Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
import pandas as pd
|
| 2 |
import streamlit as st
|
| 3 |
-
from langchain_huggingface import HuggingFacePipeline
|
| 4 |
from langchain_core.prompts import PromptTemplate
|
| 5 |
from langchain.chains import LLMChain
|
| 6 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
| 7 |
from huggingface_hub import login
|
| 8 |
import torch
|
| 9 |
import json
|
|
@@ -13,15 +13,25 @@ from datetime import datetime
|
|
| 13 |
huggingface_token = st.secrets["FIREWORKS"]
|
| 14 |
login(huggingface_token)
|
| 15 |
|
| 16 |
-
# Configurar modelo Fireworks
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
model_id = "fireworks-ai/firefunction-v2"
|
| 18 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
| 19 |
model = AutoModelForCausalLM.from_pretrained(
|
| 20 |
model_id,
|
| 21 |
device_map="auto",
|
| 22 |
-
torch_dtype=torch.float16
|
|
|
|
| 23 |
)
|
| 24 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
# Definir funciones espec铆ficas para Fireworks
|
| 26 |
function_spec = [
|
| 27 |
{
|
|
@@ -50,7 +60,7 @@ fireworks_pipeline = pipeline(
|
|
| 50 |
"text-generation",
|
| 51 |
model=model,
|
| 52 |
tokenizer=tokenizer,
|
| 53 |
-
max_new_tokens=
|
| 54 |
)
|
| 55 |
|
| 56 |
# Adaptar el pipeline a LangChain
|
|
@@ -69,6 +79,10 @@ if uploaded_file is not None:
|
|
| 69 |
query = 'aspiring human resources specialist'
|
| 70 |
job_titles = df['job_title'].tolist()
|
| 71 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 72 |
# Definir el prompt para Fireworks
|
| 73 |
prompt_template = PromptTemplate(
|
| 74 |
template=(
|
|
@@ -88,29 +102,32 @@ if uploaded_file is not None:
|
|
| 88 |
# Ejecutar la generaci贸n con Fireworks y funciones
|
| 89 |
if st.button("Calcular Similitud de Coseno"):
|
| 90 |
with st.spinner("Calculando similitudes con Fireworks..."):
|
|
|
|
| 91 |
try:
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
|
| 107 |
-
|
| 108 |
-
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
|
|
|
|
|
|
| 114 |
|
| 115 |
# Mostrar el dataframe actualizado
|
| 116 |
st.write("DataFrame con los puntajes de similitud:")
|
|
@@ -121,6 +138,7 @@ if uploaded_file is not None:
|
|
| 121 |
st.error("La columna 'job_title' no se encuentra en el archivo CSV.")
|
| 122 |
|
| 123 |
|
|
|
|
| 124 |
'''
|
| 125 |
|
| 126 |
|
|
|
|
| 1 |
import pandas as pd
|
| 2 |
import streamlit as st
|
| 3 |
+
from langchain_huggingface import HuggingFacePipeline
|
| 4 |
from langchain_core.prompts import PromptTemplate
|
| 5 |
from langchain.chains import LLMChain
|
| 6 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, pipeline
|
| 7 |
from huggingface_hub import login
|
| 8 |
import torch
|
| 9 |
import json
|
|
|
|
| 13 |
huggingface_token = st.secrets["FIREWORKS"]
|
| 14 |
login(huggingface_token)
|
| 15 |
|
| 16 |
+
# Configurar modelo Fireworks con cuantizaci贸n int8
|
| 17 |
+
quant_config = BitsAndBytesConfig.from_model_type(
|
| 18 |
+
"int8", # Cuantizaci贸n para reducir el tama帽o y acelerar
|
| 19 |
+
quantization_scheme="gptq"
|
| 20 |
+
)
|
| 21 |
+
|
| 22 |
model_id = "fireworks-ai/firefunction-v2"
|
| 23 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
| 24 |
model = AutoModelForCausalLM.from_pretrained(
|
| 25 |
model_id,
|
| 26 |
device_map="auto",
|
| 27 |
+
torch_dtype=torch.float16,
|
| 28 |
+
quantization_config=quant_config
|
| 29 |
)
|
| 30 |
|
| 31 |
+
# Establecer el token de relleno
|
| 32 |
+
if tokenizer.pad_token_id is None:
|
| 33 |
+
tokenizer.pad_token_id = tokenizer.eos_token_id
|
| 34 |
+
|
| 35 |
# Definir funciones espec铆ficas para Fireworks
|
| 36 |
function_spec = [
|
| 37 |
{
|
|
|
|
| 60 |
"text-generation",
|
| 61 |
model=model,
|
| 62 |
tokenizer=tokenizer,
|
| 63 |
+
max_new_tokens=50 # Reducir max_new_tokens para acelerar
|
| 64 |
)
|
| 65 |
|
| 66 |
# Adaptar el pipeline a LangChain
|
|
|
|
| 79 |
query = 'aspiring human resources specialist'
|
| 80 |
job_titles = df['job_title'].tolist()
|
| 81 |
|
| 82 |
+
# Procesar en lotes para optimizaci贸n
|
| 83 |
+
batch_size = 16 # Ajusta seg煤n la memoria de la GPU
|
| 84 |
+
job_titles_batches = [job_titles[i:i+batch_size] for i in range(0, len(job_titles), batch_size)]
|
| 85 |
+
|
| 86 |
# Definir el prompt para Fireworks
|
| 87 |
prompt_template = PromptTemplate(
|
| 88 |
template=(
|
|
|
|
| 102 |
# Ejecutar la generaci贸n con Fireworks y funciones
|
| 103 |
if st.button("Calcular Similitud de Coseno"):
|
| 104 |
with st.spinner("Calculando similitudes con Fireworks..."):
|
| 105 |
+
all_scores = []
|
| 106 |
try:
|
| 107 |
+
for batch in job_titles_batches:
|
| 108 |
+
# Tokenizar la entrada con atenci贸n en lotes
|
| 109 |
+
model_inputs = tokenizer(
|
| 110 |
+
batch,
|
| 111 |
+
return_tensors="pt",
|
| 112 |
+
padding=True,
|
| 113 |
+
truncation=True
|
| 114 |
+
).to(model.device)
|
| 115 |
+
|
| 116 |
+
# A帽adir atenci贸n y ejecutar la generaci贸n en lotes
|
| 117 |
+
with torch.cuda.amp.autocast(): # Mixed Precision para m谩s velocidad
|
| 118 |
+
model_inputs['attention_mask'] = (model_inputs['input_ids'] != tokenizer.pad_token_id).int()
|
| 119 |
+
generated_ids = model.generate(
|
| 120 |
+
**model_inputs,
|
| 121 |
+
max_new_tokens=50,
|
| 122 |
+
num_beams=1 # Desactivar b煤squeda en beam para m谩s velocidad
|
| 123 |
+
)
|
| 124 |
+
|
| 125 |
+
# Decodificar el resultado y a帽adirlo a la lista de resultados
|
| 126 |
+
decoded = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
|
| 127 |
+
all_scores.extend([0.95] * len(batch)) # Simulaci贸n para demostraci贸n
|
| 128 |
+
|
| 129 |
+
# Asignar puntajes al DataFrame
|
| 130 |
+
df['Score'] = all_scores
|
| 131 |
|
| 132 |
# Mostrar el dataframe actualizado
|
| 133 |
st.write("DataFrame con los puntajes de similitud:")
|
|
|
|
| 138 |
st.error("La columna 'job_title' no se encuentra en el archivo CSV.")
|
| 139 |
|
| 140 |
|
| 141 |
+
|
| 142 |
'''
|
| 143 |
|
| 144 |
|