diff --git a/.github/workflows/daily_tests.yaml b/.github/workflows/daily_tests.yaml
new file mode 100644
index 00000000..83c16c36
--- /dev/null
+++ b/.github/workflows/daily_tests.yaml
@@ -0,0 +1,27 @@
+name: basic-model-tests-daily
+on:
+  pull_request:
+  workflow_dispatch:
+  schedule:
+    - cron:  '48 19 * * *'
+env:
+  SLURM_PARTITION: llm_s
+
+jobs:
+  HF_model:
+    runs-on: [t_cluster]
+    steps:
+    - uses: actions/checkout@v3
+
+    - name: load_hf_model
+      run: |
+        conda create -n internlm-model-latest --clone internlm-model-base
+        source activate internlm-model-latest
+        pip install transformers
+        pip install sentencepiece
+        srun -p ${SLURM_PARTITION} --kill-on-bad-exit=1 --job-name=${GITHUB_RUN_ID}-${GITHUB_JOB} --gpus-per-task=2 pytest -s -v --color=yes ./tests/test_hf_model.py
+    
+     - name: clear_env
+       run: |
+         conda deactivate 
+         conda env remove --name internlm-model-latest
diff --git a/tests/test_hf_model.py b/tests/test_hf_model.py
new file mode 100644
index 00000000..114bf21a
--- /dev/null
+++ b/tests/test_hf_model.py
@@ -0,0 +1,58 @@
+import pytest
+import torch
+
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+prompts = [
+  "你好",
+  "what's your name"
+]
+
+def assert_model(response):
+    assert len(response) != 0
+    assert "user" not in response
+    assert "bot" not in response
+    assert "UNUSED_TOKEN" not in response
+
+
+class TestChat:
+    @pytest.mark.parametrize("model_name", [
+        "internlm/internlm2-chat-7b",
+        "internlm/internlm2-chat-7b-sft",
+    ])
+    def test_demo_default(self, model_name):
+        tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+        # Set `torch_dtype=torch.float16` to load model in float16, otherwise it will be loaded as float32 and might cause OOM Error.
+        model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, trust_remote_code=True).cuda()
+        model = model.eval()
+        for prompt in prompts:
+            response, history = model.chat(tokenizer, prompt, history=[])
+            print(response)
+            assert_model(response)
+
+        for prompt in prompts:
+            length = 0
+            for response, history in model.stream_chat(tokenizer, prompt, history=[]):
+                print(response[length:], flush=True, end="")
+                length = len(response)
+                assert_model(response)
+
+
+class TestBase:
+    @pytest.mark.parametrize("model_name", [
+        "internlm/internlm2-7b",
+        "internlm/internlm2-base-7b",
+    ])
+    def test_demo_default(self, model_name):
+        tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+        # Set `torch_dtype=torch.float16` to load model in float16, otherwise it will be loaded as float32 and might cause OOM Error.
+        model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, trust_remote_code=True).cuda()
+        for prompt in prompts:
+            inputs = tokenizer(prompt, return_tensors="pt")
+            for k,v in inputs.items():
+                inputs[k] = v.cuda()
+            gen_kwargs = {"max_length": 16280, "top_p": 10, "temperature": 1.0, "do_sample": True, "repetition_penalty": 1.0}
+            output = model.generate(**inputs, **gen_kwargs)
+            output = tokenizer.decode(output[0].tolist(), skip_special_tokens=True)
+            print(output)
+            assert_model(output)