An early checkpoint ~ Contextually obedient in answering, training curriculum is currently at training for contextual-refusals when the information provided is not sufficent to answer
a34c2f6 verified | { | |
| "lora_name": "Contextual-Codestral", | |
| "always_override": true, | |
| "q_proj_en": true, | |
| "v_proj_en": true, | |
| "k_proj_en": true, | |
| "o_proj_en": true, | |
| "gate_proj_en": true, | |
| "down_proj_en": true, | |
| "up_proj_en": true, | |
| "save_steps": 0, | |
| "micro_batch_size": 42, | |
| "batch_size": 24, | |
| "epochs": 1.42, | |
| "learning_rate": "6e-7", | |
| "lr_scheduler_type": "polynomial", | |
| "lora_rank": 8, | |
| "lora_alpha": 16, | |
| "lora_dropout": 0.05, | |
| "cutoff_len": 8192, | |
| "dataset": "0_Contextual", | |
| "eval_dataset": "1_I_Cannot_Answer_That_For_You_From_Provided_Context", | |
| "format": "Mermaid-format", | |
| "eval_steps": 100, | |
| "raw_text_file": "None", | |
| "overlap_len": 128, | |
| "newline_favor_len": 128, | |
| "higher_rank_limit": false, | |
| "warmup_steps": 1, | |
| "optimizer": "adagrad", | |
| "hard_cut_string": "\\n\\n\\n", | |
| "train_only_after": "", | |
| "stop_at_loss": 0.2, | |
| "add_eos_token": true, | |
| "min_chars": 0, | |
| "report_to": "None" | |
| } |