diff --git a/pyproject.toml b/pyproject.toml
index 2aaa1c0..c4fbbca 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,16 +1,16 @@
 [tool.poetry]
 name = "sllim"
-version = "0.1.25"
+version = "0.1.26"
 description = "Fixing the openai api"
 authors = ["Kaiser Pister <kaiser@pister.dev>"]
 readme = "README.md"
 homepage = "https://github.com/kpister/sllim"
 
 [tool.poetry.dependencies]
-python = "^3.9"
-openai = "^0.27.8"
-tiktoken = "^0.4.0"
-pyspellchecker = "^0.7.2"
+python = ">= 3.9"
+openai = ">= 0.28"
+tiktoken = ">= 0.4.0"
+pyspellchecker = ">= 0.7.2"
 
 [tool.poetry.scripts]
 check = "sllim.check:run"
diff --git a/sllim/__init__.py b/sllim/__init__.py
index 22794f0..c330625 100644
--- a/sllim/__init__.py
+++ b/sllim/__init__.py
@@ -23,12 +23,15 @@
 
 prompt_tokens, completion_tokens = 0, 0
 
+
 def load_template(filepath: str) -> tuple[str, str]:
     with open(filepath, "r") as f:
         description, text = f.read().split("\n", 1)
         if not description.startswith("#"):
             # Not a proper prompt file
-            logger.warning(f"File {filepath} does not start with a `# description line`.")
+            logger.warning(
+                f"File {filepath} does not start with a `# description line`."
+            )
             text = description + "\n" + text
             description = ""
         return description, text.strip()
@@ -38,11 +41,13 @@ class Message(TypedDict):
     role: str
     content: str
 
+
 class FunctionT(TypedDict):
     name: str
     description: str
     parameters: dict[str, dict[str, str | dict]]
 
+
 Prompt = TypeVar("Prompt", str, list[Message])
 
 API_PARAMS = dict(
@@ -76,6 +81,12 @@ def get_token_counts():
     return prompt_tokens, completion_tokens
 
 
+def reset_token_counts():
+    global prompt_tokens, completion_tokens
+
+    prompt_tokens, completion_tokens = 0, 0
+
+
 def try_make(folder_name):
     try:
         os.makedirs(folder_name, exist_ok=True)
@@ -93,6 +104,7 @@ def close(self, *args, **kwargs):
     def read(self, *args, **kwargs):
         return "{}"
 
+
 @contextmanager
 def try_open(filename, mode="r"):
     try:
@@ -107,7 +119,6 @@ def try_open(filename, mode="r"):
         f.close()
 
 
-
 def cache(fn):
     folder_name = ".cache"
     cache_file = os.path.join(folder_name, f"{fn.__name__}.json")
@@ -202,7 +213,9 @@ def chat(
     }
     max_tokens_str = "infinity" if max_tokens is None else str(max_tokens)
     model_str = model if model else deployment_id
-    logger.info(f"Calling {model_str} using at most {max_tokens_str} with messages: {messages}")
+    logger.info(
+        f"Calling {model_str} using at most {max_tokens_str} with messages: {messages}"
+    )
 
     response = openai.ChatCompletion.create(
         messages=messages,
@@ -263,12 +276,13 @@ def call(
     message = response.choices[0].message
     prompt_tokens += response.usage.prompt_tokens
     completion_tokens += response.usage.completion_tokens
-    
+
     if message.get("function_call"):
         return message.function_call
-    
+
     raise Exception("No function call found in response. %s" % str(message))
 
+
 @catch
 @cache
 def complete(
@@ -311,19 +325,23 @@ def complete(
 
 @catch
 @cache
-def embed(text: str | list[str], engine="text-embedding-ada-002", deployment_id: str=None) -> list[float] | list[list[float]]:
+def embed(
+    text: str | list[str], engine="text-embedding-ada-002", deployment_id: str = None
+) -> list[float] | list[list[float]]:
     if deployment_id:
         engine = None
-    kwargs = {k: v for k, v in locals().items() if k in ["engine", "deployment_id"] and v is not None}
+    kwargs = {
+        k: v
+        for k, v in locals().items()
+        if k in ["engine", "deployment_id"] and v is not None
+    }
     if isinstance(text, list):
         text = [t.replace("\n", " ") for t in text]
         response = openai.Embedding.create(input=text, **kwargs)["data"]
         return [x["embedding"] for x in response]
     else:
         text = text.replace("\n", " ")
-        return openai.Embedding.create(input=[text], **kwargs)["data"][0][
-            "embedding"
-        ]
+        return openai.Embedding.create(input=[text], **kwargs)["data"][0]["embedding"]
 
 
 def estimate(messages_or_prompt: Prompt, model="gpt-3.5-turbo"):
@@ -465,6 +483,7 @@ def map_reduce(template: Prompt, n=8, **kwargs):
     finally:
         collate_caches(params["__function"])
 
+
 def thread_map_reduce(template: Prompt, n=8, **kwargs):
     params = {}
     for key, value in kwargs.items():
@@ -490,9 +509,12 @@ def thread_map_reduce(template: Prompt, n=8, **kwargs):
     iterator = to_slices(template, iters, constants)
 
     results = ["" for _ in range(max_len)]
+
     def thread_call(message, idx):
         try:
-            results[idx] = fn(message, **{k: v for k, v in params.items()}, nocache=True)
+            results[idx] = fn(
+                message, **{k: v for k, v in params.items()}, nocache=True
+            )
         except RateLimitError:
             results[idx] = ""
 
@@ -509,7 +531,7 @@ def thread_call(message, idx):
 
     for thread in threads:
         thread.join()
-    
+
     return results
 
 
@@ -534,16 +556,17 @@ def collate_caches(function_name):
     with try_open(cache_file, "w") as w:
         json.dump(cache, w)
 
+
 def to_type_name(_type: str):
     if "list" in _type or "tuple" in _type:
         return "array", {"items": {"type": "string"}}
-    
 
     return {
         "str": "string",
         "int": "number",
     }.get(_type, _type), {}
 
+
 def parse_doc(doc: str):
     if not doc:
         return "", {}, []
@@ -576,6 +599,7 @@ def parse_doc(doc: str):
 
     return fn_description, properties, required
 
+
 def create_function_call(fn: Callable) -> FunctionT:
     description, properties, required = parse_doc(fn.__doc__)
     return {
@@ -585,10 +609,11 @@ def create_function_call(fn: Callable) -> FunctionT:
             "type": "object",
             "properties": properties,
             "required": required,
-        }
+        },
     }
 
+
 def format(s: str, **kwargs):
     for key, value in kwargs.items():
         s = s.replace("{" + key + "}", value)
-    return s
\ No newline at end of file
+    return s
diff --git a/sllim/check.py b/sllim/check.py
index b6bcd9c..099e386 100644
--- a/sllim/check.py
+++ b/sllim/check.py
@@ -12,7 +12,7 @@
 logger = logging.getLogger(__name__)
 
 spell = SpellChecker()
-spell.word_frequency.load_words(['schema', 'bulleted'])
+spell.word_frequency.load_words(["schema", "bulleted"])
 
 
 @dataclass
@@ -25,6 +25,7 @@ class PromptFile:
     extra_variables: list[str] = field(default_factory=list)
     missing_variables: list[str] = field(default_factory=list)
 
+
 @dataclass
 class PythonFile:
     filename: str
@@ -53,13 +54,18 @@ def gather(directory, ignore=None, ext=".txt"):
 
     return out_files
 
+
 def check_spelling(pfile: PromptFile):
     results = spell.unknown(spell.split_words(pfile.text))
-    pfile.typos = list(filter(lambda x: len(x) > 3 and "_" not in x and spell.candidates(x), results))
+    pfile.typos = list(
+        filter(lambda x: len(x) > 3 and "_" not in x and spell.candidates(x), results)
+    )
+
 
 def load_variables(pfile: PromptFile):
     pfile.variables = set(re.findall(r"\{(\w+)\}", pfile.text))
 
+
 def check_usage(tfile: PromptFile, pyfiles: list[PythonFile]):
     for pyfile in pyfiles:
         for _, (filename, usages) in pyfile.uses.items():
@@ -67,14 +73,19 @@ def check_usage(tfile: PromptFile, pyfiles: list[PythonFile]):
                 continue
 
             for variables in usages:
-                for variable in (set(variables) - set(tfile.variables)):
-                    logger.warning(f"{os.path.relpath(pyfile.filename)}: Variable {variable} not defined in {os.path.relpath(tfile.filename)}.")
+                for variable in set(variables) - set(tfile.variables):
+                    logger.warning(
+                        f"{os.path.relpath(pyfile.filename)}: Variable {variable} not defined in {os.path.relpath(tfile.filename)}."
+                    )
                     tfile.extra_variables.append(variable)
-                
-                for variable in (set(tfile.variables) - set(variables)):
-                    logger.warning(f"{os.path.relpath(pyfile.filename)}: Variable {variable} not used in {os.path.relpath(tfile.filename)}.")
+
+                for variable in set(tfile.variables) - set(variables):
+                    logger.warning(
+                        f"{os.path.relpath(pyfile.filename)}: Variable {variable} not used in {os.path.relpath(tfile.filename)}."
+                    )
                     tfile.missing_variables.append(variable)
 
+
 def check_file(filepath, pyfiles: list[PythonFile]):
     description, prompt = load_template(filepath)
 
@@ -84,6 +95,7 @@ def check_file(filepath, pyfiles: list[PythonFile]):
     check_usage(pfile, pyfiles)
     return pfile
 
+
 def load_python_file(filepath):
     try:
         with open(filepath, "r") as f:
@@ -91,10 +103,10 @@ def load_python_file(filepath):
     except UnicodeDecodeError:
         logger.warning(f"Unable to read {filepath}.")
         return None
-    
+
     if "sllim" not in text or "load_template" not in text:
         return None
-    
+
     track_ids = walk(ast.parse(text))
     python_file = PythonFile(filename=filepath, text=text, uses=track_ids)
     return python_file
@@ -112,16 +124,26 @@ def get_arg(node: ast.Call):
         elif isinstance(node.keywords[0], ast.Name):
             return node.keywords[0].id
     return None
-    
+
+
 def handle_load_template(node: ast.AST):
-    if isinstance(node, ast.Call) and hasattr(node.func, "id") and node.func.id == "load_template":
+    if (
+        isinstance(node, ast.Call)
+        and hasattr(node.func, "id")
+        and node.func.id == "load_template"
+    ):
         return get_arg(node)
 
     return None
 
+
 def handle_use_template(ancestor: ast.AST, node: ast.AST, track_ids):
     if isinstance(node, ast.Name) and node.id in track_ids:
-        if isinstance(ancestor, ast.Call) and hasattr(ancestor.func, "id") and ancestor.func.id == "format":
+        if (
+            isinstance(ancestor, ast.Call)
+            and hasattr(ancestor.func, "id")
+            and ancestor.func.id == "format"
+        ):
             return get_arg(ancestor), ancestor.keywords
     elif isinstance(node, ast.Attribute) and node.attr == "format":
         if isinstance(ancestor, ast.Call):
@@ -129,6 +151,7 @@ def handle_use_template(ancestor: ast.AST, node: ast.AST, track_ids):
 
     return None, None
 
+
 def walk(tree: ast.AST):
     track_ids = {}
     for ancestor in ast.walk(tree):
@@ -139,13 +162,14 @@ def walk(tree: ast.AST):
                 else:
                     tid = ancestor.targets[0].id
                 track_ids[tid] = (value, [])
-            else: 
+            else:
                 key, keywords = handle_use_template(ancestor, child, track_ids)
                 if key and keywords and key in track_ids:
                     track_ids[key][1].append([x.arg for x in keywords])
 
     return track_ids
 
+
 def run_checks(text_files, python_files):
     tfiles = []
     pyfiles = []
@@ -155,14 +179,14 @@ def run_checks(text_files, python_files):
 
     for filepath in text_files:
         tfiles.append(check_file(filepath, pyfiles))
-    
+
     spelling_errors = False
     variable_errors = False
     for tfile in tfiles:
         if tfile.typos:
             logger.warning(f"Possible typos in {tfile.filename}: {tfile.typos}")
             spelling_errors = True
-        
+
         if tfile.extra_variables or tfile.missing_variables:
             variable_errors = True
 
@@ -171,17 +195,28 @@ def run_checks(text_files, python_files):
 
     if not variable_errors:
         logger.info("No variable errors found.")
-    
+
 
 def run():
     """Run check command."""
     # All the logic of argparse goes in this function
-    parser = argparse.ArgumentParser(description='Basic prompt checking.')
-    parser.add_argument('directory', type=str, help='the name of the directory to search for prompts', default=".")
-    parser.add_argument('--ignore', type=str, help='a directory or file to exclude', default=None, required=False)
+    parser = argparse.ArgumentParser(description="Basic prompt checking.")
+    parser.add_argument(
+        "directory",
+        type=str,
+        help="the name of the directory to search for prompts",
+        default=".",
+    )
+    parser.add_argument(
+        "--ignore",
+        type=str,
+        help="a directory or file to exclude",
+        default=None,
+        required=False,
+    )
 
     args = parser.parse_args()
     prompt_files = gather(args.directory, args.ignore, ext=".txt")
     prompt_files += gather(args.directory, args.ignore, ext=".prompt")
     python_files = gather(args.directory, args.ignore, ext=".py")
-    run_checks(prompt_files, python_files)
\ No newline at end of file
+    run_checks(prompt_files, python_files)