DSPy: The framework for programming—not prompting—language models
Python
Make sure OpenAI and OpenAI sentence vectorizer work with 0.28 and 1.0
Omar Khattabc4e5aef74f65
diff --git a/dsp/modules/gpt3.py b/dsp/modules/gpt3.pyindex 46315d6b9..487d3b4f3 100644--- a/dsp/modules/gpt3.py+++ b/dsp/modules/gpt3.py@@ -1,6 +1,6 @@import functoolsimport json-from typing import Any, Literal, Optional+from typing import Any, Literal, Optional, castimport backoffimport openai@@ -8,6 +8,19 @@ import openaifrom dsp.modules.cache_utils import CacheMemory, NotebookCacheMemory, cache_turn_onfrom dsp.modules.lm import LM+try:+ OPENAI_LEGACY = int(openai.version.__version__[0]) == 0+except Exception:+ OPENAI_LEGACY = True++try:+ from openai.openai_object import OpenAIObject+ import openai.error+ ERRORS = (openai.error.RateLimitError, openai.error.ServiceUnavailableError, openai.error.APIError)+except Exception:+ ERRORS = (openai.RateLimitError, openai.APIError)+ OpenAIObject = dict+def backoff_hdlr(details):"""Handler from https://pypi.org/project/backoff/"""@@ -90,11 +103,11 @@ class GPT3(LM):# caching mechanism requires hashable kwargskwargs["messages"] = [{"role": "user", "content": prompt}]kwargs = {"stringify_request": json.dumps(kwargs)}- response = cached_gpt3_turbo_request(**kwargs).model_dump()+ response = chat_request(**kwargs)else:kwargs["prompt"] = prompt- response = cached_gpt3_request(**kwargs).model_dump()+ response = completions_request(**kwargs)history = {"prompt": prompt,@@ -108,7 +121,7 @@ class GPT3(LM):@backoff.on_exception(backoff.expo,- (openai.RateLimitError, openai.APIError),+ ERRORS,max_time=1000,on_backoff=backoff_hdlr,)@@ -183,32 +196,57 @@ class GPT3(LM):return completions+@CacheMemory.cachedef cached_gpt3_request_v2(**kwargs):- return openai.completions.create(**kwargs)-+ return openai.Completion.create(**kwargs)@functools.lru_cache(maxsize=None if cache_turn_on else 0)@NotebookCacheMemory.cachedef cached_gpt3_request_v2_wrapped(**kwargs):return cached_gpt3_request_v2(**kwargs)+def _cached_gpt3_turbo_request_v2(**kwargs) -> OpenAIObject:+ if "stringify_request" in kwargs:+ kwargs = json.loads(kwargs["stringify_request"])+ return cast(OpenAIObject, openai.ChatCompletion.create(**kwargs))-cached_gpt3_request = cached_gpt3_request_v2_wrapped[email protected]_cache(maxsize=None if cache_turn_on else 0)+def _cached_gpt3_turbo_request_v2_wrapped(**kwargs) -> OpenAIObject:+ return _cached_gpt3_turbo_request_v2(**kwargs)+def v1_cached_gpt3_request_v2(**kwargs):+ return openai.completions.create(**kwargs)+[email protected]_cache(maxsize=None if cache_turn_on else 0)+def v1_cached_gpt3_request_v2_wrapped(**kwargs):+ return v1_cached_gpt3_request_v2(**kwargs)@CacheMemory.cache-def _cached_gpt3_turbo_request_v2(**kwargs):+def v1_cached_gpt3_turbo_request_v2(**kwargs):if "stringify_request" in kwargs:kwargs = json.loads(kwargs["stringify_request"])-return openai.chat.completions.create(**kwargs)-@functools.lru_cache(maxsize=None if cache_turn_on else 0)@NotebookCacheMemory.cache-def _cached_gpt3_turbo_request_v2_wrapped(**kwargs):- return _cached_gpt3_turbo_request_v2(**kwargs)+def v1_cached_gpt3_turbo_request_v2_wrapped(**kwargs):+ return v1_cached_gpt3_turbo_request_v2(**kwargs)++++def chat_request(**kwargs):+ if OPENAI_LEGACY:+ return _cached_gpt3_turbo_request_v2_wrapped(**kwargs)++ return v1_cached_gpt3_turbo_request_v2_wrapped(**kwargs).model_dump()+def completions_request(**kwargs):+ if OPENAI_LEGACY:+ return cached_gpt3_request_v2_wrapped(**kwargs)-cached_gpt3_turbo_request = _cached_gpt3_turbo_request_v2_wrapped+ return v1_cached_gpt3_request_v2_wrapped(**kwargs).model_dump()diff --git a/dsp/modules/sentence_vectorizer.py b/dsp/modules/sentence_vectorizer.pyindex 2be913074..5666efe22 100644--- a/dsp/modules/sentence_vectorizer.py+++ b/dsp/modules/sentence_vectorizer.py@@ -2,9 +2,7 @@ import abcfrom typing import List, Optionalimport numpy as np-from openai import OpenAI--+import openaiclass BaseSentenceVectorizer(abc.ABC):@@ -111,6 +109,12 @@ class NaiveGetFieldVectorizer(BaseSentenceVectorizer):return embeddings+try:+ OPENAI_LEGACY = int(openai.version.__version__[0]) == 0+except Exception:+ OPENAI_LEGACY = True++class OpenAIVectorizer(BaseSentenceVectorizer):'''This vectorizer uses OpenAI API to convert texts to embeddings. Changing `model` is not@@ -125,10 +129,14 @@ class OpenAIVectorizer(BaseSentenceVectorizer):):self.model = modelself.embed_batch_size = embed_batch_size- if api_key:- self.client = OpenAI(api_key=api_key)++ if OPENAI_LEGACY:+ self.Embedding = openai.Embeddingelse:- self.client = OpenAI()+ self.Embedding = openai.embeddings++ if api_key:+ openai.api_key = api_keydef __call__(self, inp_examples: List["Example"]) -> np.ndarray:text_to_vectorize = self._extract_text_from_examples(inp_examples)@@ -141,10 +149,13 @@ class OpenAIVectorizer(BaseSentenceVectorizer):end_idx = (cur_batch_idx + 1) * self.embed_batch_sizecur_batch = text_to_vectorize[start_idx: end_idx]# OpenAI API call:- response = self.client.embeddings.create(model=self.model, input=cur_batch)+ response = self.Embedding.create(+ model=self.model,+ input=cur_batch+ )- cur_batch_embeddings = [cur_obj.embedding for cur_obj in response.data]+ cur_batch_embeddings = [cur_obj['embedding'] for cur_obj in response['data']]embeddings_list.extend(cur_batch_embeddings)embeddings = np.array(embeddings_list, dtype=np.float32)- return embeddings+ return embeddings\ No newline at end of file