Spaces:
Running
Running
jbdel
commited on
Commit
•
a74e67c
1
Parent(s):
aa8c7c6
gemini
Browse files- app.py +8 -8
- paper_chat_tab.py +73 -23
- requirements.txt +1 -0
app.py
CHANGED
@@ -184,14 +184,14 @@ with gr.Blocks(css_paths="style.css") as demo:
|
|
184 |
wrap=True,
|
185 |
)
|
186 |
|
187 |
-
with gr.Tab("Edit papers", id="tab-pr"):
|
188 |
-
|
189 |
-
|
190 |
-
with gr.Tab("Leaderboards", id="tab-leaderboards"):
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
|
196 |
|
197 |
# chat with paper
|
|
|
184 |
wrap=True,
|
185 |
)
|
186 |
|
187 |
+
# with gr.Tab("Edit papers", id="tab-pr"):
|
188 |
+
# pr_paper_central_tab(paper_central_df.df_raw)
|
189 |
+
#
|
190 |
+
# with gr.Tab("Leaderboards", id="tab-leaderboards"):
|
191 |
+
# with gr.Tab("Authors"):
|
192 |
+
# author_leaderboard_tab()
|
193 |
+
# with gr.Tab("Contributors"):
|
194 |
+
# author_resource_leaderboard_tab()
|
195 |
|
196 |
|
197 |
# chat with paper
|
paper_chat_tab.py
CHANGED
@@ -1,3 +1,5 @@
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
from PyPDF2 import PdfReader
|
3 |
from bs4 import BeautifulSoup
|
@@ -17,6 +19,19 @@ tokenizer_cache = {}
|
|
17 |
|
18 |
# Global variables for providers
|
19 |
PROVIDERS = {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
"SambaNova": {
|
21 |
"name": "SambaNova",
|
22 |
"logo": "https://venturebeat.com/wp-content/uploads/2020/02/SambaNovaLogo_H_F.jpg",
|
@@ -24,6 +39,7 @@ PROVIDERS = {
|
|
24 |
"api_key_env_var": "SAMBANOVA_API_KEY",
|
25 |
"models": [
|
26 |
"Meta-Llama-3.1-70B-Instruct",
|
|
|
27 |
],
|
28 |
"type": "tuples",
|
29 |
"max_total_tokens": "50000",
|
@@ -191,6 +207,7 @@ def create_chat_interface(provider_dropdown, model_dropdown, paper_content, hf_t
|
|
191 |
message_tokens_list = []
|
192 |
total_tokens = context_token_length
|
193 |
|
|
|
194 |
for user_msg, assistant_msg in history:
|
195 |
user_tokens = tokenizer.encode(user_msg)
|
196 |
messages.append({"role": "user", "content": user_msg})
|
@@ -208,6 +225,7 @@ def create_chat_interface(provider_dropdown, model_dropdown, paper_content, hf_t
|
|
208 |
message_tokens_list.append(len(message_tokens))
|
209 |
total_tokens += len(message_tokens)
|
210 |
|
|
|
211 |
if total_tokens > max_total_tokens:
|
212 |
available_tokens = max_total_tokens - (total_tokens - context_token_length)
|
213 |
if available_tokens > 0:
|
@@ -227,35 +245,67 @@ def create_chat_interface(provider_dropdown, model_dropdown, paper_content, hf_t
|
|
227 |
|
228 |
final_messages = []
|
229 |
if context:
|
230 |
-
final_messages.append(
|
|
|
231 |
final_messages.extend(messages)
|
232 |
|
233 |
api_key = hf_token_value or os.environ.get(api_key_env_var)
|
234 |
if not api_key:
|
235 |
raise ValueError("API token is not provided.")
|
236 |
|
237 |
-
|
238 |
-
|
239 |
-
|
240 |
-
|
241 |
-
|
242 |
-
|
243 |
-
|
244 |
-
|
245 |
-
|
246 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
247 |
)
|
248 |
-
|
249 |
-
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
259 |
|
260 |
chatbot = gr.Chatbot(label="Chatbot", scale=1, height=800, autoscroll=True)
|
261 |
chat_interface = gr.ChatInterface(
|
@@ -299,7 +349,7 @@ def paper_chat_tab(paper_id, paper_from, paper_central_df):
|
|
299 |
|
300 |
# Add a textbox for user to enter a paper_id (arxiv_id)
|
301 |
paper_id_input = gr.Textbox(
|
302 |
-
label="Or enter a 🤗 paper_id
|
303 |
placeholder="e.g. 1234.56789"
|
304 |
)
|
305 |
|
|
|
1 |
+
import google.generativeai as genai
|
2 |
+
|
3 |
import gradio as gr
|
4 |
from PyPDF2 import PdfReader
|
5 |
from bs4 import BeautifulSoup
|
|
|
19 |
|
20 |
# Global variables for providers
|
21 |
PROVIDERS = {
|
22 |
+
"Gemini": {
|
23 |
+
"name": "Gemini",
|
24 |
+
"logo": "https://www.gstatic.com/lamda/images/gemini_thumbnail_c362e5eadc46ca9f617e2.png",
|
25 |
+
"endpoint": "https://example-gemini-endpoint", # not need
|
26 |
+
# Not necessarily needed for Gemini since we use google.generativeai directly
|
27 |
+
"api_key_env_var": "GEMINI_API_KEY", # If using env vars for key storage
|
28 |
+
"models": [
|
29 |
+
"gemini-2.0-flash-exp",
|
30 |
+
"gemini-1.5-flash",
|
31 |
+
],
|
32 |
+
"type": "tuples",
|
33 |
+
"max_total_tokens": "50000",
|
34 |
+
},
|
35 |
"SambaNova": {
|
36 |
"name": "SambaNova",
|
37 |
"logo": "https://venturebeat.com/wp-content/uploads/2020/02/SambaNovaLogo_H_F.jpg",
|
|
|
39 |
"api_key_env_var": "SAMBANOVA_API_KEY",
|
40 |
"models": [
|
41 |
"Meta-Llama-3.1-70B-Instruct",
|
42 |
+
"Meta-Llama-3.3-70B-Instruct",
|
43 |
],
|
44 |
"type": "tuples",
|
45 |
"max_total_tokens": "50000",
|
|
|
207 |
message_tokens_list = []
|
208 |
total_tokens = context_token_length
|
209 |
|
210 |
+
# Reconstruct the conversation from history and current user message
|
211 |
for user_msg, assistant_msg in history:
|
212 |
user_tokens = tokenizer.encode(user_msg)
|
213 |
messages.append({"role": "user", "content": user_msg})
|
|
|
225 |
message_tokens_list.append(len(message_tokens))
|
226 |
total_tokens += len(message_tokens)
|
227 |
|
228 |
+
# Token truncation logic
|
229 |
if total_tokens > max_total_tokens:
|
230 |
available_tokens = max_total_tokens - (total_tokens - context_token_length)
|
231 |
if available_tokens > 0:
|
|
|
245 |
|
246 |
final_messages = []
|
247 |
if context:
|
248 |
+
final_messages.append(
|
249 |
+
{"role": "system" if not provider_name_value == "Gemini" else "user", "content": f"{context}"})
|
250 |
final_messages.extend(messages)
|
251 |
|
252 |
api_key = hf_token_value or os.environ.get(api_key_env_var)
|
253 |
if not api_key:
|
254 |
raise ValueError("API token is not provided.")
|
255 |
|
256 |
+
# Gemini logic
|
257 |
+
if provider_name_value == "Gemini":
|
258 |
+
import google.generativeai as genai
|
259 |
+
genai.configure(api_key=api_key)
|
260 |
+
|
261 |
+
# According to the docs, model should be instantiated with full model name, e.g. "models/gemini-1.5-flash"
|
262 |
+
# Ensure your PROVIDERS dict sets the model_name_value accordingly (e.g. "models/gemini-1.5-flash")
|
263 |
+
model = genai.GenerativeModel(model_name=model_name_value)
|
264 |
+
|
265 |
+
# Convert final_messages into Gemini's format:
|
266 |
+
# Gemini expects a list of messages: [{"role": "user"/"assistant"/"system", "parts": ["..."]}, ...]
|
267 |
+
gemini_messages = []
|
268 |
+
for m in final_messages:
|
269 |
+
gemini_messages.append({"role": m["role"], "parts": [m["content"]]})
|
270 |
+
|
271 |
+
# Now call generate_content with stream=True
|
272 |
+
try:
|
273 |
+
response = model.generate_content(gemini_messages, stream=True)
|
274 |
+
response_text = ""
|
275 |
+
for chunk in response:
|
276 |
+
if chunk.text:
|
277 |
+
response_text += chunk.text
|
278 |
+
yield response_text
|
279 |
+
except Exception as ex:
|
280 |
+
yield f"Error calling Gemini: {ex}"
|
281 |
+
else:
|
282 |
+
# Default OpenAI-compatible logic
|
283 |
+
from openai import OpenAI
|
284 |
+
import openai
|
285 |
+
import json
|
286 |
+
|
287 |
+
client = OpenAI(
|
288 |
+
base_url=endpoint,
|
289 |
+
api_key=api_key,
|
290 |
)
|
291 |
+
|
292 |
+
try:
|
293 |
+
completion = client.chat.completions.create(
|
294 |
+
model=model_name_value,
|
295 |
+
messages=final_messages,
|
296 |
+
stream=True,
|
297 |
+
)
|
298 |
+
response_text = ""
|
299 |
+
for chunk in completion:
|
300 |
+
delta = chunk.choices[0].delta.content or ""
|
301 |
+
response_text += delta
|
302 |
+
yield response_text
|
303 |
+
except json.JSONDecodeError as e:
|
304 |
+
yield f"JSON decoding error: {e.msg}"
|
305 |
+
except openai.OpenAIError as openai_err:
|
306 |
+
yield f"OpenAI error: {openai_err}"
|
307 |
+
except Exception as ex:
|
308 |
+
yield f"Unexpected error: {ex}"
|
309 |
|
310 |
chatbot = gr.Chatbot(label="Chatbot", scale=1, height=800, autoscroll=True)
|
311 |
chat_interface = gr.ChatInterface(
|
|
|
349 |
|
350 |
# Add a textbox for user to enter a paper_id (arxiv_id)
|
351 |
paper_id_input = gr.Textbox(
|
352 |
+
label="Or enter a 🤗 paper_id directly",
|
353 |
placeholder="e.g. 1234.56789"
|
354 |
)
|
355 |
|
requirements.txt
CHANGED
@@ -7,6 +7,7 @@ PyPDF2
|
|
7 |
transformers
|
8 |
beautifulsoup4
|
9 |
openai
|
|
|
10 |
# Set the primary index URL to PyTorch's CPU wheels
|
11 |
--index-url https://download.pytorch.org/whl/cpu
|
12 |
|
|
|
7 |
transformers
|
8 |
beautifulsoup4
|
9 |
openai
|
10 |
+
google-generativeai
|
11 |
# Set the primary index URL to PyTorch's CPU wheels
|
12 |
--index-url https://download.pytorch.org/whl/cpu
|
13 |
|