DEVICE = "cuda:0" DEBUG = False config = { "model": { "codellama": { "base_model_id": "codellama/CodeLlama-7b-hf", "quantitize": "int8", "dataset": "Arithmetic_Simple", "data_collator": "DataCollatorForSeq2Seq", "peft_config": { "lora_config": { "r": 16, "lora_alpha": 16, "target_modules": [ "q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj", ], "lora_dropout": 0.05, "bias": "none", "task_type": "CAUSAL_LM", }, }, "training_args": { "output_dir": "codellama-output", "warmup_steps": 100, "per_device_train_batch_size": 1, "per_device_eval_batch_size": 1, "gradient_accumulation_steps": 4, "max_steps": 10000, "learning_rate": 3e-4, "optim": "adamw_torch", "logging_dir": "codellama-output-logs", "logging_steps": 10, "save_strategy": "steps", "save_steps": 500, "load_best_model_at_end": False, "group_by_length": True, "fp16": True, "evaluation_strategy": "steps", "eval_steps": 1000, # Uncomment this line to set a custom integration to report the results and logs to # With transformers v4, the default value is "all" # With transformers v5, the default value will be "none" # "report_to": "wandb", # Uncomment this line to set a custom run name (default ones like "eternal-brook-20" # will be used if not set) # "run_name": "phi2-code-finetune", # Uncomment the following lines to trigger (Hugging Face built-in) evaluation after # every X steps of training # "evaluation_strategy": "steps", # "eval_steps": 200, # "do_eval": True, }, "tokenizer": { "tokenize_config": { "truncation": True, "max_length": 192, "padding": "max_length", }, "prompt_template": "config/qa_template.txt", }, }, "phi-2": { "base_model_id": "microsoft/phi-2", "quantitize": "fp16", "dataset": "Arithmetic_Simple", "data_collator": "DataCollatorForLanguageModeling", "peft_config":{ "lora_config": { "r": 32, "lora_alpha": 64, "target_modules": [ "q_proj", "k_proj", "v_proj", "dense", "fc1", "fc2", ], "bias": "none", "lora_dropout": 0.05, "task_type": "CAUSAL_LM", }, }, "training_args": { "output_dir": "phi2-output", "warmup_steps": 500, # fp16: ~21.5GiB VRAM; ~40h to finish "per_device_train_batch_size": 1, "per_device_eval_batch_size": 1, "gradient_accumulation_steps": 4, "max_steps": 100000, "learning_rate": 5e-5, "optim": "adamw_torch", "logging_dir": "phi2-output-logs", "logging_steps": 100, "save_strategy": "steps", "save_steps": 500, "evaluation_strategy": "steps", "eval_steps": 500, "fp16": True, }, "tokenizer": { "tokenize_config": { "truncation": True, "max_length": 512, "padding": "max_length", }, "prompt_template": "config/qa_template.txt", }, }, "deepseek": { "base_model_id": "deepseek-ai/deepseek-coder-1.3b-instruct", # Quantization config guidelines for DeepSeek # If running on T4, RTX 20xx, or anything older like V100: quantitize="fp16" # If running on L4, A100, RTX 30xx/40xx, or anything Ampere or later: quantitize="bf16" "quantitize": "fp16", "dataset": "Arithmetic_Hard_Third", "data_collator": "DataCollatorForLanguageModeling", "peft_config":{ "lora": { # trainable params = 30.0 M "r": 32, "lora_alpha": 64, "target_modules": [ "q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj", ], "bias": "none", "lora_dropout": 0.05, "task_type": "CAUSAL_LM", }, "lora_large": { # trainable params = not checked yet "r": 128, "lora_alpha": 256, "target_modules": [ "q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj", ], "bias": "none", "lora_dropout": 0.05, "task_type": "CAUSAL_LM", }, }, "p_tuning_config": { # Doesn't work, PEFT interface issues "num_virtual_tokens": 16, "num_transformer_submodules": 1, "token_dim": 2048, # NOTE(Shih-Lun): should change w/ base LLM "encoder_hidden_size": 2048, "task_type": "CAUSAL_LM", }, "training_args": { "warmup_steps": 500, # bf16: ~21.0GiB VRAM; ~21h to finish "per_device_train_batch_size": 4, "per_device_eval_batch_size": 4, "gradient_accumulation_steps": 1, "max_steps": 35000, "learning_rate": 2e-5, "optim": "adamw_torch", "logging_steps": 100//10, "save_strategy": "steps", "save_total_limit": 10, "save_steps": 500, "evaluation_strategy": "steps", "eval_steps": 500, "weight_decay": 0.01, "report_to": "wandb", "dataloader_num_workers": 4, "load_best_model_at_end": True, # fp16/bf16 config guidelines for DeepSeek # If running full tuning: Don't set fp16/bf16 # If running LoRA: Match the quantization setting (either fp16=True or bf16=True) "fp16": True, }, "tokenizer": { "tokenize_config": { "truncation": True, "max_length": 512, "padding": "max_length", }, "prompt_template": "config/qa_template.txt", }, }, }, "dataset": { "simple_dataset": { "type": "huggingface", # Public datasets on the Hugging Face Hub (only for testing) "dataset_purpose": "downstream", "name": "b-mc2/sql-create-context", "train_split": 0.9, "max_train_size": 100, "filling_field": ["question", "context", "answer"], }, "testdset": { "type": "local", # Local files "dataset_purpose": "downstream", "train_file": "data/Test/TestDataset.json", "val_file": "data/Test/TestDataset.json", "test_file": "data/Test/TestDataset.json", "filling_field": ["prompted_question", "answer"], }, "deepmind_mathematics": { "type": "local", "dataset_purpose": "downstream", "filling_field": ["question", "answer"], "epoch_length":{ "train": 54*4*50, # Module * BatchSize * lesson_num in each module "val": 54*4*1, }, }, "mixture_codegen": { "filling_field": ["Question", "Answer"], "dataset_purpose": "downstream", }, "Arithmetic_Hard_prompt_C12_with_gh": { "filling_field": ["Question", "Answer"], "dataset_purpose": "downstream", }, "MathQA_Python_loader": { "type": "list-like", # List-like objects (we're going to use this for ablations) "dataset_purpose": "downstream", "train": "data/MathQA_Python_processed/mathqa_python_train_clean_final.json", "val": "data/MathQA_Python_processed/mathqa_python_dev_clean_final.json", "test": "data/MathQA_Python_processed/mathqa_python_test_clean_final.json", "filling_field": ["Question", "Answer"], }, "APPS_loader": { "type": "list-like", # List-like objects (we're going to use this for ablations) "dataset_purpose": "downstream", "train": "data/APPS/apps_train.json", "val": "data/APPS/apps_dev.json", "test": "data/APPS/test/apps_test_75.json", "filling_field": ["Question", "Answer"], }, "MBPP_loader": { "type": "list-like", "dataset_purpose": "downstream", "train": "data/MBPP/mbpp_train.json", "val": "data/MBPP/mbpp_dev.json", "test": "data/MBPP/mbpp_test.json", "filling_field": ["Question", "Answer"], }, "Arithmetic_Simple": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops1_Bounds0_100", "Max_Ops1_Bounds0_1000", "Max_Ops2_Bounds0_100", "Max_Ops2_Bounds0_1000", "Max_Ops3_Bounds0_100", # "Max_Ops3_Bounds0_1000", # "Max_Ops4_Bounds0_100", # "Max_Ops4_Bounds0_1000", # "Max_Ops5_Bounds0_100", # "Max_Ops5_Bounds0_1000", ] }, "train": "data/Arithmetic/Curriculum_Simple", "val": "data/Arithmetic/Curriculum_Simple", "test": "data/Arithmetic/Curriculum_Simple", "filling_field": ["Question", "Answer"], }, "Arithmetic_Simple_First_Half": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops1_Bounds0_100", "Max_Ops1_Bounds0_1000", "Max_Ops2_Bounds0_100", "Max_Ops2_Bounds0_1000", "Max_Ops3_Bounds0_100", ] }, "train": "data/Arithmetic/Curriculum_Simple", "val": "data/Arithmetic/Curriculum_Simple", "test": "data/Arithmetic/Curriculum_Simple", "filling_field": ["Question", "Answer"], }, "Arithmetic_Simple_Second_Half": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops3_Bounds0_1000", "Max_Ops4_Bounds0_100", "Max_Ops4_Bounds0_1000", "Max_Ops5_Bounds0_100", "Max_Ops5_Bounds0_1000", ] }, "train": "data/Arithmetic/Curriculum_Simple", "val": "data/Arithmetic/Curriculum_Simple", "test": "data/Arithmetic/Curriculum_Simple", "filling_field": ["Question", "Answer"], }, "Arithmetic_Hard": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops1_Bounds-1000_1000", "Max_Ops1_Bounds-100_100", "Max_Ops1_Bounds0_100", "Max_Ops1_Bounds0_1000", "Max_Ops2_Bounds-1000_1000", "Max_Ops2_Bounds-100_100", "Max_Ops2_Bounds0_100", "Max_Ops2_Bounds0_1000", "Max_Ops3_Bounds-1000_1000", "Max_Ops3_Bounds-100_100", "Max_Ops3_Bounds0_100", "Max_Ops3_Bounds0_1000", "Max_Ops4_Bounds-1000_1000", "Max_Ops4_Bounds-100_100", "Max_Ops4_Bounds0_100", "Max_Ops4_Bounds0_1000", "Max_Ops5_Bounds-1000_1000", "Max_Ops5_Bounds-100_100", "Max_Ops5_Bounds0_100", "Max_Ops5_Bounds0_1000", "Max_Ops6_Bounds-1000_1000", "Max_Ops6_Bounds-100_100", "Max_Ops6_Bounds0_100", "Max_Ops6_Bounds0_1000", "Max_Ops7_Bounds-1000_1000", "Max_Ops7_Bounds-100_100", "Max_Ops7_Bounds0_100", "Max_Ops7_Bounds0_1000", "Max_Ops8_Bounds-1000_1000", "Max_Ops8_Bounds-100_100", "Max_Ops8_Bounds0_100", "Max_Ops8_Bounds0_1000", "Max_Ops9_Bounds-1000_1000", "Max_Ops9_Bounds-100_100", "Max_Ops9_Bounds0_100", "Max_Ops9_Bounds0_1000", "Max_Ops10_Bounds-1000_1000", "Max_Ops10_Bounds-100_100", "Max_Ops10_Bounds0_100", "Max_Ops10_Bounds0_1000", ] }, "train": "data/Arithmetic/Curriculum_Hard", "val": "data/Arithmetic/Curriculum_Hard", "test": "data/Arithmetic/Curriculum_Hard", "filling_field": ["Question", "Answer"], }, "Arithmetic_Hard_First": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops1_Bounds-1000_1000", "Max_Ops1_Bounds-100_100", "Max_Ops1_Bounds0_100", "Max_Ops1_Bounds0_1000", "Max_Ops2_Bounds-1000_1000", "Max_Ops2_Bounds-100_100", "Max_Ops2_Bounds0_100", "Max_Ops2_Bounds0_1000", "Max_Ops3_Bounds-1000_1000", "Max_Ops3_Bounds-100_100", "Max_Ops3_Bounds0_100", "Max_Ops3_Bounds0_1000", "Max_Ops4_Bounds-1000_1000", "Max_Ops4_Bounds-100_100", "Max_Ops4_Bounds0_100", "Max_Ops4_Bounds0_1000", ] }, "train": "data/Arithmetic/Curriculum_Hard", "val": "data/Arithmetic/Curriculum_Hard", "test": "data/Arithmetic/Curriculum_Hard", "filling_field": ["Question", "Answer"], }, "Arithmetic_Hard_Second": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops5_Bounds-1000_1000", "Max_Ops5_Bounds-100_100", "Max_Ops5_Bounds0_100", "Max_Ops5_Bounds0_1000", "Max_Ops6_Bounds-1000_1000", "Max_Ops6_Bounds-100_100", "Max_Ops6_Bounds0_100", "Max_Ops6_Bounds0_1000", "Max_Ops7_Bounds-1000_1000", "Max_Ops7_Bounds-100_100", "Max_Ops7_Bounds0_100", "Max_Ops7_Bounds0_1000", ] }, "train": "data/Arithmetic/Curriculum_Hard", "val": "data/Arithmetic/Curriculum_Hard", "test": "data/Arithmetic/Curriculum_Hard", "filling_field": ["Question", "Answer"], }, "Arithmetic_Hard_Third": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops8_Bounds-1000_1000", "Max_Ops8_Bounds-100_100", "Max_Ops8_Bounds0_100", "Max_Ops8_Bounds0_1000", "Max_Ops9_Bounds-1000_1000", "Max_Ops9_Bounds-100_100", "Max_Ops9_Bounds0_100", "Max_Ops9_Bounds0_1000", "Max_Ops10_Bounds-1000_1000", "Max_Ops10_Bounds-100_100", "Max_Ops10_Bounds0_100", "Max_Ops10_Bounds0_1000", ] }, "train": "data/Arithmetic/Curriculum_Hard", "val": "data/Arithmetic/Curriculum_Hard", "test": "data/Arithmetic/Curriculum_Hard", "filling_field": ["Question", "Answer"], }, "Arithmetic_Hard_prompt_C11": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops1_Bounds-1000_1000", "Max_Ops1_Bounds-100_100", "Max_Ops1_Bounds0_100", "Max_Ops1_Bounds0_1000", "Max_Ops2_Bounds-1000_1000", "Max_Ops2_Bounds-100_100", "Max_Ops2_Bounds0_100", "Max_Ops2_Bounds0_1000", "Max_Ops3_Bounds-1000_1000", "Max_Ops3_Bounds-100_100", "Max_Ops3_Bounds0_100", "Max_Ops3_Bounds0_1000", "Max_Ops4_Bounds-1000_1000", "Max_Ops4_Bounds-100_100", "Max_Ops4_Bounds0_100", "Max_Ops4_Bounds0_1000", "Max_Ops5_Bounds-1000_1000", "Max_Ops5_Bounds-100_100", "Max_Ops5_Bounds0_100", "Max_Ops5_Bounds0_1000", "Max_Ops6_Bounds-1000_1000", "Max_Ops6_Bounds-100_100", "Max_Ops6_Bounds0_100", "Max_Ops6_Bounds0_1000", "Max_Ops7_Bounds-1000_1000", "Max_Ops7_Bounds-100_100", "Max_Ops7_Bounds0_100", "Max_Ops7_Bounds0_1000", "Max_Ops8_Bounds-1000_1000", "Max_Ops8_Bounds-100_100", "Max_Ops8_Bounds0_100", "Max_Ops8_Bounds0_1000", "Max_Ops9_Bounds-1000_1000", "Max_Ops9_Bounds-100_100", "Max_Ops9_Bounds0_100", "Max_Ops9_Bounds0_1000", "Max_Ops10_Bounds-1000_1000", "Max_Ops10_Bounds-100_100", "Max_Ops10_Bounds0_100", "Max_Ops10_Bounds0_1000", ] }, "train": "data/Arithmetic/Curriculum_Hard", "val": "data/Arithmetic/Curriculum_Hard", "test": "data/Arithmetic/Curriculum_Hard", "filling_field": ["Question", "Answer"], }, "Arithmetic_Hard_prompt_C12": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops1_Bounds-1000_1000", "Max_Ops1_Bounds-100_100", "Max_Ops1_Bounds0_100", "Max_Ops1_Bounds0_1000", "Max_Ops2_Bounds-1000_1000", "Max_Ops2_Bounds-100_100", "Max_Ops2_Bounds0_100", "Max_Ops2_Bounds0_1000", "Max_Ops3_Bounds-1000_1000", "Max_Ops3_Bounds-100_100", "Max_Ops3_Bounds0_100", "Max_Ops3_Bounds0_1000", "Max_Ops4_Bounds-1000_1000", "Max_Ops4_Bounds-100_100", "Max_Ops4_Bounds0_100", "Max_Ops4_Bounds0_1000", "Max_Ops5_Bounds-1000_1000", "Max_Ops5_Bounds-100_100", "Max_Ops5_Bounds0_100", "Max_Ops5_Bounds0_1000", "Max_Ops6_Bounds-1000_1000", "Max_Ops6_Bounds-100_100", "Max_Ops6_Bounds0_100", "Max_Ops6_Bounds0_1000", "Max_Ops7_Bounds-1000_1000", "Max_Ops7_Bounds-100_100", "Max_Ops7_Bounds0_100", "Max_Ops7_Bounds0_1000", "Max_Ops8_Bounds-1000_1000", "Max_Ops8_Bounds-100_100", "Max_Ops8_Bounds0_100", "Max_Ops8_Bounds0_1000", "Max_Ops9_Bounds-1000_1000", "Max_Ops9_Bounds-100_100", "Max_Ops9_Bounds0_100", "Max_Ops9_Bounds0_1000", "Max_Ops10_Bounds-1000_1000", "Max_Ops10_Bounds-100_100", "Max_Ops10_Bounds0_100", "Max_Ops10_Bounds0_1000", ] }, "train": "data/Arithmetic/Curriculum_Hard", "val": "data/Arithmetic/Curriculum_Hard", "test": "data/Arithmetic/Curriculum_Hard", "filling_field": ["Question", "Answer"], }, "Arithmetic_Hard_prompt_C12_intermediate": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops1_Bounds-1000_1000", "Max_Ops1_Bounds-100_100", "Max_Ops1_Bounds0_100", "Max_Ops1_Bounds0_1000", "Max_Ops2_Bounds-1000_1000", "Max_Ops2_Bounds-100_100", "Max_Ops2_Bounds0_100", "Max_Ops2_Bounds0_1000", "Max_Ops3_Bounds-1000_1000", "Max_Ops3_Bounds-100_100", "Max_Ops3_Bounds0_100", "Max_Ops3_Bounds0_1000", "Max_Ops4_Bounds-1000_1000", "Max_Ops4_Bounds-100_100", "Max_Ops4_Bounds0_100", "Max_Ops4_Bounds0_1000", "Max_Ops5_Bounds-1000_1000", "Max_Ops5_Bounds-100_100", "Max_Ops5_Bounds0_100", "Max_Ops5_Bounds0_1000", "Max_Ops6_Bounds-1000_1000", "Max_Ops6_Bounds-100_100", "Max_Ops6_Bounds0_100", "Max_Ops6_Bounds0_1000", "Max_Ops7_Bounds-1000_1000", "Max_Ops7_Bounds-100_100", "Max_Ops7_Bounds0_100", "Max_Ops7_Bounds0_1000", "Max_Ops8_Bounds-1000_1000", "Max_Ops8_Bounds-100_100", "Max_Ops8_Bounds0_100", "Max_Ops8_Bounds0_1000", "Max_Ops9_Bounds-1000_1000", "Max_Ops9_Bounds-100_100", "Max_Ops9_Bounds0_100", "Max_Ops9_Bounds0_1000", "Max_Ops10_Bounds-1000_1000", "Max_Ops10_Bounds-100_100", "Max_Ops10_Bounds0_100", "Max_Ops10_Bounds0_1000", ] }, "train": "data/Arithmetic/Curriculum_Hard", "val": "data/Arithmetic/Curriculum_Hard", "test": "data/Arithmetic/Curriculum_Hard", "filling_field": ["Question", "Answer"], }, "Arithmetic_XHard": { "type": "list-like", "dataset_purpose": "downstream", "attributes": { "subjects": [1, 2, 3, 4, 5, 6, 7, 8, 9], "lessons": [ "Max_Ops10_Bounds0_10000.json", "Max_Ops10_Bounds0_1000.json", "Max_Ops10_Bounds-10000_10000.json", "Max_Ops10_Bounds-1000_1000.json", "Max_Ops11_Bounds0_10000.json", "Max_Ops11_Bounds0_1000.json", "Max_Ops11_Bounds-10000_10000.json", "Max_Ops11_Bounds-1000_1000.json", "Max_Ops12_Bounds0_10000.json", "Max_Ops12_Bounds0_1000.json", "Max_Ops12_Bounds-10000_10000.json", "Max_Ops12_Bounds-1000_1000.json", "Max_Ops13_Bounds0_10000.json", "Max_Ops13_Bounds0_1000.json", "Max_Ops13_Bounds-10000_10000.json", "Max_Ops13_Bounds-1000_1000.json", "Max_Ops14_Bounds0_10000.json", "Max_Ops14_Bounds0_1000.json", "Max_Ops14_Bounds-10000_10000.json", "Max_Ops14_Bounds-1000_1000.json", "Max_Ops15_Bounds0_10000.json", "Max_Ops15_Bounds0_1000.json", "Max_Ops15_Bounds-10000_10000.json", "Max_Ops15_Bounds-1000_1000.json", "Max_Ops16_Bounds0_10000.json", "Max_Ops16_Bounds0_1000.json", "Max_Ops16_Bounds-10000_10000.json", "Max_Ops16_Bounds-1000_1000.json", "Max_Ops17_Bounds0_10000.json", "Max_Ops17_Bounds0_1000.json", "Max_Ops17_Bounds-10000_10000.json", "Max_Ops17_Bounds-1000_1000.json", "Max_Ops18_Bounds0_10000.json", "Max_Ops18_Bounds0_1000.json", "Max_Ops18_Bounds-10000_10000.json", "Max_Ops18_Bounds-1000_1000.json", "Max_Ops19_Bounds0_10000.json", "Max_Ops19_Bounds0_1000.json", "Max_Ops19_Bounds-10000_10000.json", "Max_Ops19_Bounds-1000_1000.json", "Max_Ops1_Bounds0_10000.json", "Max_Ops1_Bounds0_1000.json", "Max_Ops1_Bounds-10000_10000.json", "Max_Ops1_Bounds-1000_1000.json", "Max_Ops20_Bounds0_10000.json", "Max_Ops20_Bounds0_1000.json", "Max_Ops20_Bounds-10000_10000.json", "Max_Ops20_Bounds-1000_1000.json", "Max_Ops2_Bounds0_10000.json", "Max_Ops2_Bounds0_1000.json", "Max_Ops2_Bounds-10000_10000.json", "Max_Ops2_Bounds-1000_1000.json", "Max_Ops3_Bounds0_10000.json", "Max_Ops3_Bounds0_1000.json", "Max_Ops3_Bounds-10000_10000.json", "Max_Ops3_Bounds-1000_1000.json", "Max_Ops4_Bounds0_10000.json", "Max_Ops4_Bounds0_1000.json", "Max_Ops4_Bounds-10000_10000.json", "Max_Ops4_Bounds-1000_1000.json", "Max_Ops5_Bounds0_10000.json", "Max_Ops5_Bounds0_1000.json", "Max_Ops5_Bounds-10000_10000.json", "Max_Ops5_Bounds-1000_1000.json", "Max_Ops6_Bounds0_10000.json", "Max_Ops6_Bounds0_1000.json", "Max_Ops6_Bounds-10000_10000.json", "Max_Ops6_Bounds-1000_1000.json", "Max_Ops7_Bounds0_10000.json", "Max_Ops7_Bounds0_1000.json", "Max_Ops7_Bounds-10000_10000.json", "Max_Ops7_Bounds-1000_1000.json", "Max_Ops8_Bounds0_10000.json", "Max_Ops8_Bounds0_1000.json", "Max_Ops8_Bounds-10000_10000.json", "Max_Ops8_Bounds-1000_1000.json", "Max_Ops9_Bounds0_10000.json", "Max_Ops9_Bounds0_1000.json", "Max_Ops9_Bounds-10000_10000.json", "Max_Ops9_Bounds-1000_1000.json", ] }, "train": "data/Arithmetic/Curriculum_XHard", "val": "data/Arithmetic/Curriculum_XHard", "test": "data/Arithmetic/Curriculum_XHard", "filling_field": ["Question", "Answer"], }, "GSM8K": { "type": "local", "dataset_purpose": "downstream", "train_file": "data/GSM8K/GSM8K_train.json", "val_file": "data/GSM8K/GSM8K_test.json", "test_file": "data/GSM8K/GSM8K_dev.json", "filling_field": ["Body", "Question", "Answer"], }, "APPS": { "type": "local", "dataset_purpose": "downstream", "train_file": "data/APPS/apps_train.json", "val_file": "data/APPS/apps_test.json", "test_file": "data/APPS/apps_dev.json", "filling_field": ["Body", "Question", "Answer"], }, "ghcode_python": { "type": "huggingface", "dataset_purpose": "pretrain", "name": "slseanwu/ghcode_python_split_700k", "max_eval_size": 1000, "max_train_size": 160000, "filling_field": ["code"], }, }, } if DEBUG: config.epochs = 100 config.save_steps = 10 config.train_dataset = "local-test-train" config.val_dataset = "local-test-dev" config.test_dataset = "test-clean"