Qwen2.5 fine-tuned

scripts/eval-mgtv-qwen2.5.sh

@@ -0,0 +1,29 @@
+#!/bin/sh
+
+BASEDIR=$(dirname "$0")
+cd $BASEDIR/..
+echo Current Directory:
+pwd
+
+BASEDIR=`pwd`
+
+nvidia-smi
+uname -a
+cat /etc/os-release
+lscpu
+grep MemTotal /proc/meminfo
+
+export LOGICAL_REASONING_DATA_PATH=datasets/mgtv
+export RESIZE_TOKEN_EMBEDDINGS=true
+export USING_LLAMA_FACTORY=true
+export USING_P1_PROMPT_TEMPLATE=false
+#export LOAD_IN_4BIT=true
+
+export MODEL_PREFIX=Qwen2.5-3B-Instruct
+export MODEL_NAME=qwen/$MODEL_PREFIX
+
+export LOGICAL_REASONING_RESULTS_PATH=data/${MODEL_PREFIX}_results.csv
+export ADAPTER_PATH_BASE=llama-factory/saves/$MODEL_PREFIX
+
+echo "Eval $MODEL_NAME with $ADAPTER_PATH_BASE"
+python llm_toolkit/eval_logical_reasoning_all_epochs.py

scripts/eval-mgtv-shots-qwen2.5.sh

@@ -0,0 +1,20 @@
+#!/bin/sh
+
+BASEDIR=$(dirname "$0")
+cd $BASEDIR/..
+echo Current Directory:
+pwd
+
+BASEDIR=`pwd`
+
+nvidia-smi
+uname -a
+cat /etc/os-release
+lscpu
+grep MemTotal /proc/meminfo
+
+export START_NUM_SHOTS=5
+
+$BASEDIR/scripts/eval-shots.sh Qwen Qwen2.5-3B-Instruct
+
+#$BASEDIR/scripts/eval-shots.sh Qwen Qwen2.5-7B-Instruct

scripts/tune-mgtv-bf16.sh

@@ -25,8 +25,10 @@ export LOGICAL_REASONING_RESULTS_PATH=results/mgtv-results_bf16.csv
 
 # $BASEDIR/scripts/tune-lf_v2.sh  internlm internlm2_5-7b-chat intern2
 
-$BASEDIR/scripts/tune-lf_v2.sh  Qwen Qwen2.5-3B-Instruct qwen
+$BASEDIR/scripts/tune-lf_v2.sh  Qwen Qwen2.5-7B-Instruct qwen
 
-$BASEDIR/scripts/tune-lf_v2.sh  Qwen Qwen2.5-1.5B-Instruct qwen
+#$BASEDIR/scripts/tune-lf_v2.sh  Qwen Qwen2.5-3B-Instruct qwen
 
-$BASEDIR/scripts/tune-lf_v2.sh  Qwen Qwen2.5-0.5B-Instruct qwen
+#$BASEDIR/scripts/tune-lf_v2.sh  Qwen Qwen2.5-1.5B-Instruct qwen
+
+#$BASEDIR/scripts/tune-lf_v2.sh  Qwen Qwen2.5-0.5B-Instruct qwen