Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,50 +1,47 @@
|
|
1 |
-
from haystack
|
2 |
-
from
|
3 |
-
from
|
4 |
-
from haystack.pipelines import Pipeline
|
5 |
|
6 |
-
from
|
7 |
-
from haystack.schema import Document
|
8 |
|
9 |
-
from
|
10 |
-
|
11 |
-
from huggingface_hub import get_inference_endpoint
|
12 |
from datasets import load_dataset
|
13 |
from time import perf_counter
|
14 |
import gradio as gr
|
15 |
-
import
|
16 |
import requests
|
17 |
import os
|
18 |
|
19 |
-
|
20 |
-
|
21 |
|
22 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
23 |
RANKER_URL = os.getenv("RANKER_URL")
|
24 |
-
|
25 |
|
26 |
-
|
27 |
-
"fastrag-
|
28 |
)
|
29 |
RANKER_IE = get_inference_endpoint(
|
30 |
"fastrag-ranker", namespace="optimum-intel", token=HF_TOKEN
|
31 |
)
|
32 |
|
|
|
33 |
def check_inference_endpoints():
|
34 |
-
|
35 |
RANKER_IE.update()
|
36 |
|
37 |
messages = []
|
38 |
|
39 |
-
if
|
40 |
messages += [
|
41 |
-
f"
|
42 |
]
|
43 |
-
elif
|
44 |
messages += [
|
45 |
-
f"
|
46 |
]
|
47 |
-
|
48 |
|
49 |
if RANKER_IE.status in ["initializing", "pending"]:
|
50 |
messages += [
|
@@ -62,7 +59,6 @@ def check_inference_endpoints():
|
|
62 |
return None
|
63 |
|
64 |
|
65 |
-
|
66 |
def post(url, payload):
|
67 |
response = requests.post(
|
68 |
url,
|
@@ -85,81 +81,47 @@ def method_timer(method):
|
|
85 |
return timed
|
86 |
|
87 |
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
scale_score: bool = True,
|
95 |
-
):
|
96 |
-
self.document_store = document_store
|
97 |
-
self.top_k = top_k
|
98 |
-
self.batch_size = batch_size
|
99 |
-
self.scale_score = scale_score
|
100 |
-
|
101 |
-
@method_timer
|
102 |
-
def embed_queries(self, queries: List[str]) -> np.ndarray:
|
103 |
-
payload = {"queries": queries, "inputs": ""}
|
104 |
-
response = post(RETRIEVER_URL, payload)
|
105 |
|
106 |
if "error" in response:
|
107 |
raise gr.Error(response["error"])
|
108 |
|
109 |
-
|
110 |
-
|
111 |
|
112 |
-
|
113 |
-
|
|
|
|
|
114 |
documents = [d.to_dict() for d in documents]
|
115 |
-
for doc in documents:
|
116 |
-
doc["embedding"] = None
|
117 |
|
118 |
payload = {"documents": documents, "inputs": ""}
|
119 |
-
response = post(
|
120 |
|
121 |
if "error" in response:
|
122 |
raise gr.Error(response["error"])
|
123 |
|
124 |
-
|
125 |
-
return arrays
|
126 |
|
127 |
|
128 |
-
|
129 |
-
|
130 |
-
def
|
131 |
-
self
|
132 |
-
) -> List[Document]:
|
133 |
-
documents = [d.to_dict() for d in documents]
|
134 |
-
for doc in documents:
|
135 |
-
doc["embedding"] = None
|
136 |
-
|
137 |
-
payload = {"query": query, "documents": documents, "top_k": top_k, "inputs": ""}
|
138 |
-
response = post(RANKER_URL, payload)
|
139 |
-
|
140 |
-
if "error" in response:
|
141 |
-
raise gr.Error(response["error"])
|
142 |
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
def predict_batch(
|
147 |
-
self,
|
148 |
-
queries: List[str],
|
149 |
-
documents: List[List[Document]],
|
150 |
-
batch_size: Optional[int] = None,
|
151 |
-
top_k: Optional[int] = None,
|
152 |
-
) -> List[List[Document]]:
|
153 |
-
documents = [[d.to_dict() for d in docs] for docs in documents]
|
154 |
-
for docs in documents:
|
155 |
-
for doc in docs:
|
156 |
-
doc["embedding"] = None
|
157 |
|
158 |
payload = {
|
159 |
-
"
|
160 |
"documents": documents,
|
161 |
-
"
|
162 |
-
"top_k": top_k,
|
163 |
"inputs": "",
|
164 |
}
|
165 |
response = post(RANKER_URL, payload)
|
@@ -167,49 +129,58 @@ class Ranker(BaseRanker):
|
|
167 |
if "error" in response:
|
168 |
raise gr.Error(response["error"])
|
169 |
|
170 |
-
return [
|
171 |
|
172 |
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
except FileNotFoundError:
|
192 |
-
pass
|
193 |
-
|
194 |
-
document_store = FAISSDocumentStore(
|
195 |
-
sql_url="sqlite:////data/faiss_document_store.db",
|
196 |
return_embedding=True,
|
|
|
197 |
embedding_dim=384,
|
198 |
)
|
199 |
-
|
200 |
-
|
201 |
-
)
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
|
210 |
pipe = Pipeline()
|
211 |
-
|
212 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
213 |
|
214 |
|
215 |
def run(query: str) -> dict:
|
@@ -221,11 +192,12 @@ def run(query: str) -> dict:
|
|
221 |
<p>{message}</p>
|
222 |
"""
|
223 |
|
224 |
-
pipe_output = pipe.run(query
|
225 |
|
226 |
-
output =
|
227 |
|
228 |
-
for i, doc in enumerate(pipe_output["documents"]):
|
|
|
229 |
output += f"""
|
230 |
<h3>Document {i + 1}</h3>
|
231 |
<p><strong>ID:</strong> {doc.id}</p>
|
@@ -251,6 +223,7 @@ input_text = gr.components.Textbox(
|
|
251 |
)
|
252 |
output_html = gr.components.HTML(label="Documents")
|
253 |
|
|
|
254 |
gr.Interface(
|
255 |
fn=run,
|
256 |
inputs=input_text,
|
@@ -259,9 +232,9 @@ gr.Interface(
|
|
259 |
cache_examples=False,
|
260 |
allow_flagging="never",
|
261 |
title="End-to-End Retrieval & Ranking with Hugging Face Inference Endpoints and Spaces",
|
262 |
-
description="""## A [haystack](https://haystack.deepset.ai/) pipeline with the following components
|
263 |
-
- <strong>Document Store</strong>: A [
|
264 |
-
- <strong>
|
265 |
- <strong>Ranker</strong>: [Quantized FastRAG Retriever](https://huggingface.co/optimum-intel/fastrag-ranker) deployed on [Inference Endpoints](https://huggingface.co/docs/inference-endpoints/index) + Intel Sapphire Rapids CPU.
|
266 |
|
267 |
This Space is based on the optimizations demonstrated in the blog [CPU Optimized Embeddings with π€ Optimum Intel and fastRAG](https://huggingface.co/blog/intel-fast-embedding)
|
|
|
1 |
+
from haystack import Document, Pipeline, component
|
2 |
+
from haystack_integrations.document_stores.qdrant import QdrantDocumentStore
|
3 |
+
from haystack_integrations.components.retrievers.qdrant import QdrantEmbeddingRetriever
|
|
|
4 |
|
5 |
+
from typing import List
|
|
|
6 |
|
7 |
+
from huggingface_hub import get_inference_endpoint, get_token
|
|
|
|
|
8 |
from datasets import load_dataset
|
9 |
from time import perf_counter
|
10 |
import gradio as gr
|
11 |
+
import shutil
|
12 |
import requests
|
13 |
import os
|
14 |
|
15 |
+
RETRIEVER_TOP_K = 5
|
16 |
+
RANKER_TOP_K = 2
|
17 |
|
18 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
19 |
RANKER_URL = os.getenv("RANKER_URL")
|
20 |
+
EMBEDDER_URL = os.getenv("EMBEDDER_URL")
|
21 |
|
22 |
+
EMBEDDER_IE = get_inference_endpoint(
|
23 |
+
"fastrag-embedder", namespace="optimum-intel", token=HF_TOKEN
|
24 |
)
|
25 |
RANKER_IE = get_inference_endpoint(
|
26 |
"fastrag-ranker", namespace="optimum-intel", token=HF_TOKEN
|
27 |
)
|
28 |
|
29 |
+
|
30 |
def check_inference_endpoints():
|
31 |
+
EMBEDDER_IE.update()
|
32 |
RANKER_IE.update()
|
33 |
|
34 |
messages = []
|
35 |
|
36 |
+
if EMBEDDER_IE.status in ["initializing", "pending"]:
|
37 |
messages += [
|
38 |
+
f"Embedder Inference Endpoint is {EMBEDDER_IE.status}. Please wait a few seconds and try again."
|
39 |
]
|
40 |
+
elif EMBEDDER_IE.status in ["paused", "scaledToZero"]:
|
41 |
messages += [
|
42 |
+
f"Embedder Inference Endpoint is {EMBEDDER_IE.status}. Resuming it. Please wait a few seconds and try again."
|
43 |
]
|
44 |
+
EMBEDDER_IE.resume()
|
45 |
|
46 |
if RANKER_IE.status in ["initializing", "pending"]:
|
47 |
messages += [
|
|
|
59 |
return None
|
60 |
|
61 |
|
|
|
62 |
def post(url, payload):
|
63 |
response = requests.post(
|
64 |
url,
|
|
|
81 |
return timed
|
82 |
|
83 |
|
84 |
+
@component
|
85 |
+
class InferenceEndpointTextEmbedder:
|
86 |
+
@component.output_types(embedding=List[float])
|
87 |
+
def run(self, text: str):
|
88 |
+
payload = {"text": text, "inputs": ""}
|
89 |
+
response = post(EMBEDDER_URL, payload)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
90 |
|
91 |
if "error" in response:
|
92 |
raise gr.Error(response["error"])
|
93 |
|
94 |
+
return {"embedding": response["embedding"]}
|
95 |
+
|
96 |
|
97 |
+
@component
|
98 |
+
class InferenceEndpointDocumentEmbedder:
|
99 |
+
@component.output_types(documents=List[Document])
|
100 |
+
def run(self, documents: List[Document]):
|
101 |
documents = [d.to_dict() for d in documents]
|
|
|
|
|
102 |
|
103 |
payload = {"documents": documents, "inputs": ""}
|
104 |
+
response = post(EMBEDDER_URL, payload)
|
105 |
|
106 |
if "error" in response:
|
107 |
raise gr.Error(response["error"])
|
108 |
|
109 |
+
return {"documents": [Document.from_dict(doc) for doc in response["documents"]]}
|
|
|
110 |
|
111 |
|
112 |
+
@component
|
113 |
+
class InferenceEndpointRanker:
|
114 |
+
def __init__(self, top_k: int):
|
115 |
+
self.top_k = top_k
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
116 |
|
117 |
+
@component.output_types(documents=List[Document])
|
118 |
+
def run(self, query: str, documents: List[Document]):
|
119 |
+
documents = [d.to_dict() for d in documents]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
120 |
|
121 |
payload = {
|
122 |
+
"query": query,
|
123 |
"documents": documents,
|
124 |
+
"top_k": self.top_k,
|
|
|
125 |
"inputs": "",
|
126 |
}
|
127 |
response = post(RANKER_URL, payload)
|
|
|
129 |
if "error" in response:
|
130 |
raise gr.Error(response["error"])
|
131 |
|
132 |
+
return {"documents": [Document.from_dict(doc) for doc in response["documents"]]}
|
133 |
|
134 |
|
135 |
+
document_store = None
|
136 |
+
|
137 |
+
if os.path.exists("data/qdrant"):
|
138 |
+
try:
|
139 |
+
document_store = QdrantDocumentStore(
|
140 |
+
path="./data/qdrant",
|
141 |
+
return_embedding=True,
|
142 |
+
recreate_index=False,
|
143 |
+
embedding_dim=384,
|
144 |
+
)
|
145 |
+
except Exception:
|
146 |
+
shutil.rmtree("data/qdrant", ignore_errors=True)
|
147 |
+
|
148 |
+
if document_store is None:
|
149 |
+
shutil.rmtree("data/qdrant", ignore_errors=True)
|
150 |
+
|
151 |
+
document_store = QdrantDocumentStore(
|
152 |
+
path="./data/qdrant",
|
|
|
|
|
|
|
|
|
|
|
153 |
return_embedding=True,
|
154 |
+
recreate_index=True,
|
155 |
embedding_dim=384,
|
156 |
)
|
157 |
+
dataset = load_dataset("bilgeyucel/seven-wonders")
|
158 |
+
documents = [Document(**doc) for doc in dataset["train"]]
|
159 |
+
documents_embedder = InferenceEndpointDocumentEmbedder()
|
160 |
+
documents_with_embedding = documents_embedder.run(documents)["documents"]
|
161 |
+
document_store.write_documents(documents_with_embedding)
|
162 |
+
|
163 |
+
print(
|
164 |
+
"Number of embedded documents in DocumentStore:",
|
165 |
+
document_store.count_documents(),
|
166 |
+
)
|
167 |
|
168 |
pipe = Pipeline()
|
169 |
+
|
170 |
+
embedder = InferenceEndpointTextEmbedder()
|
171 |
+
ranker = InferenceEndpointRanker(top_k=RANKER_TOP_K)
|
172 |
+
retriever = QdrantEmbeddingRetriever(
|
173 |
+
document_store=document_store, top_k=RETRIEVER_TOP_K
|
174 |
+
)
|
175 |
+
|
176 |
+
pipe.add_component("retriever", retriever)
|
177 |
+
pipe.add_component("embedder", embedder)
|
178 |
+
pipe.add_component("ranker", ranker)
|
179 |
+
|
180 |
+
pipe.connect("retriever", "ranker.documents")
|
181 |
+
pipe.connect("embedder", "retriever")
|
182 |
+
|
183 |
+
print(pipe)
|
184 |
|
185 |
|
186 |
def run(query: str) -> dict:
|
|
|
192 |
<p>{message}</p>
|
193 |
"""
|
194 |
|
195 |
+
pipe_output = pipe.run({"embedder": {"text": query}, "ranker": {"query": query}})
|
196 |
|
197 |
+
output = """<h2>Top Ranked Documents</h2>"""
|
198 |
|
199 |
+
for i, doc in enumerate(pipe_output["ranker"]["documents"]):
|
200 |
+
# limit content to 100 characters
|
201 |
output += f"""
|
202 |
<h3>Document {i + 1}</h3>
|
203 |
<p><strong>ID:</strong> {doc.id}</p>
|
|
|
223 |
)
|
224 |
output_html = gr.components.HTML(label="Documents")
|
225 |
|
226 |
+
|
227 |
gr.Interface(
|
228 |
fn=run,
|
229 |
inputs=input_text,
|
|
|
232 |
cache_examples=False,
|
233 |
allow_flagging="never",
|
234 |
title="End-to-End Retrieval & Ranking with Hugging Face Inference Endpoints and Spaces",
|
235 |
+
description="""## A [haystack](https://haystack.deepset.ai/) V2 pipeline with the following components
|
236 |
+
- <strong>Document Store</strong>: A [Qdrant document store](https://github.com/qdrant/qdrant) containing the [`seven-wonders` dataset](https://huggingface.co/datasets/bilgeyucel/seven-wonders), created on this Space's [persistent storage](https://huggingface.co/docs/hub/en/spaces-storage).
|
237 |
+
- <strong>Embedder</strong>: [Quantized FastRAG Embedder](https://huggingface.co/optimum-intel/fastrag-embedder) deployed on [Inference Endpoints](https://huggingface.co/docs/inference-endpoints/index) + Intel Sapphire Rapids CPU.
|
238 |
- <strong>Ranker</strong>: [Quantized FastRAG Retriever](https://huggingface.co/optimum-intel/fastrag-ranker) deployed on [Inference Endpoints](https://huggingface.co/docs/inference-endpoints/index) + Intel Sapphire Rapids CPU.
|
239 |
|
240 |
This Space is based on the optimizations demonstrated in the blog [CPU Optimized Embeddings with π€ Optimum Intel and fastRAG](https://huggingface.co/blog/intel-fast-embedding)
|