set max_new_tokens to 300

.env.example

@@ -1,5 +1,6 @@
 MODEL_NAME=internlm/internlm2_5-7b-chat-1m
 BATCH_SIZE=2
+MAX_NEW_TOKENS=300
 
 HF_TOKEN=
 

llm_toolkit/eval.py

@@ -26,9 +26,18 @@ load_in_4bit = os.getenv("LOAD_IN_4BIT") == "true"
 data_path = os.getenv("DATA_PATH")
 results_path = os.getenv("RESULTS_PATH")
 batch_size = int(os.getenv("BATCH_SIZE", 1))
+use_english_datasets = os.getenv("USE_ENGLISH_DATASETS") == "true"
+max_new_tokens = int(os.getenv("MAX_NEW_TOKENS", 2048))
 
 print(
-    model_name, adapter_name_or_path, load_in_4bit, data_path, results_path, batch_size
+    model_name,
+    adapter_name_or_path,
+    load_in_4bit,
+    data_path,
+    results_path,
+    use_english_datasets,
+    max_new_tokens,
+    batch_size,
 )
 
 if is_cuda:
@@ -88,6 +97,7 @@ evaluate_model_with_repetition_penalty(
     end_repetition_penalty=1.3,
     step_repetition_penalty=0.02,
     batch_size=batch_size,
+    max_new_tokens=max_new_tokens,
     device=device,
 )
 

llm_toolkit/llm_utils.py

@@ -224,6 +224,7 @@ def evaluate_model_with_repetition_penalty(
     end_repetition_penalty=1.3,
     step_repetition_penalty=0.02,
     batch_size=1,
+    max_new_tokens=2048,
     device="cuda",
 ):
     print(f"Evaluating model: {model_name} on {device}")
@@ -243,6 +244,7 @@ def evaluate_model_with_repetition_penalty(
             device=device,
             repetition_penalty=repetition_penalty,
             batch_size=batch_size,
+            max_new_tokens=max_new_tokens,
         )
 
         model_name_with_rp = f"{model_name}/rpp-{repetition_penalty:.2f}"

scripts/eval-mac.sh

@@ -14,12 +14,12 @@ grep MemTotal /proc/meminfo
 # pip install torch torchvision torchaudio
 # pip install -r requirements.txt
 
-#./scripts/eval-model.sh shenzhi-wang/Mistral-7B-v0.3-Chinese-Chat
+./scripts/eval-model.sh shenzhi-wang/Mistral-7B-v0.3-Chinese-Chat
 
 #./scripts/eval-model.sh Qwen/Qwen2-7B-Instruct
 
 #./scripts/eval-model.sh shenzhi-wang/Llama3.1-8B-Chinese-Chat
 
-#./scripts/eval-model.sh 01-ai/Yi-1.5-9B-Chat
+./scripts/eval-model.sh 01-ai/Yi-1.5-9B-Chat
 
 ./scripts/eval-model.sh internlm/internlm2_5-7b-chat