Spaces:
Running
Running
Upload folder using huggingface_hub
Browse files
app.py
CHANGED
@@ -19,7 +19,7 @@ file_example = """Please upload a JSON file with a "text" field (with optional "
|
|
19 |
{"title": "Title B", "text": "This an example text with the title"},
|
20 |
]
|
21 |
```
|
22 |
-
Due to the computation resources, please test with small scale data.
|
23 |
"""
|
24 |
|
25 |
|
@@ -42,6 +42,12 @@ def upload_file_fn(
|
|
42 |
try:
|
43 |
with open(file_path) as f:
|
44 |
document_data = json.load(f)
|
|
|
|
|
|
|
|
|
|
|
|
|
45 |
documents = []
|
46 |
for obj in document_data:
|
47 |
text = obj["title"] + "\n" + obj["text"] if obj.get("title") else obj["text"]
|
@@ -55,18 +61,13 @@ def upload_file_fn(
|
|
55 |
gr.Error(str(e))
|
56 |
return None, gr.update(interactive=False)
|
57 |
|
58 |
-
if len(documents) <
|
59 |
-
gr.Error("Please upload more than
|
60 |
return None, gr.update(interactive=False)
|
61 |
-
|
62 |
-
gr.Info(f"Upload {len(documents)} documents.")
|
63 |
-
if len(documents) > 1000:
|
64 |
-
gr.Info(f"Cut uploaded documents to 1000 due to the computation resource.")
|
65 |
-
documents = documents[: 1000]
|
66 |
|
67 |
# documents_embeddings = model.encode(documents, show_progress_bar=True)
|
68 |
documents_embeddings = []
|
69 |
-
batch_size =
|
70 |
for i in tqdm(range(0, len(documents), batch_size)):
|
71 |
batch_documents = documents[i: i+batch_size]
|
72 |
batch_embeddings = model.encode(batch_documents, show_progress_bar=True)
|
@@ -87,7 +88,7 @@ def clear_file_fn():
|
|
87 |
|
88 |
|
89 |
def retrieve_document_fn(question, document_states, instruct):
|
90 |
-
num_retrieval_doc =
|
91 |
|
92 |
if document_states is None:
|
93 |
gr.Warning("Please upload documents first!")
|
@@ -95,11 +96,16 @@ def retrieve_document_fn(question, document_states, instruct):
|
|
95 |
|
96 |
document_data, document_index = document_states["document_data"], document_states["document_index"]
|
97 |
|
98 |
-
|
|
|
|
|
|
|
|
|
|
|
99 |
batch_scores, batch_inxs = document_index.search(question_embedding, k=min(len(document_data), 150))
|
100 |
|
101 |
answers = [document_data[i]["text"] for i in batch_inxs[0][:num_retrieval_doc]]
|
102 |
-
return answers[0], answers[1], answers[2], document_states
|
103 |
|
104 |
|
105 |
def main(args):
|
@@ -126,9 +132,10 @@ def main(args):
|
|
126 |
retrieval_interface = gr.Interface(
|
127 |
fn=retrieve_document_fn,
|
128 |
inputs=[gr.Textbox(label="Query"), document_state],
|
129 |
-
outputs=[gr.Text(label="Recall-1"), gr.Text(label="Recall-2"), gr.Text(label="Recall-3"), gr.State()],
|
130 |
additional_inputs=[gr.Textbox("Instruct: Given a query, retrieve documents that answer the query. \n Query: ", label="Instruct of Query", lines=2)],
|
131 |
concurrency_limit=1,
|
|
|
132 |
)
|
133 |
# retrieval_interface.input_components[0] = gr.update(interactive=False)
|
134 |
|
@@ -153,7 +160,7 @@ def main(args):
|
|
153 |
if __name__ == "__main__":
|
154 |
parser = argparse.ArgumentParser()
|
155 |
parser.add_argument("--model_name_or_path", type=str, default="HIT-TMG/KaLM-embedding-multilingual-mini-instruct-v1.5")
|
156 |
-
parser.add_argument("--revision", type=str, default=
|
157 |
|
158 |
args = parser.parse_args()
|
159 |
main(args)
|
|
|
19 |
{"title": "Title B", "text": "This an example text with the title"},
|
20 |
]
|
21 |
```
|
22 |
+
Due to the computation resources, please test with small scale data (<1000).
|
23 |
"""
|
24 |
|
25 |
|
|
|
42 |
try:
|
43 |
with open(file_path) as f:
|
44 |
document_data = json.load(f)
|
45 |
+
|
46 |
+
gr.Info(f"Upload {len(document_data)} documents.")
|
47 |
+
if len(document_data) > 1000:
|
48 |
+
gr.Info(f"Cut uploaded documents to 1000 due to the computation resource.")
|
49 |
+
document_data = document_data[: 1000]
|
50 |
+
|
51 |
documents = []
|
52 |
for obj in document_data:
|
53 |
text = obj["title"] + "\n" + obj["text"] if obj.get("title") else obj["text"]
|
|
|
61 |
gr.Error(str(e))
|
62 |
return None, gr.update(interactive=False)
|
63 |
|
64 |
+
if len(documents) < 5:
|
65 |
+
gr.Error("Please upload more than 53 documents.")
|
66 |
return None, gr.update(interactive=False)
|
|
|
|
|
|
|
|
|
|
|
67 |
|
68 |
# documents_embeddings = model.encode(documents, show_progress_bar=True)
|
69 |
documents_embeddings = []
|
70 |
+
batch_size = 16
|
71 |
for i in tqdm(range(0, len(documents), batch_size)):
|
72 |
batch_documents = documents[i: i+batch_size]
|
73 |
batch_embeddings = model.encode(batch_documents, show_progress_bar=True)
|
|
|
88 |
|
89 |
|
90 |
def retrieve_document_fn(question, document_states, instruct):
|
91 |
+
num_retrieval_doc = 5
|
92 |
|
93 |
if document_states is None:
|
94 |
gr.Warning("Please upload documents first!")
|
|
|
96 |
|
97 |
document_data, document_index = document_states["document_data"], document_states["document_index"]
|
98 |
|
99 |
+
question_with_inst = str(instruct) + str(question)
|
100 |
+
if len(question_with_inst.strip()) == 0:
|
101 |
+
gr.Warning("Please enter a non-empty query.")
|
102 |
+
return None, None, None, None, None, document_states
|
103 |
+
|
104 |
+
question_embedding = model.encode([question_with_inst])
|
105 |
batch_scores, batch_inxs = document_index.search(question_embedding, k=min(len(document_data), 150))
|
106 |
|
107 |
answers = [document_data[i]["text"] for i in batch_inxs[0][:num_retrieval_doc]]
|
108 |
+
return answers[0], answers[1], answers[2], answers[3], answers[4],document_states
|
109 |
|
110 |
|
111 |
def main(args):
|
|
|
132 |
retrieval_interface = gr.Interface(
|
133 |
fn=retrieve_document_fn,
|
134 |
inputs=[gr.Textbox(label="Query"), document_state],
|
135 |
+
outputs=[gr.Text(label="Recall-1"), gr.Text(label="Recall-2"), gr.Text(label="Recall-3"), gr.Text(label="Recall-4"), gr.Text(label="Recall-5"), gr.State()],
|
136 |
additional_inputs=[gr.Textbox("Instruct: Given a query, retrieve documents that answer the query. \n Query: ", label="Instruct of Query", lines=2)],
|
137 |
concurrency_limit=1,
|
138 |
+
allow_flagging="never",
|
139 |
)
|
140 |
# retrieval_interface.input_components[0] = gr.update(interactive=False)
|
141 |
|
|
|
160 |
if __name__ == "__main__":
|
161 |
parser = argparse.ArgumentParser()
|
162 |
parser.add_argument("--model_name_or_path", type=str, default="HIT-TMG/KaLM-embedding-multilingual-mini-instruct-v1.5")
|
163 |
+
parser.add_argument("--revision", type=str, default=None)
|
164 |
|
165 |
args = parser.parse_args()
|
166 |
main(args)
|