Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
import gradio as gr
|
2 |
import time
|
3 |
-
from transformers import pipeline
|
4 |
from sentence_transformers import SentenceTransformer
|
5 |
from sklearn.metrics.pairwise import cosine_similarity
|
6 |
from TTS.api import TTS # Coqui TTS library
|
@@ -9,7 +9,10 @@ import PyPDF2
|
|
9 |
# Initialize Models
|
10 |
stt_model = pipeline("automatic-speech-recognition", model="openai/whisper-tiny")
|
11 |
embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
|
12 |
-
|
|
|
|
|
|
|
13 |
|
14 |
tts_model = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC", progress_bar=False, gpu=False)
|
15 |
|
@@ -28,16 +31,18 @@ def process_inputs(resume, job_desc):
|
|
28 |
job_desc_embedding = embedding_model.encode(job_desc)
|
29 |
return resume_embeddings, job_desc_embedding
|
30 |
|
31 |
-
# Generate a follow-up question
|
32 |
-
def generate_question(response, resume_embeddings):
|
33 |
user_embedding = embedding_model.encode(response)
|
34 |
similarities = {
|
35 |
section: cosine_similarity([user_embedding], [embedding])[0][0]
|
36 |
for section, embedding in resume_embeddings.items()
|
37 |
}
|
38 |
most_relevant_section = max(similarities, key=similarities.get)
|
39 |
-
|
40 |
-
|
|
|
|
|
41 |
return question
|
42 |
|
43 |
# Generate TTS audio for a question
|
@@ -71,7 +76,7 @@ class MockInterview:
|
|
71 |
return "No response detected. Please try again.", None
|
72 |
|
73 |
# Generate the next question
|
74 |
-
self.current_question = generate_question(transcription, self.resume_embeddings)
|
75 |
return transcription, generate_audio(self.current_question)
|
76 |
|
77 |
def end_interview(self):
|
@@ -111,4 +116,3 @@ with interface:
|
|
111 |
|
112 |
if __name__ == "__main__":
|
113 |
interface.launch()
|
114 |
-
|
|
|
1 |
import gradio as gr
|
2 |
import time
|
3 |
+
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
|
4 |
from sentence_transformers import SentenceTransformer
|
5 |
from sklearn.metrics.pairwise import cosine_similarity
|
6 |
from TTS.api import TTS # Coqui TTS library
|
|
|
9 |
# Initialize Models
|
10 |
stt_model = pipeline("automatic-speech-recognition", model="openai/whisper-tiny")
|
11 |
embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
|
12 |
+
|
13 |
+
gpt_model_name = "google/flan-t5-base"
|
14 |
+
gpt_tokenizer = AutoTokenizer.from_pretrained(gpt_model_name)
|
15 |
+
gpt_model = AutoModelForCausalLM.from_pretrained(gpt_model_name)
|
16 |
|
17 |
tts_model = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC", progress_bar=False, gpu=False)
|
18 |
|
|
|
31 |
job_desc_embedding = embedding_model.encode(job_desc)
|
32 |
return resume_embeddings, job_desc_embedding
|
33 |
|
34 |
+
# Generate a follow-up question using Flan-T5
|
35 |
+
def generate_question(response, resume_embeddings, job_desc):
|
36 |
user_embedding = embedding_model.encode(response)
|
37 |
similarities = {
|
38 |
section: cosine_similarity([user_embedding], [embedding])[0][0]
|
39 |
for section, embedding in resume_embeddings.items()
|
40 |
}
|
41 |
most_relevant_section = max(similarities, key=similarities.get)
|
42 |
+
prompt = f"You are a hiring manager. Based on the candidate's experience in {most_relevant_section} and the job description, ask a follow-up question."
|
43 |
+
inputs = gpt_tokenizer(prompt, return_tensors="pt", truncation=True)
|
44 |
+
outputs = gpt_model.generate(**inputs, max_length=50, num_beams=3, early_stopping=True)
|
45 |
+
question = gpt_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
46 |
return question
|
47 |
|
48 |
# Generate TTS audio for a question
|
|
|
76 |
return "No response detected. Please try again.", None
|
77 |
|
78 |
# Generate the next question
|
79 |
+
self.current_question = generate_question(transcription, self.resume_embeddings, self.job_desc_embedding)
|
80 |
return transcription, generate_audio(self.current_question)
|
81 |
|
82 |
def end_interview(self):
|
|
|
116 |
|
117 |
if __name__ == "__main__":
|
118 |
interface.launch()
|
|