Skip to content

armgit / dspy

Public Imported from GitHub
0

DSPy: The framework for programming—not prompting—language models

Python

[wip] gsm8k

manishshettymd4c5e2a9216c
diff --git a/assert_gsm8k.py b/assert_gsm8k.py
new file mode 100644
index 000000000..6367e49e3
--- /dev/null
+++ b/assert_gsm8k.py
@@ -0,0 +1,46 @@
+import dspy
+from dsp.utils import deduplicate
+from dspy.teleprompt import LabeledFewShot, BootstrapFewShotWithRandomSearch
+
+import datasets
+
+# pipeline configs
+turbo = dspy.OpenAI(model="gpt-3.5-turbo")
+dspy.settings.configure(lm=turbo)
+
+
+# load up the gsm8k dataset from the huggingface hub
+dataset = datasets.load_dataset("gsm8k")
+
+# split into train and test
+trainset = dataset["train"]
+testset = dataset["test"]
+
+print(trainset)
+
+# # each example should have a question and answer
+# trainset = []
+# testset = []
+
+# def gsm8k_accuracy(example, pred, trace=None):
+# if not dspy.evaluate.answer_exact_match(example, pred):
+# return False
+# return True
+
+# class ThoughtReflection(dspy.Module):
+# def __init__(self, num_attempts):
+# self.predict = dspy.ChainOfThought("question -> answer", n=num_attempts)
+# self.compare = dspy.MultiChainComparison("question -> answer", M=num_attempts)
+
+# def forward(self, question):
+# completions = self.predict(question=question).completions
+# answer = self.compare(question=question, completions=completions)
+# return answer
+
+
+# telemprompter = LabeledFewShot(k=8)
+# # teleprompter = BootstrapFewShotWithRandomSearch(metric=gsm8k_accuracy)
+
+
+# prog = ThoughtReflection(num_attempts=5)
+# compiled_prog = telemprompter.compile(prog, trainset=trainset)
\ No newline at end of file
diff --git a/assert_hotpotqa.py b/assert_hotpotqa.py
index 83b6d6597..54ad31e66 100644
--- a/assert_hotpotqa.py
+++ b/assert_hotpotqa.py
@@ -1,5 +1,3 @@
-import os
-
import dspy
from dsp.utils import deduplicate
from dspy.datasets import HotPotQA