created scripts/tune-mgtv.sh

llama-factory/config/internlm2_5_7b_lora_sft_bf16_p1_full.yaml

@@ -0,0 +1,47 @@
+### model
+model_name_or_path: internlm/internlm2_5-7b-chat-1m
+
+### method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: all
+# quantization_bit: 4                     # use 4-bit QLoRA
+loraplus_lr_ratio: 16.0                 # use LoRA+ with lambda=16.0
+# use_unsloth: true                       # use UnslothAI's LoRA optimization for 2x faster training
+upcast_layernorm: true
+
+### dataset
+dataset: alpaca_mgtv_p1
+template: chatml
+cutoff_len: 4096
+max_samples: 25000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+### output
+output_dir: saves/internlm2_5_7b/lora/sft_bf16_p1_full
+logging_steps: 100
+save_steps: 562
+plot_loss: true
+overwrite_output_dir: true
+# resume_from_checkpoint: true
+
+### train
+per_device_train_batch_size: 16
+gradient_accumulation_steps: 8
+learning_rate: 1.0e-4
+num_train_epochs: 6.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+
+### eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+eval_strategy: steps
+eval_steps: 562
+
+report_to: wandb
+run_name: internlm2_5_7b # optional

llama-factory/config/internlm2_5_7b_lora_sft_bf16_p2_full.yaml

@@ -0,0 +1,47 @@
+### model
+model_name_or_path: internlm/internlm2_5-7b-chat-1m
+
+### method
+stage: sft
+do_train: true
+finetuning_type: lora
+lora_target: all
+# quantization_bit: 4                     # use 4-bit QLoRA
+loraplus_lr_ratio: 16.0                 # use LoRA+ with lambda=16.0
+# use_unsloth: true                       # use UnslothAI's LoRA optimization for 2x faster training
+upcast_layernorm: true
+
+### dataset
+dataset: alpaca_mgtv_p2
+template: chatml
+cutoff_len: 4096
+max_samples: 25000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+### output
+output_dir: saves/internlm2_5_7b/lora/sft_bf16_p2_full
+logging_steps: 100
+save_steps: 562
+plot_loss: true
+overwrite_output_dir: true
+# resume_from_checkpoint: true
+
+### train
+per_device_train_batch_size: 16
+gradient_accumulation_steps: 8
+learning_rate: 1.0e-4
+num_train_epochs: 6.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+
+### eval
+val_size: 0.1
+per_device_eval_batch_size: 1
+eval_strategy: steps
+eval_steps: 562
+
+report_to: wandb
+run_name: internlm2_5_7b # optional

scripts/tune-mgtv.sh

@@ -0,0 +1,18 @@
+#!/bin/sh
+
+BASEDIR=$(dirname "$0")
+cd $BASEDIR/..
+echo Current Directory:
+pwd
+
+export LOGICAL_REASONING_DATA_PATH=datasets/mgtv
+
+export LOGICAL_REASONING_RESULTS_PATH=results/mgtv-results_p1_full.csv
+export CONFIG_FILE=configs/internlm2_5_7b_lora_sft_bf16_p1_full.yaml
+echo "Tuning with $CONFIG_FILE"
+./scripts/tune-lf.sh $CONFIG_FILE
+
+export LOGICAL_REASONING_RESULTS_PATH=results/mgtv-results_p2_full.csv
+export CONFIG_FILE=configs/internlm2_5_7b_lora_sft_bf16_p2_full.yaml
+cho "Tuning with $CONFIG_FILE"
+./scripts/tune-lf.sh $CONFIG_FILE