DSPy: The framework for programming—not prompting—language models
Python
Linting of dummy
Thomas D Ahlea021bd26763a
diff --git a/dspy/utils/dummies.py b/dspy/utils/dummies.pyindex a0c997145..3a3738807 100644--- a/dspy/utils/dummies.py+++ b/dspy/utils/dummies.py@@ -1,6 +1,5 @@import randomfrom dsp.modules import LM-from typing import List, Union, Dictimport numpy as npfrom dsp.utils.utils import dotdictimport re@@ -9,9 +8,9 @@ import reclass DummyLM(LM):"""Dummy language model for unit testing purposes."""- def __init__(self, answers: Union[List[str], Dict[str,str]], follow_examples: bool = False):- """- Initializes the dummy language model.+ def __init__(self, answers: list[str] | dict[str, str], follow_examples: bool = False):+ """Initializes the dummy language model.+Parameters:- answers: A list of strings or a dictionary with string keys and values.- follow_examples: If True, and the prompt contains an example exactly equal to the prompt,@@ -24,7 +23,7 @@ class DummyLM(LM):self.answers = answersself.follow_examples = follow_examples- def basic_request(self, prompt, n=1, **kwargs):+ def basic_request(self, prompt, n=1, **kwargs) -> dict[str, list[dict[str, str]]]:"""Generates a dummy response based on the prompt."""dummy_response = {"choices": []}for _ in range(n):@@ -55,12 +54,14 @@ class DummyLM(LM):answer = "No more responses"# Mimic the structure of a real language model response.- dummy_response["choices"].append({- "text": answer,- "finish_reason": "simulated completion",- })-- RED, GREEN, RESET = '\033[91m', '\033[92m', '\033[0m'+ dummy_response["choices"].append(+ {+ "text": answer,+ "finish_reason": "simulated completion",+ },+ )++ RED, GREEN, RESET = "\033[91m", "\033[92m", "\033[0m"print("=== DummyLM ===")print(prompt, end="")print(f"{RED}{answer}{RESET}")@@ -77,67 +78,68 @@ class DummyLM(LM):return dummy_response- def __call__(self, prompt, only_completed=True, return_sorted=False, **kwargs):+ def __call__(self, prompt, _only_completed=True, _return_sorted=False, **kwargs):"""Retrieves dummy completions."""response = self.basic_request(prompt, **kwargs)choices = response["choices"]# Filter choices and return text completions.- completions = [choice["text"] for choice in choices]-- return completions+ return [choice["text"] for choice in choices]- def get_convo(self, index):- """Get the prompt + anwer from the ith message"""- return self.history[index]['prompt'] \- + " " \- + self.history[index]['response']['choices'][0]['text']+ def get_convo(self, index) -> str:+ """Get the prompt + anwer from the ith message."""+ return self.history[index]["prompt"] + " " + self.history[index]["response"]["choices"][0]["text"]-def dummy_rm(passages=()):+def dummy_rm(passages=()) -> callable:if not passages:- def inner(query:str, *, k:int, **kwargs):++ def inner(query: str, *, k: int, **kwargs):assert False, "No passages defined"+return innermax_length = max(map(len, passages)) + 100vectorizer = DummyVectorizer(max_length)passage_vecs = vectorizer(passages)- def inner(query:str, *, k:int, **kwargs):++ def inner(query: str, *, k: int, **kwargs):assert k <= len(passages)query_vec = vectorizer([query])[0]scores = passage_vecs @ query_veclargest_idx = (-scores).argsort()[:k]- #return dspy.Prediction(passages=[passages[i] for i in largest_idx])+ # return dspy.Prediction(passages=[passages[i] for i in largest_idx])return [dotdict(dict(long_text=passages[i])) for i in largest_idx]+return innerclass DummyVectorizer:- """Simple vectorizer based on n-grams"""+ """Simple vectorizer based on n-grams."""+def __init__(self, max_length=100, n_gram=2):self.max_length = max_lengthself.n_gram = n_gramself.P = 10**9 + 7 # A large prime numberrandom.seed(123)self.coeffs = [random.randrange(1, self.P) for _ in range(n_gram)]-+def _hash(self, gram):- """Hashes a string using a polynomial hash function"""+ """Hashes a string using a polynomial hash function."""h = 1for coeff, c in zip(self.coeffs, gram):h = h * coeff + ord(c)h %= self.Preturn h % self.max_length- def __call__(self, texts: List[str]) -> np.ndarray:+ def __call__(self, texts: list[str]) -> np.ndarray:vecs = []for text in texts:- grams = [text[i:i+self.n_gram] for i in range(len(text) - self.n_gram + 1)]+ grams = [text[i : i + self.n_gram] for i in range(len(text) - self.n_gram + 1)]vec = [0] * self.max_lengthfor gram in grams:vec[self._hash(gram)] += 1vecs.append(vec)-+vecs = np.array(vecs, dtype=np.float32)vecs -= np.mean(vecs, axis=1, keepdims=True)vecs /= np.linalg.norm(vecs, axis=1, keepdims=True) + 1e-10 # Added epsilon to avoid division by zero