add notify

.github/workflows/daily_tests.yaml

@@ -5,23 +5,41 @@ on:
   schedule:
     - cron:  '48 19 * * *'
 env:
+  WORKSPACE_PREFIX: $(echo $GITHUB_WORKSPACE |cut -d '/' -f 1-4)
   SLURM_PARTITION: llm_s
+  CONDA_BASE_ENV: python310-torch212-cu117
 
 jobs:
   HF_model:
     runs-on: [t_cluster]
     steps:
+    - name: mask env
+      run: |
+        echo "::add-mask::${{env.WORKSPACE_PREFIX}}"
+    steps:
     - uses: actions/checkout@v3
 
     - name: load_hf_model
       run: |
-        conda create -n internlm-model-latest --clone internlm-model-base
+        conda create -n internlm-model-latest --clone ${CONDA_BASE_ENV}
         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
+     - name: clear_env
+       run: |
+         conda deactivate 
+         conda env remove --name internlm-model-latest
+
+  notify_to_feishu:
+    if: ${{ always() && !cancelled() && contains(needs.*.result, 'failure') && (github.ref_name == 'develop' || github.ref_name == 'main') }}
+    needs: [HF_model]
+    runs-on: [test_machine]
+    steps:
+    - name: mask env
       run: |
-        conda deactivate 
+        echo "::add-mask::${{env.WORKSPACE_PREFIX}}"
-        conda env remove --name internlm-model-latest
+    - name: notify
+      run: |
+        curl -X POST -H "Content-Type: application/json" -d '{"msg_type":"post","content":{"post":{"zh_cn":{"title":"Internlm GitHubAction Failed","content":[[{"tag":"text","text":""},{"tag":"a","text":"Please click here for details ","href":"https://github.com/pjlab-sys4nlp/train_internlm/actions/runs/'${GITHUB_RUN_ID}'"},{"tag":"at","user_id":"'${{ secrets.USER_ID }}'"}]]}}}}'  ${{ secrets.WEBHOOK_URL }}

tests/test_hf_model.py

@@ -1,12 +1,9 @@
 import pytest
 import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
-from transformers import AutoTokenizer, AutoModelForCausalLM
+prompts = ["你好", "what's your name"]
 
-prompts = [
-  "你好",
-  "what's your name"
-]
 
 def assert_model(response):
     assert len(response) != 0
@@ -16,14 +13,24 @@ def assert_model(response):
 
 
 class TestChat:
-    @pytest.mark.parametrize("model_name", [
+    """
-        "internlm/internlm2-chat-7b",
+    Test cases for chat model.
-        "internlm/internlm2-chat-7b-sft",
+    """
-    ])
+
+    @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.
+        # Set `torch_dtype=torch.float16` to load model in float16, otherwise
-        model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, trust_remote_code=True).cuda()
+        # 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=[])
@@ -39,19 +46,35 @@ class TestChat:
 
 
 class TestBase:
-    @pytest.mark.parametrize("model_name", [
+    """
-        "internlm/internlm2-7b",
+    Test cases for base model.
-        "internlm/internlm2-base-7b",
+    """
-    ])
+
+    @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.
+        # Set `torch_dtype=torch.float16` to load model in float16, otherwise
-        model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, trust_remote_code=True).cuda()
+        # 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():
+            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}
+            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)