llama3 8b base model results

llama-factory/config/llama3-8b_lora_sft_bf16-p1.yaml

@@ -1,5 +1,6 @@
 ### model
-model_name_or_path: FlagAlpha/Llama3-Chinese-8B-Instruct
+model_name_or_path: shenzhi-wang/Llama3-8B-Chinese-Chat
+#model_name_or_path: FlagAlpha/Llama3-Chinese-8B-Instruct
 
 ### method
 stage: sft

llama-factory/config/llama3-8b_lora_sft_bf16-p2.yaml

@@ -1,5 +1,6 @@
 ### model
-model_name_or_path: FlagAlpha/Llama3-Chinese-8B-Instruct
+model_name_or_path: shenzhi-wang/Llama3-8B-Chinese-Chat
+#model_name_or_path: FlagAlpha/Llama3-Chinese-8B-Instruct
 
 ### method
 stage: sft

scripts/eval-mgtv.sh

@@ -1 +1 @@
-eval-mgtv-llama3.sh
+eval-mgtv-llama3_8b.sh

scripts/tune-mgtv-llama3_8b.sh

@@ -20,24 +20,20 @@ grep MemTotal /proc/meminfo
 
 export LOGICAL_REASONING_DATA_PATH=datasets/mgtv
 
-export MODEL_NAME=FlagAlpha/Llama3-Chinese-8B-Instruct
+#export MODEL_NAME=FlagAlpha/Llama3-Chinese-8B-Instruct
+export MODEL_NAME=shenzhi-wang/Llama3-8B-Chinese-Chat
+
 export MODEL_PREFIX=llama3-8b_lora_sft_bf16
 
 export CONFIG_FILE=config/$MODEL_PREFIX-p1.yaml
-echo "Tuning with $CONFIG_FILE"
-#$BASEDIR/scripts/tune-lf.sh $CONFIG_FILE
-
-export LOGICAL_REASONING_RESULTS_PATH=results/$MODEL_PREFIX-p1.csv
-export ADAPTER_PATH_BASE=llama-factory/saves/qwen2-72b/lora/sft_4bit_p1_full
-echo "Eval $MODEL_NAME with $ADAPTER_PATH_BASE"
-#python llm_toolkit/eval_logical_reasoning_all_epochs.py
+echo "Tuning $MODEL_NAME with $CONFIG_FILE"
+$BASEDIR/scripts/tune-lf.sh $CONFIG_FILE
 
 
 export CONFIG_FILE=config/$MODEL_PREFIX-p2.yaml
-echo "Tuning with $CONFIG_FILE"
+echo "Tuning $MODEL_NAME with $CONFIG_FILE"
 $BASEDIR/scripts/tune-lf.sh $CONFIG_FILE
 
-export LOGICAL_REASONING_RESULTS_PATH=results/$MODEL_PREFIX-p2.csv
-export ADAPTER_PATH_BASE=llama-factory/saves/qwen2-72b/lora/sft_4bit_p2_full
-echo "Eval $MODEL_NAME with $ADAPTER_PATH_BASE"
-#python llm_toolkit/eval_logical_reasoning_all_epochs.py
+
+$BASEDIR/scripts/eval-mgtv-llama3_8b.sh
+