Описание модели:

Tikhomirov, M.M. and Chernyshev, D.I., 2024. Improving Large Language Model Russian Adaptation with Preliminary Vocabulary Optimization. Lobachevskii Journal of Mathematics (will be soon) ruadapt Solar-10.7 tuned on train sets using https://github.com/IlyaGusev/rulm&

Описание параметров:

Tested using https://github.com/IlyaGusev/rulm repo Training config: { "trainer": { "evaluation_strategy": "steps", "per_device_train_batch_size": 1, "per_device_eval_batch_size": 1, "gradient_accumulation_steps": 128, "eval_steps": 100, "save_steps": 100, "logging_steps": 5, "learning_rate": 0.00025, "num_train_epochs": 3, "lr_scheduler_type": "cosine", "warmup_steps": 30, "fp16": true, "bf16": false, "torch_compile": false, "optim": "adamw_torch" }, "lora": { "r": 16, "lora_alpha": 16, "lora_dropout": 0.05, "bias": "none", "target_modules": ["q_proj", "v_proj", "k_proj", "o_proj"], "task_type": "CAUSAL_LM" }, "load_in_8bit": false, "only_target_loss": true, "mode": "chat", "templates_path": "internal_prompts/saiga_v2.json", "model_name": "/data/models/gpt/solar/ruadapt_solar_10.7_darulm_unigram_proj_init_part2_v3_alpha_scale_2", "model_type": "causal", "max_tokens_count": 1024 }