best eval LF

llm_toolkit/eval_logical_reasoning.py

@@ -24,6 +24,7 @@ data_path = os.getenv("LOGICAL_REASONING_DATA_PATH")
 results_path = os.getenv("LOGICAL_REASONING_RESULTS_PATH")
 use_english_datasets = os.getenv("USE_ENGLISH_DATASETS") == "true"
 using_p1 = os.getenv("USING_P1_PROMPT_TEMPLATE") == "true"
+using_llama_factory = os.getenv("USING_LLAMA_FACTORY") == "true"
 max_new_tokens = int(os.getenv("MAX_NEW_TOKENS", 16))
 repetition_penalty = float(os.getenv("REPETITION_PENALTY", 1.0))
 
@@ -43,6 +44,7 @@ model, tokenizer = load_model(
     model_name,
     load_in_4bit=load_in_4bit,
     adapter_name_or_path=adapter_name_or_path,
+    using_llama_factory=using_llama_factory,
     dtype=dtype,
 )
 

scripts/eval-mgtv-best.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
+
+#pip install -r requirements.txt
+#cd ../LLaMA-Factory && pip install -e .[torch,bitsandbytes]
+
+pip install transformers==4.41.2
+
+export USING_LLAMA_FACTORY=true
+
+export MODEL_NAME=internlm/internlm2_5-7b-chat-1m
+export ADAPTER_NAME_OR_PATH=inflaton-ai/InternLM_2_5-7b_LoRA-Adapter
+export LOGICAL_REASONING_DATA_PATH=datasets/mgtv
+export LOGICAL_REASONING_RESULTS_PATH=results/mgtv-results_internlm_best.csv
+
+echo "Eval $MODEL_NAME with $ADAPTER_NAME_OR_PATH"
+python llm_toolkit/eval_logical_reasoning.py

scripts/eval-mgtv.sh

@@ -1 +1 @@
-eval-mgtv-glm-4-9b.sh
+eval-mgtv-best.sh