added chat, langchat
Browse files- app.py +34 -16
- explore_1.py +37 -0
- explore_2.py +43 -0
- langchat.py +141 -0
- utils.py +0 -2
app.py
CHANGED
@@ -1,20 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
1 |
# import torch
|
2 |
import pickle
|
3 |
-
import
|
4 |
-
from transformers import Conversation, pipeline
|
5 |
from upload import get_file, upload_file
|
6 |
from utils import clear_uploader, undo, restart
|
7 |
|
8 |
|
|
|
|
|
9 |
share_keys = ["messages", "model_name"]
|
10 |
MODELS = [
|
11 |
-
"
|
12 |
-
"
|
13 |
-
"google/flan-t5-
|
14 |
-
"google/flan-t5-
|
15 |
-
"google/flan-t5-
|
|
|
|
|
16 |
]
|
17 |
-
default_model =
|
|
|
18 |
|
19 |
st.set_page_config(
|
20 |
page_title="LLM",
|
@@ -25,10 +35,18 @@ if "model_name" not in st.session_state:
|
|
25 |
st.session_state.model_name = default_model
|
26 |
|
27 |
|
|
|
28 |
def get_pipeline(model_name):
|
29 |
-
|
30 |
-
|
31 |
-
chatbot = pipeline(model=model_name, task="conversational", device=device)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
return chatbot
|
33 |
|
34 |
chatbot = get_pipeline(st.session_state.model_name)
|
@@ -60,7 +78,7 @@ with st.sidebar:
|
|
60 |
st.title(":blue[LLM Only]")
|
61 |
|
62 |
st.subheader("Model")
|
63 |
-
model_name = st.selectbox("Model", MODELS,
|
64 |
|
65 |
if st.button("Share", use_container_width=True):
|
66 |
share()
|
@@ -94,12 +112,12 @@ if prompt := st.chat_input("Type a message", key="chat_input"):
|
|
94 |
|
95 |
if not append:
|
96 |
with st.chat_message("assistant"):
|
97 |
-
|
98 |
for m in st.session_state.messages:
|
99 |
-
|
100 |
-
print(
|
101 |
with st.spinner("Generating response..."):
|
102 |
-
response = chatbot(
|
103 |
response = response[-1]["content"]
|
104 |
st.write(response)
|
105 |
|
|
|
1 |
+
|
2 |
+
import streamlit as st
|
3 |
+
import os
|
4 |
+
os.environ['HF_HOME'] = '/scratch/sroydip1/cache/hf/'
|
5 |
+
os.environ["HUGGINGFACEHUB_API_TOKEN"] = st.secrets["HF_TOKEN"]
|
6 |
# import torch
|
7 |
import pickle
|
8 |
+
import torch
|
9 |
+
from transformers import Conversation, pipeline, AutoTokenizer, AutoModelForCausalLM
|
10 |
from upload import get_file, upload_file
|
11 |
from utils import clear_uploader, undo, restart
|
12 |
|
13 |
|
14 |
+
TOKEN = st.secrets["HF_TOKEN"]
|
15 |
+
|
16 |
share_keys = ["messages", "model_name"]
|
17 |
MODELS = [
|
18 |
+
"meta-llama/Llama-2-7b-chat-hf",
|
19 |
+
"mistralai/Mistral-7B-Instruct-v0.2",
|
20 |
+
# "google/flan-t5-small",
|
21 |
+
# "google/flan-t5-base",
|
22 |
+
# "google/flan-t5-large",
|
23 |
+
# "google/flan-t5-xl",
|
24 |
+
# "google/flan-t5-xxl",
|
25 |
]
|
26 |
+
default_model = MODELS[0]
|
27 |
+
# default_model = "meta-llama/Llama-2-7b-chat-hf"
|
28 |
|
29 |
st.set_page_config(
|
30 |
page_title="LLM",
|
|
|
35 |
st.session_state.model_name = default_model
|
36 |
|
37 |
|
38 |
+
@st.cache_resource
|
39 |
def get_pipeline(model_name):
|
40 |
+
device = 0 if torch.cuda.is_available() else -1
|
41 |
+
# if True or model_name == "meta-llama/Llama-2-7b-chat-hf" or model_name == "mistralai/Mistral-7B-Instruct-v0.2":
|
42 |
+
# chatbot = pipeline(model=model_name, task="conversational", device=device)#, model_kwargs=model_kwargs)
|
43 |
+
# else:
|
44 |
+
# chatbot = pipeline(model=model_name, task="text-generation", device=device)
|
45 |
+
|
46 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, token=TOKEN)
|
47 |
+
model = AutoModelForCausalLM.from_pretrained(model_name, token=TOKEN, load_in_8bit=True)
|
48 |
+
# chatbot = pipeline("conversational", model=model, tokenizer=tokenizer, device=device)
|
49 |
+
chatbot = pipeline("conversational", model=model, tokenizer=tokenizer)
|
50 |
return chatbot
|
51 |
|
52 |
chatbot = get_pipeline(st.session_state.model_name)
|
|
|
78 |
st.title(":blue[LLM Only]")
|
79 |
|
80 |
st.subheader("Model")
|
81 |
+
model_name = st.selectbox("Model", MODELS, key="model_name")
|
82 |
|
83 |
if st.button("Share", use_container_width=True):
|
84 |
share()
|
|
|
112 |
|
113 |
if not append:
|
114 |
with st.chat_message("assistant"):
|
115 |
+
chat = Conversation()
|
116 |
for m in st.session_state.messages:
|
117 |
+
chat.add_message(m)
|
118 |
+
print(chat)
|
119 |
with st.spinner("Generating response..."):
|
120 |
+
response = chatbot(chat)
|
121 |
response = response[-1]["content"]
|
122 |
st.write(response)
|
123 |
|
explore_1.py
ADDED
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
os.environ["HUGGINGFACEHUB_API_TOKEN"] = ""
|
3 |
+
|
4 |
+
from langchain_community.llms import HuggingFaceHub
|
5 |
+
|
6 |
+
llm = HuggingFaceHub(
|
7 |
+
repo_id="meta-llama/Llama-2-7b-chat-hf",
|
8 |
+
task="text-generation",
|
9 |
+
model_kwargs={
|
10 |
+
"max_new_tokens": 512,
|
11 |
+
"temperature": 0.1,
|
12 |
+
"seed": 42,
|
13 |
+
},
|
14 |
+
)
|
15 |
+
|
16 |
+
from langchain.schema import (
|
17 |
+
HumanMessage,
|
18 |
+
SystemMessage,
|
19 |
+
AIMessage,
|
20 |
+
)
|
21 |
+
from langchain_community.chat_models.huggingface import ChatHuggingFace
|
22 |
+
|
23 |
+
messages = [
|
24 |
+
SystemMessage(content="You're a helpful assistant"),
|
25 |
+
]
|
26 |
+
|
27 |
+
chat_model = ChatHuggingFace(llm=llm)
|
28 |
+
|
29 |
+
|
30 |
+
while True:
|
31 |
+
question = input("You: ")
|
32 |
+
messages.append(HumanMessage(content=question))
|
33 |
+
response = chat_model.invoke(messages)
|
34 |
+
print(response)
|
35 |
+
response = response.content
|
36 |
+
messages.append(AIMessage(content=response))
|
37 |
+
print(f"Bot: {response}")
|
explore_2.py
ADDED
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
os.environ["HUGGINGFACEHUB_API_TOKEN"] = ""
|
3 |
+
|
4 |
+
from langchain.prompts import PromptTemplate
|
5 |
+
from langchain.chains import LLMChain
|
6 |
+
from langchain.memory import ConversationBufferMemory
|
7 |
+
from langchain_community.llms import HuggingFaceHub
|
8 |
+
|
9 |
+
template = """You are a friendly chatbot engaging in a conversation with a human.
|
10 |
+
|
11 |
+
Previous conversation:
|
12 |
+
{chat_history}
|
13 |
+
|
14 |
+
New human question: {question}
|
15 |
+
Response:"""
|
16 |
+
|
17 |
+
def get_pipeline(model_name):
|
18 |
+
llm = HuggingFaceHub(
|
19 |
+
repo_id=model_name,
|
20 |
+
task="text-generation",
|
21 |
+
model_kwargs={
|
22 |
+
"max_new_tokens": 250,
|
23 |
+
"top_k": 30,
|
24 |
+
"temperature": 0.1,
|
25 |
+
"repetition_penalty": 1.03,
|
26 |
+
},
|
27 |
+
)
|
28 |
+
return llm
|
29 |
+
|
30 |
+
|
31 |
+
chatbot = get_pipeline("mistralai/Mistral-7B-Instruct-v0.2")
|
32 |
+
memory = ConversationBufferMemory(memory_key="chat_history")
|
33 |
+
prompt_template = PromptTemplate.from_template(template)
|
34 |
+
conversation = LLMChain(llm=chatbot, prompt=prompt_template, verbose=True, memory=memory)
|
35 |
+
|
36 |
+
while True:
|
37 |
+
question = input("You: ")
|
38 |
+
response = conversation({"question": question})
|
39 |
+
print("-" * 50)
|
40 |
+
print(response)
|
41 |
+
print(response["text"])
|
42 |
+
print("-" * 50)
|
43 |
+
print()
|
langchat.py
ADDED
@@ -0,0 +1,141 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
|
3 |
+
os.environ["HF_HOME"] = "/scratch/sroydip1/cache/hf/"
|
4 |
+
os.environ["HUGGINGFACEHUB_API_TOKEN"] = ""
|
5 |
+
# import torch
|
6 |
+
import pickle
|
7 |
+
import torch
|
8 |
+
import streamlit as st
|
9 |
+
from transformers import Conversation, pipeline
|
10 |
+
from upload import get_file, upload_file
|
11 |
+
from utils import clear_uploader, undo, restart
|
12 |
+
|
13 |
+
from langchain.prompts import PromptTemplate
|
14 |
+
from langchain.chains import LLMChain
|
15 |
+
from langchain.memory import ConversationBufferMemory
|
16 |
+
from langchain_community.llms import HuggingFaceHub
|
17 |
+
|
18 |
+
|
19 |
+
share_keys = ["messages", "model_name"]
|
20 |
+
MODELS = [
|
21 |
+
"mistralai/Mistral-7B-Instruct-v0.2",
|
22 |
+
"google/flan-t5-small",
|
23 |
+
"google/flan-t5-base",
|
24 |
+
"google/flan-t5-large",
|
25 |
+
"google/flan-t5-xl",
|
26 |
+
"google/flan-t5-xxl",
|
27 |
+
]
|
28 |
+
default_model = "mistralai/Mistral-7B-Instruct-v0.2"
|
29 |
+
# default_model = "meta-llama/Llama-2-7b-chat-hf"
|
30 |
+
|
31 |
+
st.set_page_config(
|
32 |
+
page_title="LLM",
|
33 |
+
page_icon="π",
|
34 |
+
)
|
35 |
+
|
36 |
+
if "model_name" not in st.session_state:
|
37 |
+
st.session_state.model_name = default_model
|
38 |
+
|
39 |
+
template = """You are a friendly chatbot engaging in a conversation with a human.
|
40 |
+
|
41 |
+
Previous conversation:
|
42 |
+
{chat_history}
|
43 |
+
|
44 |
+
New human question: {question}
|
45 |
+
Response:"""
|
46 |
+
|
47 |
+
|
48 |
+
def get_pipeline(model_name):
|
49 |
+
llm = HuggingFaceHub(
|
50 |
+
repo_id=model_name,
|
51 |
+
task="text-generation",
|
52 |
+
model_kwargs={
|
53 |
+
"max_new_tokens": 512,
|
54 |
+
"top_k": 30,
|
55 |
+
"temperature": 0.1,
|
56 |
+
"repetition_penalty": 1.03,
|
57 |
+
},
|
58 |
+
)
|
59 |
+
return llm
|
60 |
+
|
61 |
+
|
62 |
+
chatbot = get_pipeline(st.session_state.model_name)
|
63 |
+
memory = ConversationBufferMemory(memory_key="chat_history")
|
64 |
+
prompt_template = PromptTemplate.from_template(template)
|
65 |
+
conversation = LLMChain(llm=chatbot, prompt=prompt_template, verbose=True, memory=memory)
|
66 |
+
|
67 |
+
|
68 |
+
if "messages" not in st.session_state:
|
69 |
+
st.session_state.messages = []
|
70 |
+
|
71 |
+
if len(st.session_state.messages) == 0 and "id" in st.query_params:
|
72 |
+
with st.spinner("Loading chat..."):
|
73 |
+
id = st.query_params["id"]
|
74 |
+
data = get_file(id)
|
75 |
+
obj = pickle.loads(data)
|
76 |
+
for k, v in obj.items():
|
77 |
+
st.session_state[k] = v
|
78 |
+
|
79 |
+
|
80 |
+
def share():
|
81 |
+
obj = {}
|
82 |
+
for k in share_keys:
|
83 |
+
if k in st.session_state:
|
84 |
+
obj[k] = st.session_state[k]
|
85 |
+
data = pickle.dumps(obj)
|
86 |
+
id = upload_file(data)
|
87 |
+
url = f"https://umbc-nlp-chat-llm.hf.space/?id={id}"
|
88 |
+
st.markdown(f"[share](/?id={id})")
|
89 |
+
st.success(f"Share URL: {url}")
|
90 |
+
|
91 |
+
|
92 |
+
with st.sidebar:
|
93 |
+
st.title(":blue[LLM Only]")
|
94 |
+
|
95 |
+
st.subheader("Model")
|
96 |
+
model_name = st.selectbox(
|
97 |
+
"Model", MODELS, index=MODELS.index(st.session_state.model_name)
|
98 |
+
)
|
99 |
+
|
100 |
+
if st.button("Share", use_container_width=True):
|
101 |
+
share()
|
102 |
+
|
103 |
+
cols = st.columns(2)
|
104 |
+
with cols[0]:
|
105 |
+
if st.button("Restart", type="primary", use_container_width=True):
|
106 |
+
restart()
|
107 |
+
|
108 |
+
with cols[1]:
|
109 |
+
if st.button("Undo", use_container_width=True):
|
110 |
+
undo()
|
111 |
+
|
112 |
+
append = st.checkbox("Append to previous message", value=False)
|
113 |
+
|
114 |
+
|
115 |
+
for message in st.session_state.messages:
|
116 |
+
with st.chat_message(message["role"]):
|
117 |
+
st.markdown(message["content"])
|
118 |
+
|
119 |
+
|
120 |
+
def push_message(role, content):
|
121 |
+
message = {"role": role, "content": content}
|
122 |
+
st.session_state.messages.append(message)
|
123 |
+
return message
|
124 |
+
|
125 |
+
|
126 |
+
if prompt := st.chat_input("Type a message", key="chat_input"):
|
127 |
+
push_message("user", prompt)
|
128 |
+
with st.chat_message("user"):
|
129 |
+
st.markdown(prompt)
|
130 |
+
|
131 |
+
if not append:
|
132 |
+
with st.chat_message("assistant"):
|
133 |
+
print(conversation)
|
134 |
+
with st.spinner("Generating response..."):
|
135 |
+
response = conversation({"question": prompt})
|
136 |
+
print(response)
|
137 |
+
response = response["text"]
|
138 |
+
st.write(response)
|
139 |
+
|
140 |
+
push_message("assistant", response)
|
141 |
+
clear_uploader()
|
utils.py
CHANGED
@@ -9,8 +9,6 @@ def undo():
|
|
9 |
if len(st.session_state.messages) > 0:
|
10 |
st.query_params.clear()
|
11 |
msg = st.session_state.messages.pop()
|
12 |
-
if msg["role"] == "assistant" and "cost" in st.session_state:
|
13 |
-
st.session_state.cost.pop()
|
14 |
time.sleep(0.1)
|
15 |
st.rerun()
|
16 |
|
|
|
9 |
if len(st.session_state.messages) > 0:
|
10 |
st.query_params.clear()
|
11 |
msg = st.session_state.messages.pop()
|
|
|
|
|
12 |
time.sleep(0.1)
|
13 |
st.rerun()
|
14 |
|