chore: remove torch in main code (#439)

* use np.ndarray in costeer model evaluators

* remove package

* fix test error

---------

Co-authored-by: Xu Yang <[email protected]>

Author: SunsetWolf

constraints/3.10.txt

@@ -71,7 +71,6 @@ isort==5.13.2
 jaraco.classes==3.3.0
 jedi==0.19.1
 jeepney==0.8.0
-Jinja2==3.1.2
 joblib==1.4.2
 json5==0.9.25
 jsonpatch==1.33
@@ -102,15 +101,13 @@ loguru==0.7.2
 loguru-mypy==0.0.4
 lxml==5.0.0
 markdown-it-py==3.0.0
-MarkupSafe==2.1.3
 marshmallow==3.20.1
 matplotlib==3.9.1
 matplotlib-inline==0.1.7
 mdit-py-plugins==0.4.0
 mdurl==0.1.2
 mistune==3.0.2
 more-itertools==10.1.0
-mpmath==1.3.0
 msal==1.30.0
 msal-extensions==1.2.0
 msgpack==1.0.8
@@ -124,24 +121,11 @@ nbconvert==7.16.4
 nbformat==5.10.4
 ndindex==1.8
 nest-asyncio==1.6.0
-networkx==3.2.1
 nh3==0.2.15
 notebook==7.2.1
 notebook_shim==0.2.4
 numexpr==2.10.1
 numpy==1.26.2
-nvidia-cublas-cu12==12.1.3.1
-nvidia-cuda-cupti-cu12==12.1.105
-nvidia-cuda-nvrtc-cu12==12.1.105
-nvidia-cuda-runtime-cu12==12.1.105
-nvidia-cudnn-cu12==8.9.2.26
-nvidia-cufft-cu12==11.0.2.54
-nvidia-curand-cu12==10.3.2.106
-nvidia-cusolver-cu12==11.4.5.107
-nvidia-cusparse-cu12==12.1.0.106
-nvidia-nccl-cu12==2.18.1
-nvidia-nvjitlink-cu12==12.3.101
-nvidia-nvtx-cu12==12.1.105
 oauthlib==3.2.2
 openai==1.6.1
 overrides==7.7.0
@@ -153,7 +137,6 @@ parso==0.8.4
 pathspec==0.12.1
 patsy==0.5.6
 pexpect==4.9.0
-pillow==10.4.0
 pkginfo==1.9.6
 platformdirs==4.1.0
 pluggy==1.3.0
@@ -201,7 +184,6 @@ ruamel.yaml==0.18.5
 ruamel.yaml.clib==0.2.8
 ruff==0.4.5
 scikit-learn==1.5.1
-scipy==1.11.4
 SecretStorage==3.3.3
 semver==3.0.2
 Send2Trash==1.8.3
@@ -226,7 +208,6 @@ sphinxcontrib-serializinghtml==1.1.9
 SQLAlchemy==2.0.24
 stack-data==0.6.3
 statsmodels==0.14.2
-sympy==1.12
 tables==3.9.2
 tabulate==0.9.0
 tenacity==8.2.3
@@ -238,22 +219,18 @@ tinycss2==1.3.0
 toml-sort==0.23.1
 tomli==2.0.1
 tomlkit==0.12.3
-torch==2.1.2
-torch_geometric==2.5.3
 tornado==6.4
 tqdm==4.66.1
 traitlets==5.14.3
 tree-sitter==0.22.3
 tree-sitter-python==0.21.0
-triton==2.1.0
 twine==4.0.2
 typer==0.9.0
 types-psutil==6.0.0.20240621
 types-python-dateutil==2.9.0.20240316
 types-PyYAML==6.0.12.20240724
 types-tqdm==4.66.0.20240417
 typing-inspect==0.9.0
-typing_extensions==4.9.0
 tzdata==2023.4
 uri-template==1.3.0
 urllib3==2.1.0

constraints/3.11.txt

@@ -69,7 +69,6 @@ isort==5.13.2
 jaraco.classes==3.3.0
 jedi==0.19.1
 jeepney==0.8.0
-Jinja2==3.1.2
 joblib==1.4.2
 json5==0.9.25
 jsonpatch==1.33
@@ -100,15 +99,13 @@ loguru==0.7.2
 loguru-mypy==0.0.4
 lxml==5.0.0
 markdown-it-py==3.0.0
-MarkupSafe==2.1.3
 marshmallow==3.20.1
 matplotlib==3.9.1
 matplotlib-inline==0.1.7
 mdit-py-plugins==0.4.0
 mdurl==0.1.2
 mistune==3.0.2
 more-itertools==10.1.0
-mpmath==1.3.0
 msal==1.30.0
 msal-extensions==1.2.0
 msgpack==1.0.8
@@ -122,24 +119,11 @@ nbconvert==7.16.4
 nbformat==5.10.4
 ndindex==1.8
 nest-asyncio==1.6.0
-networkx==3.2.1
 nh3==0.2.15
 notebook==7.2.1
 notebook_shim==0.2.4
 numexpr==2.10.1
 numpy==1.26.2
-nvidia-cublas-cu12==12.1.3.1
-nvidia-cuda-cupti-cu12==12.1.105
-nvidia-cuda-nvrtc-cu12==12.1.105
-nvidia-cuda-runtime-cu12==12.1.105
-nvidia-cudnn-cu12==8.9.2.26
-nvidia-cufft-cu12==11.0.2.54
-nvidia-curand-cu12==10.3.2.106
-nvidia-cusolver-cu12==11.4.5.107
-nvidia-cusparse-cu12==12.1.0.106
-nvidia-nccl-cu12==2.18.1
-nvidia-nvjitlink-cu12==12.3.101
-nvidia-nvtx-cu12==12.1.105
 oauthlib==3.2.2
 openai==1.6.1
 overrides==7.7.0
@@ -151,7 +135,6 @@ parso==0.8.4
 pathspec==0.12.1
 patsy==0.5.6
 pexpect==4.9.0
-pillow==10.4.0
 pkginfo==1.9.6
 platformdirs==4.1.0
 pluggy==1.3.0
@@ -199,7 +182,6 @@ ruamel.yaml==0.18.5
 ruamel.yaml.clib==0.2.8
 ruff==0.4.5
 scikit-learn==1.5.1
-scipy==1.11.4
 SecretStorage==3.3.3
 semver==3.0.2
 Send2Trash==1.8.3
@@ -224,7 +206,6 @@ sphinxcontrib-serializinghtml==1.1.9
 SQLAlchemy==2.0.24
 stack-data==0.6.3
 statsmodels==0.14.2
-sympy==1.12
 tables==3.9.2
 tabulate==0.9.0
 tenacity==8.2.3
@@ -235,22 +216,18 @@ tiktoken==0.7.0
 tinycss2==1.3.0
 toml-sort==0.23.1
 tomlkit==0.12.3
-torch==2.1.2
-torch_geometric==2.5.3
 tornado==6.4
 tqdm==4.66.1
 traitlets==5.14.3
 tree-sitter==0.22.3
 tree-sitter-python==0.21.0
-triton==2.1.0
 twine==4.0.2
 typer==0.9.0
 types-psutil==6.0.0.20240621
 types-python-dateutil==2.9.0.20240316
 types-PyYAML==6.0.12.20240724
 types-tqdm==4.66.0.20240417
 typing-inspect==0.9.0
-typing_extensions==4.9.0
 tzdata==2023.4
 uri-template==1.3.0
 urllib3==2.1.0

rdagent/components/coder/model_coder/CoSTEER/evaluators.py

@@ -4,7 +4,6 @@
 from typing import List, Tuple
 
 import numpy as np
-import torch
 from jinja2 import Environment, StrictUndefined
 
 from rdagent.components.coder.model_coder.conf import MODEL_IMPL_SETTINGS
@@ -25,7 +24,7 @@
 evaluate_prompts = Prompts(file_path=Path(__file__).parent.parent / "prompts.yaml")
 
 
-def shape_evaluator(prediction: torch.Tensor | np.ndarray, target_shape: Tuple = None) -> Tuple[str, bool]:
+def shape_evaluator(prediction: np.ndarray, target_shape: Tuple = None) -> Tuple[str, bool]:
     if target_shape is None or prediction is None:
         return (
             "No output generated from the model. No shape evaluation conducted.",
@@ -42,18 +41,10 @@ def shape_evaluator(prediction: torch.Tensor | np.ndarray, target_shape: Tuple =
         )
 
 
-def reshape_tensor(original_tensor, target_shape):
-    new_tensor = torch.zeros(target_shape)
-    for i, dim in enumerate(original_tensor.shape):
-        new_tensor = new_tensor.narrow(i, 0, dim).copy_(original_tensor)
-
-    return new_tensor
-
-
 def value_evaluator(
-    prediction: torch.Tensor,
-    target: torch.Tensor,
-) -> Tuple[torch.Tensor, bool]:
+    prediction: np.ndarray,
+    target: np.ndarray,
+) -> Tuple[np.ndarray, bool]:
     if prediction is None:
         return "No output generated from the model. Skip value evaluation", False
     elif target is None:
@@ -63,7 +54,7 @@ def value_evaluator(
         )
     else:
         # Calculate the mean absolute difference
-        diff = torch.mean(torch.abs(target - prediction)).item()
+        diff = np.mean(np.abs(target - prediction))
         return (
             f"The value of the output is correct. The mean absolute difference is {diff}.",
             diff < 0.1,
@@ -273,7 +264,7 @@ def evaluate(
         param_init_value = 0.6
 
         assert isinstance(implementation, ModelFBWorkspace)
-        model_execution_feedback, gen_tensor = implementation.execute(
+        model_execution_feedback, gen_np_array = implementation.execute(
             batch_size=batch_size,
             num_features=num_features,
             num_timesteps=num_timesteps,
@@ -282,18 +273,18 @@ def evaluate(
         )
         if gt_implementation is not None:
             assert isinstance(gt_implementation, ModelFBWorkspace)
-            _, gt_tensor = gt_implementation.execute(
+            _, gt_np_array = gt_implementation.execute(
                 batch_size=batch_size,
                 num_features=num_features,
                 num_timesteps=num_timesteps,
                 input_value=input_value,
                 param_init_value=param_init_value,
             )
         else:
-            gt_tensor = None
+            gt_np_array = None
 
-        shape_feedback, shape_decision = shape_evaluator(gen_tensor, (batch_size, 1))
-        value_feedback, value_decision = value_evaluator(gen_tensor, gt_tensor)
+        shape_feedback, shape_decision = shape_evaluator(gen_np_array, (batch_size, 1))
+        value_feedback, value_decision = value_evaluator(gen_np_array, gt_np_array)
         code_feedback, _ = ModelCodeEvaluator(scen=self.scen).evaluate(
             target_task=target_task,
             implementation=implementation,

rdagent/components/coder/model_coder/model_execute_template_v1.txt

@@ -37,7 +37,7 @@ if MODEL_TYPE == "Graph":
 else:
     out = m(data)
 
-execution_model_output = out.cpu().detach()
+execution_model_output = out.cpu().detach().numpy()
 execution_feedback_str = f"Execution successful, output tensor shape: {execution_model_output.shape}"
 
 pickle.dump(execution_model_output, open("execution_model_output.pkl", "wb"))

rdagent/components/coder/model_coder/model_execute_template_v2.txt

@@ -12,7 +12,7 @@ valid_X = pd.DataFrame(np.random.randn(8, 30), columns=[f"{i}" for i in range(30
 valid_y = pd.Series(np.random.randint(0, 2, 8))
 
 model = fit(train_X, train_y, valid_X, valid_y)
-execution_model_output = predict(model, valid_X)
+execution_model_output = predict(model, valid_X).cpu().detach().numpy()
 
 execution_feedback_str = f"Execution successful, output numpy ndarray shape: {execution_model_output.shape}"
 

requirements.txt

@@ -3,7 +3,6 @@ pydantic-settings
 typer[all]
 
 cython
-scipy
 python-Levenshtein
 scikit-learn
 filelock
@@ -14,8 +13,6 @@ fuzzywuzzy
 openai
 
 ruamel-yaml
-torch
-torch_geometric
 tabulate  # Convert pandas dataframe to markdown table to make it more readable to LLM
 numpy # we use numpy as default data format. So we have to install numpy
 pandas # we use pandas as default data format. So we have to install pandas
@@ -69,5 +66,4 @@ seaborn
 setuptools-scm
 
 # This is a temporary package installed to pass the test_import test
-xgboost
 lightgbm

test/utils/test_import.py

@@ -21,6 +21,8 @@ def import_all_modules_from_directory(directory):
                 continue
             if "_template" in fstr:
                 continue
+            if "model_coder" in fstr:
+                continue
             if (
                 fstr.endswith("rdagent/log/ui/app.py")
                 or fstr.endswith("rdagent/app/cli.py")