Delta-Vector commited on
Commit
171b700
·
verified ·
1 Parent(s): 053053e

Update big.yml

Browse files
Files changed (1) hide show
  1. big.yml +6 -8
big.yml CHANGED
@@ -1,10 +1,10 @@
1
- base_model: meta-llama/Llama-3.3-70B-Instruct
2
  base_model_ignore_patterns: "*/*"
3
  model_type: AutoModelForCausalLM
4
  tokenizer_type: AutoTokenizer
5
 
6
 
7
- hub_model_id: NewEden/BigPicaro
8
  hub_strategy: "all_checkpoints"
9
  push_dataset_to_hub:
10
  hf_use_auth_token: true
@@ -56,9 +56,7 @@ lora_dropout: 0.05
56
  lora_target_linear: true
57
  lora_fan_in_fan_out:
58
  peft_use_rslora: true
59
- lora_modules_to_save:
60
- - embed_tokens
61
- - lm_head
62
 
63
  wandb_project: tavbussy
64
  wandb_entity:
@@ -68,10 +66,10 @@ wandb_log_model:
68
 
69
  gradient_accumulation_steps: 1
70
  micro_batch_size: 2
71
- num_epochs: 2
72
  optimizer: paged_adamw_8bit
73
  lr_scheduler: cosine
74
- learning_rate: 4.0e-5
75
  max_grad_norm: 0.2
76
 
77
  train_on_inputs: false
@@ -80,7 +78,7 @@ bf16: auto
80
  fp16:
81
  tf32: false
82
 
83
- gradient_checkpointing: true
84
  early_stopping_patience:
85
  resume_from_checkpoint:
86
  local_rank:
 
1
+ base_model: Qwen/Qwen2-72B-Instruct
2
  base_model_ignore_patterns: "*/*"
3
  model_type: AutoModelForCausalLM
4
  tokenizer_type: AutoTokenizer
5
 
6
 
7
+ hub_model_id: NewEden/BigPicaro-qwen
8
  hub_strategy: "all_checkpoints"
9
  push_dataset_to_hub:
10
  hf_use_auth_token: true
 
56
  lora_target_linear: true
57
  lora_fan_in_fan_out:
58
  peft_use_rslora: true
59
+
 
 
60
 
61
  wandb_project: tavbussy
62
  wandb_entity:
 
66
 
67
  gradient_accumulation_steps: 1
68
  micro_batch_size: 2
69
+ num_epochs: 4
70
  optimizer: paged_adamw_8bit
71
  lr_scheduler: cosine
72
+ learning_rate: 5e-5
73
  max_grad_norm: 0.2
74
 
75
  train_on_inputs: false
 
78
  fp16:
79
  tf32: false
80
 
81
+ gradient_checkpointing: unsloth
82
  early_stopping_patience:
83
  resume_from_checkpoint:
84
  local_rank: