add ruff, change formats, modify README.md

README.md

@@ -197,7 +197,7 @@ docker run -it --rm \
 
 ### Step 3. Submit the request to get model response
 
-Run the following command to submit a request to the fine-tuned model and get model response. Note that the port for our inference serivice is `5001`, not `5000`.
+Run the following command to submit a request to the fine-tuned model and get model response. Note that this inference service is designed to be similar to [Cohere's Chat API](https://docs.cohere.com/v1/reference/chat), and the port for this inference service is `5001`, not `5000`.
 ```commandline
 curl --request POST http://localhost:5001/inference \
     --header "Content-Type: application/json" \

pyproject.toml

@@ -18,6 +18,7 @@ dependencies = [
     "pytest>=8.3.3",
     "pytest-cov>=5.0.0",
     "pytest-xdist>=3.6.1",
+    "ruff>=0.7.2",
 ]
 requires-python = ">=3.10,<4.0"
 readme = "README.md"
@@ -31,6 +32,27 @@ build-backend = "pdm.backend"
 distribution = true
 
 [tool.pytest.ini_options]
-pythonpath = [
-  "src/cohere_finetune"
+pythonpath = ["src/cohere_finetune"]
+
+[tool.ruff]
+src = ["src/cohere_finetune"]
+line-length = 119
+extend-exclude = ["*.ipynb"]
+
+[tool.ruff.lint]
+extend-select = [
+    "C", # Complexity
+    "E", # PEP8 errors
+    "F", # PEP8 formatting
+    "I", # Import sorting
+    "UP", # Pyupgrade upgrades
+    "W", # PEP8 warnings
+    "PT009", # Pytest assertions
+]
+ignore = [
+    "C901", # Function too complex
+    "E501", # Line length (handled by ruff-format)
+    "I001", # Import block is un-sorted or un-formatted
+    "UP007", # X | Y style Unions
+    "UP015", # Unnecessary open mode parameters
 ]

src/cohere_finetune/chat_utils.py

@@ -33,9 +33,9 @@ def chat_to_str(chat: dict) -> str:
     """
     # Create a normalized chat, where the order of keys in each message is always: "role" and then "content" (if the
     # only difference between two chats is the order of keys in each message, we want to regard them as the same chat)
-    normalized_chat = {"messages": [
-        {"role": message["role"], "content": message["content"]} for message in chat["messages"]
-    ]}
+    normalized_chat = {
+        "messages": [{"role": message["role"], "content": message["content"]} for message in chat["messages"]]
+    }
     return json.dumps(normalized_chat)
 
 

src/cohere_finetune/cohere_finetune_service.py

@@ -96,7 +96,7 @@ def finetune(self) -> None:
             trc = traceback.format_exc()
             self._format_error(
                 logging_msg=f"Missing files during dataset creation: {trc}",
-                error_msg=f"Missing files during dataset creation: {e}"
+                error_msg=f"Missing files during dataset creation: {e}",
             )
             return
         except ValueError as e:
@@ -128,7 +128,7 @@ def finetune(self) -> None:
             trc = traceback.format_exc()
             self._format_error(
                 logging_msg=f"Exception during preprocessing: {trc}",
-                error_msg=f"Exception during preprocessing: {e}"
+                error_msg=f"Exception during preprocessing: {e}",
             )
             return
 
@@ -145,7 +145,8 @@ def finetune(self) -> None:
             trc = traceback.format_exc()
             self._format_error(
                 logging_msg=f"Exception during max sequence length calculation: {trc}",
-                error_msg=f"Exception during max sequence length calculation: {e}")
+                error_msg=f"Exception during max sequence length calculation: {e}",
+            )
             return
         except Exception:
             # If the above procedure fails for any other reason, use the default max_sequence_length in hyperparameters

src/cohere_finetune/finetune_backends/cohere_peft/peft_arguments.py

@@ -1,7 +1,7 @@
 from dataclasses import dataclass, field
 from transformers import TrainingArguments
 from transformers.trainer_utils import IntervalStrategy, SchedulerType
-from typing import List, Union
+from typing import Union
 
 
 @dataclass
@@ -208,7 +208,7 @@ class TrainingArgumentsDefaultChanged(TrainingArguments):
         default=1,
         metadata={"help": "Total number of training epochs to perform"},
     )
-    report_to: Union[None, str, List[str]] = field(
+    report_to: Union[None, str, list[str]] = field(
         default="none",
         metadata={"help": "The list of integrations to report the results and logs to"},
     )

src/cohere_finetune/finetune_backends/cohere_peft/peft_model.py

@@ -81,9 +81,7 @@ def create_and_prepare_model(
         See https://huggingface.co/docs/transformers/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained
         """
         torch_dtype = (
-            quant_storage_dtype
-            if quant_storage_dtype and quant_storage_dtype.is_floating_point
-            else torch.bfloat16
+            quant_storage_dtype if quant_storage_dtype and quant_storage_dtype.is_floating_point else torch.bfloat16
         )
         if os.environ.get("ACCELERATE_USE_FSDP", "false") == "true" or os.environ.get("ACCELERATE_USE_DEEPSPEED", "false") == "true":
             # Can't use device_map if you use fsdp or deepspeed

src/cohere_finetune/utils.py

@@ -123,14 +123,14 @@ def save_file(x: Any, path: str, overwrite_ok: bool = False) -> None:
             for entry in x:
                 assert isinstance(entry, dict)
                 json.dump(entry, f)
-                f.write('\n')
+                f.write("\n")
     elif ext in {".txt"}:
         if isinstance(x, list):
             with open(path, "w") as f:
                 for entry in x:
                     assert isinstance(entry, str)
                     f.write(entry)
-                    f.write('\n')
+                    f.write("\n")
         else:
             assert isinstance(x, str)
             with open(path, "w") as f: