From 839ab253d5aa38e69c475f11fc742022bcc8de27 Mon Sep 17 00:00:00 2001 From: Ziqing Yang Date: Thu, 31 Aug 2023 15:20:53 +0800 Subject: [PATCH 1/3] update --with_inst --- scripts/longbench/pred_llama2.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/scripts/longbench/pred_llama2.py b/scripts/longbench/pred_llama2.py index 6dda516..e5456c3 100644 --- a/scripts/longbench/pred_llama2.py +++ b/scripts/longbench/pred_llama2.py @@ -34,7 +34,8 @@ parser.add_argument('--gpus',type=str, default=None) parser.add_argument('--max_length',type=int, default=4096-512) parser.add_argument('--alpha', type=str, default="auto", help="The scaling factor of NTK method, can be a float or 'auto'. ") -parser.add_argument('--with_inst', choices=['true','false','auto'], default = 'auto') +parser.add_argument('--with_inst', choices=['true','false','auto'], default = 'false', + help="Whether use the system prompt and template of Chinese-Alpaca-2 when constructing the instructions.") parser.add_argument('--e', action='store_true', help="Evaluate on LongBench-E") From 1f9b603ce628ce6cebfb0dc78faeae03e1baa133 Mon Sep 17 00:00:00 2001 From: Ziqing Yang Date: Thu, 31 Aug 2023 15:21:17 +0800 Subject: [PATCH 2/3] update --with_inst --- scripts/training/test.py | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+) create mode 100644 scripts/training/test.py diff --git a/scripts/training/test.py b/scripts/training/test.py new file mode 100644 index 0000000..c4a06ac --- /dev/null +++ b/scripts/training/test.py @@ -0,0 +1,18 @@ +import torch +from transformers import LlamaForCausalLM, LlamaTokenizer +from transformers import GenerationConfig +from transformers import BitsAndBytesConfig +from peft import PeftModel +from peft import LoraConfig, TaskType, get_peft_model, PeftModel, get_peft_model_state_dict + +model = LlamaForCausalLM.from_pretrained('/Users/yangziqing/Documents/projects/llama/test/weights/llama_tiny_test',torch_dtype=torch.float16) + +peft_config = LoraConfig( + task_type=TaskType.CAUSAL_LM, + target_modules=["q_proj","v_proj",'embed_tokens','lm_head'], + inference_mode=False, + r=8, lora_alpha=4, + lora_dropout=0,) + #modules_to_save=['embed_tokens','lm_head']) +model = get_peft_model(model, peft_config) +model.print_trainable_parameters() \ No newline at end of file From be05fc293918d4aae16f6c3d4ab4d8e7bc04b8dd Mon Sep 17 00:00:00 2001 From: Ziqing Yang Date: Thu, 31 Aug 2023 15:24:19 +0800 Subject: [PATCH 3/3] remove accidentally added file --- scripts/longbench/pred_llama2.py | 2 +- scripts/training/test.py | 18 ------------------ 2 files changed, 1 insertion(+), 19 deletions(-) delete mode 100644 scripts/training/test.py diff --git a/scripts/longbench/pred_llama2.py b/scripts/longbench/pred_llama2.py index e5456c3..2a4a5b7 100644 --- a/scripts/longbench/pred_llama2.py +++ b/scripts/longbench/pred_llama2.py @@ -34,7 +34,7 @@ parser.add_argument('--gpus',type=str, default=None) parser.add_argument('--max_length',type=int, default=4096-512) parser.add_argument('--alpha', type=str, default="auto", help="The scaling factor of NTK method, can be a float or 'auto'. ") -parser.add_argument('--with_inst', choices=['true','false','auto'], default = 'false', +parser.add_argument('--with_inst', choices=['true','false','auto'], default = 'false', help="Whether use the system prompt and template of Chinese-Alpaca-2 when constructing the instructions.") parser.add_argument('--e', action='store_true', help="Evaluate on LongBench-E") diff --git a/scripts/training/test.py b/scripts/training/test.py deleted file mode 100644 index c4a06ac..0000000 --- a/scripts/training/test.py +++ /dev/null @@ -1,18 +0,0 @@ -import torch -from transformers import LlamaForCausalLM, LlamaTokenizer -from transformers import GenerationConfig -from transformers import BitsAndBytesConfig -from peft import PeftModel -from peft import LoraConfig, TaskType, get_peft_model, PeftModel, get_peft_model_state_dict - -model = LlamaForCausalLM.from_pretrained('/Users/yangziqing/Documents/projects/llama/test/weights/llama_tiny_test',torch_dtype=torch.float16) - -peft_config = LoraConfig( - task_type=TaskType.CAUSAL_LM, - target_modules=["q_proj","v_proj",'embed_tokens','lm_head'], - inference_mode=False, - r=8, lora_alpha=4, - lora_dropout=0,) - #modules_to_save=['embed_tokens','lm_head']) -model = get_peft_model(model, peft_config) -model.print_trainable_parameters() \ No newline at end of file