Upload training_config.json with huggingface_hub
Browse files- training_config.json +93 -0
training_config.json
ADDED
|
@@ -0,0 +1,93 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"project_name": "azma-hermes-pro-llama-3-8b-030524",
|
| 3 |
+
"new_model_id": "azma-hermes-pro-llama-3-8b-030524",
|
| 4 |
+
"base_model_id": "NousResearch/Hermes-2-Pro-Llama-3-8B",
|
| 5 |
+
"adapter_id": "azma-hermes-pro-llama-3-8b-030524-adapter",
|
| 6 |
+
"chat_template": "chatml",
|
| 7 |
+
"lora_config": {
|
| 8 |
+
"lora_rank": 16,
|
| 9 |
+
"lora_alpha": 32,
|
| 10 |
+
"lora_dropout": 0.1,
|
| 11 |
+
"target_modules": [
|
| 12 |
+
"k_proj",
|
| 13 |
+
"v_proj",
|
| 14 |
+
"o_proj",
|
| 15 |
+
"q_proj",
|
| 16 |
+
"up_proj",
|
| 17 |
+
"gate_proj",
|
| 18 |
+
"down_proj"
|
| 19 |
+
],
|
| 20 |
+
"modules_to_save": [
|
| 21 |
+
"lm_head"
|
| 22 |
+
],
|
| 23 |
+
"task_type": "CAUSAL_LM"
|
| 24 |
+
},
|
| 25 |
+
"load_config": {
|
| 26 |
+
"use_flash_attention": true,
|
| 27 |
+
"load_in_4bit": false,
|
| 28 |
+
"cache_dir": "./",
|
| 29 |
+
"torch_dtype": "auto",
|
| 30 |
+
"device_map": "auto",
|
| 31 |
+
"pad_to_multiples": false
|
| 32 |
+
},
|
| 33 |
+
"dataset_config": {
|
| 34 |
+
"dataset_id": [
|
| 35 |
+
"Azma-AI/azma-mermaid-dataset-single-turn-chatml",
|
| 36 |
+
"Azma-AI/azma-dataset-v2-mermaid-without-thoughts-final-chatml-8192-seq-len"
|
| 37 |
+
],
|
| 38 |
+
"max_seq_length": 8192,
|
| 39 |
+
"shuffle": true,
|
| 40 |
+
"data_collator": "default",
|
| 41 |
+
"template_version": 1.0,
|
| 42 |
+
"dataset_text_field": "text",
|
| 43 |
+
"add_custom_tokens": {
|
| 44 |
+
"bos_token": "<|begin_of_text|>",
|
| 45 |
+
"eos_token": "<|im_end|>",
|
| 46 |
+
"pad_token": "<|end_of_text|>"
|
| 47 |
+
}
|
| 48 |
+
},
|
| 49 |
+
"dpo_args": {
|
| 50 |
+
"beta": 0.6,
|
| 51 |
+
"loss_types": "sigmoid"
|
| 52 |
+
},
|
| 53 |
+
"training_args": {
|
| 54 |
+
"seed": 42,
|
| 55 |
+
"max_steps": -1,
|
| 56 |
+
"weight_decay": 0.01,
|
| 57 |
+
"num_train_epochs": 1,
|
| 58 |
+
"learning_rate": 1e-05,
|
| 59 |
+
"output_dir": "./results",
|
| 60 |
+
"optim": "paged_adamw_32bit",
|
| 61 |
+
"gradient_accumulation_steps": 2,
|
| 62 |
+
"per_device_train_batch_size": 8,
|
| 63 |
+
"per_device_eval_batch_size": 2,
|
| 64 |
+
"group_by_length": false,
|
| 65 |
+
"resume_from_checkpoint": true,
|
| 66 |
+
"gradient_checkpointing": true,
|
| 67 |
+
"gradient_checkpointing_kwargs": null,
|
| 68 |
+
"do_eval": false,
|
| 69 |
+
"eval_steps": 100,
|
| 70 |
+
"evaluation_strategy": "steps",
|
| 71 |
+
"save_steps": 100,
|
| 72 |
+
"save_total_limit": 2,
|
| 73 |
+
"save_strategy": "steps",
|
| 74 |
+
"logging_steps": 5,
|
| 75 |
+
"log_level": "info",
|
| 76 |
+
"logging_strategy": "steps",
|
| 77 |
+
"push_to_hub": true,
|
| 78 |
+
"hub_model_id": "Azma-AI/azma-hermes-pro-llama-3-8b-030524-adapter",
|
| 79 |
+
"hub_private_repo": true,
|
| 80 |
+
"hub_strategy": "checkpoint",
|
| 81 |
+
"report_to": "wandb",
|
| 82 |
+
"warmup_steps": 5,
|
| 83 |
+
"neftune_noise_alpha": 5,
|
| 84 |
+
"lr_scheduler_type": "cosine",
|
| 85 |
+
"auto_find_batch_size": true,
|
| 86 |
+
"load_best_model_at_end": true,
|
| 87 |
+
"deepspeed": null,
|
| 88 |
+
"bf16": true,
|
| 89 |
+
"fp16": false
|
| 90 |
+
},
|
| 91 |
+
"merge_final_model": true,
|
| 92 |
+
"push_to_organization": true
|
| 93 |
+
}
|