update

.github/workflows/daily_tests.yaml

@@ -41,8 +41,8 @@ jobs:
         source activate internlm-model-latest
         pip install torch==2.2.2 torchvision==0.17.2 --index-url https://download.pytorch.org/whl/cu118
         pip install /mnt/petrelfs/qa-caif-cicd/resource/flash_attn-2.5.8+cu118torch2.2cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
-        pip install sentencepiece auto-gptq==0.6.0 lmdeploy[all] beautifulsoup4 decord lxml
-        srun -p ${SLURM_PARTITION} --kill-on-bad-exit=1 --job-name=${GITHUB_RUN_ID}-${GITHUB_JOB} --gpus-per-task=2 pytest -s -v -m tmp --color=yes ./tests/test_hf_model.py
+        pip install sentencepiece auto-gptq==0.6.0 lmdeploy[all]==0.5.0 beautifulsoup4 decord lxml
+        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
         conda deactivate
     - name: remove_env
       if: always()

tests/test_hf_model.py

@@ -282,7 +282,6 @@ class InternLMXComposer2QForCausalLM(BaseGPTQForCausalLM):
     ]
 
 
-@pytest.mark.tmp
 class TestXcomposer2d5Model:
     """Test cases for base model."""
 
@@ -298,7 +297,7 @@ class TestXcomposer2d5Model:
         # init model and tokenizer
         model = AutoModel.from_pretrained(
             model_name, torch_dtype=torch.bfloat16,
-            trust_remote_code=True).cuda().eval()
+            trust_remote_code=True).cuda().eval().half()
         tokenizer = AutoTokenizer.from_pretrained(model_name,
                                                   trust_remote_code=True)
         model.tokenizer = tokenizer
@@ -394,7 +393,9 @@ class TestXcomposer2d5Model:
         model.tokenizer = tokenizer
 
         query = 'Generate the HTML code of this web image with Tailwind CSS.'
-        image = ['/mnt/petrelfs/qa-caif-cicd/github_runner/examples/screenshot.jpg']
+        image = [
+            '/mnt/petrelfs/qa-caif-cicd/github_runner/examples/screenshot.jpg'
+        ]
         with torch.autocast(device_type='cuda', dtype=torch.float16):
             response = model.resume_2_webpage(query,
                                               image,