Disable quantization in llama config

modal-labs · Feb 6, 2024 · 3aa70b0 · 3aa70b0
1 parent 5b2d0f0
commit 3aa70b0
Showing 1 changed file with 3 additions and 4 deletions.
diff --git a/config/llama-2.yml b/config/llama-2.yml
@@ -3,7 +3,7 @@ model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 is_llama_derived_model: true
 
-load_in_8bit: true
+load_in_8bit: false
 load_in_4bit: false
 strict: false
 
@@ -34,8 +34,8 @@ pad_to_sequence_len: false
 
 adapter: lora
 lora_model_dir:
-lora_r: 32
-lora_alpha: 16
+lora_r: 16
+lora_alpha: 32
 lora_dropout: 0.05
 lora_target_linear: true
 lora_fan_in_fan_out:
@@ -44,7 +44,6 @@ wandb_project:
 wandb_entity:
 wandb_watch:
 wandb_run_id:
-wandb_log_model:
 
 gradient_accumulation_steps: 1
 micro_batch_size: 32