Skip to content

Commit 1c4f050

Browse files
okhatokhat
authored
Fix imports related to the removal of dsp/* (#1937)
* Fix dspy.dsp imports * Update setup.py to exclude dsp/
1 parent 65ba23a commit 1c4f050

File tree

7 files changed

+508
-508
lines changed

7 files changed

+508
-508
lines changed

‎dspy/__init__.py‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -18,8 +18,8 @@
1818

1919
configure_dspy_loggers(__name__)
2020

21-
from dspy.dsp.modules.colbertv2 import ColBERTv2
22-
# from dspy.dsp.modules.you import You
21+
from dspy.dsp.colbertv2 import ColBERTv2
22+
# from dspy.dsp.you import You
2323

2424
configure = settings.configure
2525
context = settings.context

‎dspy/dsp/modules/cache_utils.py‎ renamed to ‎dspy/dsp/cache_utils.py‎

File renamed without changes.

‎dspy/dsp/modules/colbertv2.py‎ renamed to ‎dspy/dsp/colbertv2.py‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -3,7 +3,7 @@
33

44
import requests
55

6-
from dspy.dsp.modules.cache_utils import CacheMemory, NotebookCacheMemory
6+
from dspy.dsp.cache_utils import CacheMemory, NotebookCacheMemory
77
from dspy.dsp.utils import dotdict
88

99
# TODO: Ideally, this takes the name of the index and looks up its port.

‎dspy/retrieve/faiss_rm.py‎

Lines changed: 156 additions & 156 deletions
Original file line numberDiff line numberDiff line change
@@ -1,156 +1,156 @@
1-
"""Retriever model for faiss: https://github.com/facebookresearch/faiss.
2-
Author: Jagane Sundar: https://github.com/jagane.
3-
"""
4-
5-
import logging
6-
from typing import Optional, Union
7-
8-
import numpy as np
9-
10-
import dspy
11-
from dspy.dsp.modules.sentence_vectorizer import SentenceTransformersVectorizer
12-
from dspy.dsp.utils import dotdict
13-
14-
try:
15-
import faiss
16-
except ImportError:
17-
faiss = None
18-
19-
if faiss is None:
20-
raise ImportError(
21-
"""
22-
The faiss package is required. Install it using `pip install dspy-ai[faiss-cpu]`
23-
""",
24-
)
25-
26-
27-
logger = logging.getLogger(__name__)
28-
class FaissRM(dspy.Retrieve):
29-
"""A retrieval module that uses an in-memory Faiss to return the top passages for a given query.
30-
31-
Args:
32-
document_chunks: the input text chunks
33-
vectorizer: an object that is a subclass of BaseSentenceVectorizer
34-
k (int, optional): The number of top passages to retrieve. Defaults to 3.
35-
36-
Returns:
37-
dspy.Prediction: An object containing the retrieved passages.
38-
39-
Examples:
40-
Below is a code snippet that shows how to use this as the default retriver:
41-
```python
42-
import dspy
43-
from dspy.retrieve import faiss_rm
44-
45-
document_chunks = [
46-
"The superbowl this year was played between the San Francisco 49ers and the Kanasas City Chiefs",
47-
"Pop corn is often served in a bowl",
48-
"The Rice Bowl is a Chinese Restaurant located in the city of Tucson, Arizona",
49-
"Mars is the fourth planet in the Solar System",
50-
"An aquarium is a place where children can learn about marine life",
51-
"The capital of the United States is Washington, D.C",
52-
"Rock and Roll musicians are honored by being inducted in the Rock and Roll Hall of Fame",
53-
"Music albums were published on Long Play Records in the 70s and 80s",
54-
"Sichuan cuisine is a spicy cuisine from central China",
55-
"The interest rates for mortgages is considered to be very high in 2024",
56-
]
57-
58-
frm = faiss_rm.FaissRM(document_chunks)
59-
turbo = dspy.OpenAI(model="gpt-3.5-turbo")
60-
dspy.settings.configure(lm=turbo, rm=frm)
61-
print(frm(["I am in the mood for Chinese food"]))
62-
```
63-
64-
Below is a code snippet that shows how to use this in the forward() function of a module
65-
```python
66-
self.retrieve = FaissRM(k=num_passages)
67-
```
68-
"""
69-
70-
def __init__(self, document_chunks, vectorizer=None, k: int = 3):
71-
"""Inits the faiss retriever.
72-
73-
Args:
74-
document_chunks: a list of input strings.
75-
vectorizer: an object that is a subclass of BaseTransformersVectorizer.
76-
k: number of matches to return.
77-
"""
78-
if vectorizer:
79-
self._vectorizer = vectorizer
80-
else:
81-
self._vectorizer = SentenceTransformersVectorizer()
82-
embeddings = self._vectorizer(document_chunks)
83-
xb = np.array(embeddings)
84-
d = len(xb[0])
85-
logger.info(f"FaissRM: embedding size={d}")
86-
if len(xb) < 100:
87-
self._faiss_index = faiss.IndexFlatL2(d)
88-
self._faiss_index.add(xb)
89-
else:
90-
# if we have at least 100 vectors, we use Voronoi cells
91-
nlist = 100
92-
quantizer = faiss.IndexFlatL2(d)
93-
self._faiss_index = faiss.IndexIVFFlat(quantizer, d, nlist)
94-
self._faiss_index.train(xb)
95-
self._faiss_index.add(xb)
96-
97-
logger.info(f"{self._faiss_index.ntotal} vectors in faiss index")
98-
self._document_chunks = document_chunks # save the input document chunks
99-
100-
super().__init__(k=k)
101-
102-
def _dump_raw_results(self, queries, index_list, distance_list) -> None:
103-
for i in range(len(queries)):
104-
indices = index_list[i]
105-
distances = distance_list[i]
106-
logger.debug(f"Query: {queries[i]}")
107-
for j in range(len(indices)):
108-
logger.debug(f" Hit {j} = {indices[j]}/{distances[j]}: {self._document_chunks[indices[j]]}")
109-
return
110-
111-
def forward(self, query_or_queries: Union[str, list[str]], k: Optional[int] = None, **kwargs) -> dspy.Prediction:
112-
"""Search the faiss index for k or self.k top passages for query.
113-
114-
Args:
115-
query_or_queries (Union[str, List[str]]): The query or queries to search for.
116-
117-
Returns:
118-
dspy.Prediction: An object containing the retrieved passages.
119-
"""
120-
queries = [query_or_queries] if isinstance(query_or_queries, str) else query_or_queries
121-
queries = [q for q in queries if q] # Filter empty queries
122-
embeddings = self._vectorizer(queries)
123-
emb_npa = np.array(embeddings)
124-
# For single query, just look up the top k passages
125-
if len(queries) == 1:
126-
distance_list, index_list = self._faiss_index.search(emb_npa, k or self.k)
127-
# self._dump_raw_results(queries, index_list, distance_list)
128-
passages = [(self._document_chunks[ind], ind) for ind in index_list[0]]
129-
return [dotdict({"long_text": passage[0], "index": passage[1]}) for passage in passages]
130-
131-
distance_list, index_list = self._faiss_index.search(emb_npa, (k or self.k) * 3, **kwargs)
132-
# self._dump_raw_results(queries, index_list, distance_list)
133-
passage_scores = {}
134-
for emb in range(len(embeddings)):
135-
indices = index_list[emb] # indices of neighbors for embeddings[emb] - this is an array of k*3 integers
136-
distances = distance_list[
137-
emb
138-
] # distances of neighbors for embeddings[emb] - this is an array of k*3 floating point numbers
139-
for res in range((k or self.k) * 3):
140-
neighbor = indices[res]
141-
distance = distances[res]
142-
if neighbor in passage_scores:
143-
passage_scores[neighbor].append(distance)
144-
else:
145-
passage_scores[neighbor] = [distance]
146-
# Note re. sorting:
147-
# first degree sort: number of queries that got a hit with any particular document chunk. More
148-
# is a better match. This is len(queries)-len(x[1])
149-
# second degree sort: sum of the distances of each hit returned by faiss. Smaller distance is a better match
150-
sorted_passages = sorted(passage_scores.items(), key=lambda x: (len(queries) - len(x[1]), sum(x[1])))[
151-
: k or self.k
152-
]
153-
return [
154-
dotdict({"long_text": self._document_chunks[passage_index], "index": passage_index})
155-
for passage_index, _ in sorted_passages
156-
]
1+
# """Retriever model for faiss: https://github.com/facebookresearch/faiss.
2+
# Author: Jagane Sundar: https://github.com/jagane.
3+
# """
4+
5+
# import logging
6+
# from typing import Optional, Union
7+
8+
# import numpy as np
9+
10+
# import dspy
11+
# from dspy.dsp.modules.sentence_vectorizer import SentenceTransformersVectorizer
12+
# from dspy.dsp.utils import dotdict
13+
14+
# try:
15+
# import faiss
16+
# except ImportError:
17+
# faiss = None
18+
19+
# if faiss is None:
20+
# raise ImportError(
21+
# """
22+
# The faiss package is required. Install it using `pip install dspy-ai[faiss-cpu]`
23+
# """,
24+
# )
25+
26+
27+
# logger = logging.getLogger(__name__)
28+
# class FaissRM(dspy.Retrieve):
29+
# """A retrieval module that uses an in-memory Faiss to return the top passages for a given query.
30+
31+
# Args:
32+
# document_chunks: the input text chunks
33+
# vectorizer: an object that is a subclass of BaseSentenceVectorizer
34+
# k (int, optional): The number of top passages to retrieve. Defaults to 3.
35+
36+
# Returns:
37+
# dspy.Prediction: An object containing the retrieved passages.
38+
39+
# Examples:
40+
# Below is a code snippet that shows how to use this as the default retriver:
41+
# ```python
42+
# import dspy
43+
# from dspy.retrieve import faiss_rm
44+
45+
# document_chunks = [
46+
# "The superbowl this year was played between the San Francisco 49ers and the Kanasas City Chiefs",
47+
# "Pop corn is often served in a bowl",
48+
# "The Rice Bowl is a Chinese Restaurant located in the city of Tucson, Arizona",
49+
# "Mars is the fourth planet in the Solar System",
50+
# "An aquarium is a place where children can learn about marine life",
51+
# "The capital of the United States is Washington, D.C",
52+
# "Rock and Roll musicians are honored by being inducted in the Rock and Roll Hall of Fame",
53+
# "Music albums were published on Long Play Records in the 70s and 80s",
54+
# "Sichuan cuisine is a spicy cuisine from central China",
55+
# "The interest rates for mortgages is considered to be very high in 2024",
56+
# ]
57+
58+
# frm = faiss_rm.FaissRM(document_chunks)
59+
# turbo = dspy.OpenAI(model="gpt-3.5-turbo")
60+
# dspy.settings.configure(lm=turbo, rm=frm)
61+
# print(frm(["I am in the mood for Chinese food"]))
62+
# ```
63+
64+
# Below is a code snippet that shows how to use this in the forward() function of a module
65+
# ```python
66+
# self.retrieve = FaissRM(k=num_passages)
67+
# ```
68+
# """
69+
70+
# def __init__(self, document_chunks, vectorizer=None, k: int = 3):
71+
# """Inits the faiss retriever.
72+
73+
# Args:
74+
# document_chunks: a list of input strings.
75+
# vectorizer: an object that is a subclass of BaseTransformersVectorizer.
76+
# k: number of matches to return.
77+
# """
78+
# if vectorizer:
79+
# self._vectorizer = vectorizer
80+
# else:
81+
# self._vectorizer = SentenceTransformersVectorizer()
82+
# embeddings = self._vectorizer(document_chunks)
83+
# xb = np.array(embeddings)
84+
# d = len(xb[0])
85+
# logger.info(f"FaissRM: embedding size={d}")
86+
# if len(xb) < 100:
87+
# self._faiss_index = faiss.IndexFlatL2(d)
88+
# self._faiss_index.add(xb)
89+
# else:
90+
# # if we have at least 100 vectors, we use Voronoi cells
91+
# nlist = 100
92+
# quantizer = faiss.IndexFlatL2(d)
93+
# self._faiss_index = faiss.IndexIVFFlat(quantizer, d, nlist)
94+
# self._faiss_index.train(xb)
95+
# self._faiss_index.add(xb)
96+
97+
# logger.info(f"{self._faiss_index.ntotal} vectors in faiss index")
98+
# self._document_chunks = document_chunks # save the input document chunks
99+
100+
# super().__init__(k=k)
101+
102+
# def _dump_raw_results(self, queries, index_list, distance_list) -> None:
103+
# for i in range(len(queries)):
104+
# indices = index_list[i]
105+
# distances = distance_list[i]
106+
# logger.debug(f"Query: {queries[i]}")
107+
# for j in range(len(indices)):
108+
# logger.debug(f" Hit {j} = {indices[j]}/{distances[j]}: {self._document_chunks[indices[j]]}")
109+
# return
110+
111+
# def forward(self, query_or_queries: Union[str, list[str]], k: Optional[int] = None, **kwargs) -> dspy.Prediction:
112+
# """Search the faiss index for k or self.k top passages for query.
113+
114+
# Args:
115+
# query_or_queries (Union[str, List[str]]): The query or queries to search for.
116+
117+
# Returns:
118+
# dspy.Prediction: An object containing the retrieved passages.
119+
# """
120+
# queries = [query_or_queries] if isinstance(query_or_queries, str) else query_or_queries
121+
# queries = [q for q in queries if q] # Filter empty queries
122+
# embeddings = self._vectorizer(queries)
123+
# emb_npa = np.array(embeddings)
124+
# # For single query, just look up the top k passages
125+
# if len(queries) == 1:
126+
# distance_list, index_list = self._faiss_index.search(emb_npa, k or self.k)
127+
# # self._dump_raw_results(queries, index_list, distance_list)
128+
# passages = [(self._document_chunks[ind], ind) for ind in index_list[0]]
129+
# return [dotdict({"long_text": passage[0], "index": passage[1]}) for passage in passages]
130+
131+
# distance_list, index_list = self._faiss_index.search(emb_npa, (k or self.k) * 3, **kwargs)
132+
# # self._dump_raw_results(queries, index_list, distance_list)
133+
# passage_scores = {}
134+
# for emb in range(len(embeddings)):
135+
# indices = index_list[emb] # indices of neighbors for embeddings[emb] - this is an array of k*3 integers
136+
# distances = distance_list[
137+
# emb
138+
# ] # distances of neighbors for embeddings[emb] - this is an array of k*3 floating point numbers
139+
# for res in range((k or self.k) * 3):
140+
# neighbor = indices[res]
141+
# distance = distances[res]
142+
# if neighbor in passage_scores:
143+
# passage_scores[neighbor].append(distance)
144+
# else:
145+
# passage_scores[neighbor] = [distance]
146+
# # Note re. sorting:
147+
# # first degree sort: number of queries that got a hit with any particular document chunk. More
148+
# # is a better match. This is len(queries)-len(x[1])
149+
# # second degree sort: sum of the distances of each hit returned by faiss. Smaller distance is a better match
150+
# sorted_passages = sorted(passage_scores.items(), key=lambda x: (len(queries) - len(x[1]), sum(x[1])))[
151+
# : k or self.k
152+
# ]
153+
# return [
154+
# dotdict({"long_text": self._document_chunks[passage_index], "index": passage_index})
155+
# for passage_index, _ in sorted_passages
156+
# ]

0 commit comments

Comments
 (0)