 type: function109 function: data_strategies.commons.add_constant_columns110 function_config:111 response_seed: ""112 input_ids:113 - generate_instructions114- id: rename_task_column115 config:116 type: function117 function: data_strategies.commons.keep_only_columns118 function_config:119 columns_to_keep:120 - response_seed121 - instruction_seed122 input_ids:123 - add_constant_column124- id: annotate125 config:126 type: completions127 map: chat128 map_config:129 system_message: "You are a helpful assistant."130 user_message_column: instruction_seed131 output_column: response132 model: gpt-4o-mini133 temperature: 1.0134 batch: True135 input_ids:136 - rename_task_column137- id: convert_to_sharegpt138 config:139 type: function140 function: data_strategies.commons.convert_instruction_response_to_sharegpt141 function_config:142 input_instruction_column: instruction_seed143 input_response_column: response144 output_sharegpt_column: conversations145 input_ids:146 - annotate