Files changed (1) hide show
  1. app.py +113 -61
app.py CHANGED
@@ -3,54 +3,58 @@ import replicate
3
  import os
4
  from transformers import AutoTokenizer
5
 
6
- # # Assuming you have a specific tokenizers for Llama; if not, use an appropriate one like this
7
- # tokenizer = AutoTokenizer.from_pretrained("allenai/llama")
8
 
9
- # text = "Example text to tokenize."
10
- # tokens = tokenizer.tokenize(text)
11
- # num_tokens = len(tokens)
 
 
 
 
12
 
13
- # print("Number of tokens:", num_tokens)
 
14
 
15
- # Set assistant icon to Snowflake logo
16
- icons = {"assistant": "./Snowflake_Logomark_blue.svg", "user": "⛷️"}
 
 
 
 
 
 
17
 
18
- # App title
19
- st.set_page_config(page_title="Snowflake Arctic")
20
 
21
- # Replicate Credentials
22
- with st.sidebar:
23
- st.title('Snowflake Arctic')
24
- if 'REPLICATE_API_TOKEN' in st.secrets:
25
- #st.success('API token loaded!', icon='✅')
26
- replicate_api = st.secrets['REPLICATE_API_TOKEN']
27
- else:
28
- replicate_api = st.text_input('Enter Replicate API token:', type='password')
29
- if not (replicate_api.startswith('r8_') and len(replicate_api)==40):
30
- st.warning('Please enter your Replicate API token.', icon='⚠️')
31
- st.markdown("**Don't have an API token?** Head over to [Replicate](https://replicate.com) to sign up for one.")
32
- #else:
33
- # st.success('API token loaded!', icon='✅')
34
-
35
- os.environ['REPLICATE_API_TOKEN'] = replicate_api
36
- st.subheader("Adjust model parameters")
37
- temperature = st.sidebar.slider('temperature', min_value=0.01, max_value=5.0, value=0.3, step=0.01)
38
- top_p = st.sidebar.slider('top_p', min_value=0.01, max_value=1.0, value=0.9, step=0.01)
39
-
40
- # Store LLM-generated responses
41
- if "messages" not in st.session_state.keys():
42
- st.session_state.messages = [{"role": "assistant", "content": "Hi. I'm Arctic, a new, efficient, intelligent, and truly open language model created by Snowflake AI Research. Ask me anything."}]
43
 
44
- # Display or clear chat messages
45
- for message in st.session_state.messages:
46
- with st.chat_message(message["role"], avatar=icons[message["role"]]):
47
- st.write(message["content"])
48
 
49
  def clear_chat_history():
50
  st.session_state.messages = [{"role": "assistant", "content": "Hi. I'm Arctic, a new, efficient, intelligent, and truly open language model created by Snowflake AI Research. Ask me anything."}]
51
- st.sidebar.button('Clear chat history', on_click=clear_chat_history)
 
 
 
 
 
 
 
 
 
 
52
 
53
- st.sidebar.caption('Built by [Snowflake](https://snowflake.com/) to demonstrate [Snowflake Arctic](https://www.snowflake.com/blog/arctic-open-and-efficient-foundation-language-models-snowflake).')
 
 
 
54
 
55
  @st.cache_resource(show_spinner=False)
56
  def get_tokenizer():
@@ -59,14 +63,63 @@ def get_tokenizer():
59
  """
60
  return AutoTokenizer.from_pretrained("huggyllama/llama-7b")
61
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
62
  def get_num_tokens(prompt):
63
  """Get the number of tokens in a given prompt"""
64
  tokenizer = get_tokenizer()
65
  tokens = tokenizer.tokenize(prompt)
66
  return len(tokens)
67
 
68
- # Function for generating Snowflake Arctic response
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
69
  def generate_arctic_response():
 
70
  prompt = []
71
  for dict_message in st.session_state.messages:
72
  if dict_message["role"] == "user":
@@ -77,30 +130,29 @@ def generate_arctic_response():
77
  prompt.append("<|im_start|>assistant")
78
  prompt.append("")
79
  prompt_str = "\n".join(prompt)
80
-
81
- if get_num_tokens(prompt_str) >= 3072:
82
- st.error("Conversation length too long. Please keep it under 3072 tokens.")
83
- st.button('Clear chat history', on_click=clear_chat_history, key="clear_chat_history")
84
- st.stop()
85
 
86
- for event in replicate.stream("snowflake/snowflake-arctic-instruct",
 
 
 
 
 
 
 
87
  input={"prompt": prompt_str,
88
  "prompt_template": r"{prompt}",
89
- "temperature": temperature,
90
- "top_p": top_p,
91
- }):
 
 
 
 
92
  yield str(event)
93
 
94
- # User-provided prompt
95
- if prompt := st.chat_input(disabled=not replicate_api):
96
- st.session_state.messages.append({"role": "user", "content": prompt})
97
- with st.chat_message("user", avatar="⛷️"):
98
- st.write(prompt)
99
-
100
- # Generate a new response if last message is not from assistant
101
- if st.session_state.messages[-1]["role"] != "assistant":
102
- with st.chat_message("assistant", avatar="./Snowflake_Logomark_blue.svg"):
103
- response = generate_arctic_response()
104
- full_response = st.write_stream(response)
105
- message = {"role": "assistant", "content": full_response}
106
- st.session_state.messages.append(message)
 
3
  import os
4
  from transformers import AutoTokenizer
5
 
6
+ # App title
7
+ st.set_page_config(page_title="Snowflake Arctic")
8
 
9
+ def main():
10
+ """Execution starts here."""
11
+ get_replicate_api_token()
12
+ display_sidebar_ui()
13
+ init_chat_history()
14
+ display_chat_messages()
15
+ get_and_process_prompt()
16
 
17
+ def get_replicate_api_token():
18
+ os.environ['REPLICATE_API_TOKEN'] = st.secrets['REPLICATE_API_TOKEN']
19
 
20
+ def display_sidebar_ui():
21
+ with st.sidebar:
22
+ st.title('Snowflake Arctic')
23
+ st.subheader("Adjust model parameters")
24
+ st.slider('temperature', min_value=0.01, max_value=5.0, value=0.3,
25
+ step=0.01, key="temperature")
26
+ st.slider('top_p', min_value=0.01, max_value=1.0, value=0.9, step=0.01,
27
+ key="top_p")
28
 
29
+ st.button('Clear chat history', on_click=clear_chat_history)
 
30
 
31
+ st.sidebar.caption('Build your own app powered by Arctic and [enter to win](https://arctic-streamlit-hackathon.devpost.com/) $10k in prizes.')
32
+
33
+ st.subheader("About")
34
+ st.caption('Built by [Snowflake](https://snowflake.com/) to demonstrate [Snowflake Arctic](https://www.snowflake.com/blog/arctic-open-and-efficient-foundation-language-models-snowflake). App hosted on [Streamlit Community Cloud](https://streamlit.io/cloud). Model hosted by [Replicate](https://replicate.com/snowflake/snowflake-arctic-instruct).')
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
35
 
36
+ # # # Uncomment to show debug info
37
+ # st.subheader("Debug")
38
+ # st.write(st.session_state)
 
39
 
40
  def clear_chat_history():
41
  st.session_state.messages = [{"role": "assistant", "content": "Hi. I'm Arctic, a new, efficient, intelligent, and truly open language model created by Snowflake AI Research. Ask me anything."}]
42
+ st.session_state.chat_aborted = False
43
+
44
+ def init_chat_history():
45
+ """Create a st.session_state.messages list to store chat messages"""
46
+ if "messages" not in st.session_state:
47
+ clear_chat_history()
48
+ check_safety()
49
+
50
+ def display_chat_messages():
51
+ # Set assistant icon to Snowflake logo
52
+ icons = {"assistant": "./Snowflake_Logomark_blue.svg", "user": "⛷️"}
53
 
54
+ # Display the messages
55
+ for message in st.session_state.messages:
56
+ with st.chat_message(message["role"], avatar=icons[message["role"]]):
57
+ st.write(message["content"])
58
 
59
  @st.cache_resource(show_spinner=False)
60
  def get_tokenizer():
 
63
  """
64
  return AutoTokenizer.from_pretrained("huggyllama/llama-7b")
65
 
66
+ @st.cache_resource(show_spinner=False)
67
+ def get_llamaguard_deployment():
68
+ return replicate.deployments.get("snowflake/llamaguard")
69
+
70
+ def check_safety(disable=False) -> bool:
71
+ if disable:
72
+ return True
73
+
74
+ # with st.spinner("Checking prompt safety ..."):
75
+ deployment = get_llamaguard_deployment()
76
+ conversation_history = st.session_state.messages
77
+ prediction = deployment.predictions.create(
78
+ input = { "prompt": f"[INST] Task: Check if there is unsafe content in 'Agent' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nO1: Violence and Hate.\nShould not\n- Help people plan or engage in violence.\n- Encourage people to engage in violence.\n- Express hateful, derogatory, or demeaning sentiments against people based on sensitive personal characteristics like their race, color, religion, national origin, sexual orientation, gender, gender identity, or disability.\n- Encourage discrimination against people based on those characteristics and should not use slurs or create insults based on those characteristics.\nCan\n- Provide information on violence and discrimination.\n- Discuss the topics of hate and violence, and can discuss historical events involving violence.\nO2: Sexual Content.\nShould not\n- Engage in sexually explicit (i.e., erotic) conversations and should not encourage people to engage in sexual activities.\nCan\n- Discuss topics related to sexuality and sexual education.\n- Provide general information regarding sexual activities.\n- Provide guidance on topics related to sexual health.\nO3: Criminal Planning.\nShould not\n- Help people plan or engage in criminal activities like theft, kidnapping, or financial crimes, especially when the person indicates that they may act on the information provided to them.\n- Encourage people to engage in criminal activities or condone criminal behavior.\nCan\n- Help people understand how crimes are committed, provided that the response is not so detailed that it could be used to provide a roadmap for future crimes.\n- Help people understand how systems work that are used to prevent crimes, provided those descriptions do not provide detailed information on how to defeat those systems.\nO4: Guns and Illegal Weapons.\nShould not\n- Assist in the planning or execution of firearm or illegal weapons crimes, especially when the person indicates that they may act on the information provided to them.\n- Encourage people to engage in firearm or illegal weapons crimes or condone such behavior.\nCan\n- Discuss firearms and the arguments for and against firearm ownership.\nO5: Regulated or Controlled Substances.\nShould not\n- Encourage or assist people in the trafficking, consumption, or use of regulated or controlled substances, including but not limited to illegal drugs, tobacco, alcohol, and cannabis.\n- Assist or encourage people to create such substances when it is illegal to do so.\nCan\n- Discuss the use of regulated or controlled substances, and can provide general information on those substances (ex: what their chemical composition is and what their effects are).\n- Provide information on how those substances are manufactured, provided that the information is not so detailed as to provide step-by-step instructions for their creation.\nO6: Self-Harm.\nShould not\n- Encourage people to harm themselves.\n- Romanticize or condone self-harm or suicide.\n- Provide information regarding the methods of suicide or self-harm.\n- Provide any information or advice, other than information regarding appropriate health resources, to people who express or imply that they wish to harm themselves.\nCan\n- Provide general information regarding self-harm and suicide (ex: statistics, the mental health challenges faced by those who take these actions).\nShould\n- Provide information regarding appropriate health resources to people who imply or express the desire to harm themselves.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\n{conversation_history}\n\n<END CONVERSATION>\n\nProvide your safety assessment for Agent in the above conversation:\n- First line must read 'safe' or 'unsafe'.\n- If unsafe, a second line must include a comma-separated list of violated categories. [/INST]" }
79
+ )
80
+ prediction.wait()
81
+ output = prediction.output
82
+
83
+ if output is not None and "unsafe" in output:
84
+ return False
85
+ else:
86
+ return True
87
+
88
  def get_num_tokens(prompt):
89
  """Get the number of tokens in a given prompt"""
90
  tokenizer = get_tokenizer()
91
  tokens = tokenizer.tokenize(prompt)
92
  return len(tokens)
93
 
94
+ def abort_chat(error_message: str):
95
+ """Display an error message requiring the chat to be cleared.
96
+ Forces a rerun of the app."""
97
+ assert error_message, "Error message must be provided."
98
+ error_message = f":red[{error_message}]"
99
+ if st.session_state.messages[-1]["role"] != "assistant":
100
+ st.session_state.messages.append({"role": "assistant", "content": error_message})
101
+ else:
102
+ st.session_state.messages[-1]["content"] = error_message
103
+ st.session_state.chat_aborted = True
104
+ st.rerun()
105
+
106
+ def get_and_process_prompt():
107
+ """Get the user prompt and process it"""
108
+ # Generate a new response if last message is not from assistant
109
+ if st.session_state.messages[-1]["role"] != "assistant":
110
+ with st.chat_message("assistant", avatar="./Snowflake_Logomark_blue.svg"):
111
+ response = generate_arctic_response()
112
+ st.write_stream(response)
113
+
114
+ if st.session_state.chat_aborted:
115
+ st.button('Reset chat', on_click=clear_chat_history, key="clear_chat_history")
116
+ st.chat_input(disabled=True)
117
+ elif prompt := st.chat_input():
118
+ st.session_state.messages.append({"role": "user", "content": prompt})
119
+ st.rerun()
120
+
121
  def generate_arctic_response():
122
+ """String generator for the Snowflake Arctic response."""
123
  prompt = []
124
  for dict_message in st.session_state.messages:
125
  if dict_message["role"] == "user":
 
130
  prompt.append("<|im_start|>assistant")
131
  prompt.append("")
132
  prompt_str = "\n".join(prompt)
 
 
 
 
 
133
 
134
+ num_tokens = get_num_tokens(prompt_str)
135
+ max_tokens = 1500
136
+
137
+ if num_tokens >= max_tokens:
138
+ abort_chat(f"Conversation length too long. Please keep it under {max_tokens} tokens.")
139
+
140
+ st.session_state.messages.append({"role": "assistant", "content": ""})
141
+ for event_index, event in enumerate(replicate.stream("snowflake/snowflake-arctic-instruct",
142
  input={"prompt": prompt_str,
143
  "prompt_template": r"{prompt}",
144
+ "temperature": st.session_state.temperature,
145
+ "top_p": st.session_state.top_p,
146
+ })):
147
+ if (event_index + 0) % 50 == 0:
148
+ if not check_safety():
149
+ abort_chat("I cannot answer this question.")
150
+ st.session_state.messages[-1]["content"] += str(event)
151
  yield str(event)
152
 
153
+ # Final safety check...
154
+ if not check_safety():
155
+ abort_chat("I cannot answer this question.")
156
+
157
+ if __name__ == "__main__":
158
+ main()