Update big.yml

big.yml

@@ -1,10 +1,10 @@
-base_model: meta-llama/Llama-3.3-70B-Instruct
+base_model: Qwen/Qwen2-72B-Instruct
 base_model_ignore_patterns: "*/*"
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
 
 
-hub_model_id: NewEden/BigPicaro
+hub_model_id: NewEden/BigPicaro-qwen
 hub_strategy: "all_checkpoints"
 push_dataset_to_hub:
 hf_use_auth_token: true
@@ -56,9 +56,7 @@ lora_dropout: 0.05
 lora_target_linear: true
 lora_fan_in_fan_out:
 peft_use_rslora: true
-lora_modules_to_save:
-  - embed_tokens
-  - lm_head
+
 
 wandb_project: tavbussy
 wandb_entity:
@@ -68,10 +66,10 @@ wandb_log_model:
 
 gradient_accumulation_steps: 1
 micro_batch_size: 2
-num_epochs: 2
+num_epochs: 4
 optimizer: paged_adamw_8bit
 lr_scheduler: cosine
-learning_rate: 4.0e-5
+learning_rate: 5e-5
 max_grad_norm: 0.2
 
 train_on_inputs: false
@@ -80,7 +78,7 @@ bf16: auto
 fp16:
 tf32: false
 
-gradient_checkpointing: true
+gradient_checkpointing: unsloth
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank: