Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,121 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
Chat engine.
|
3 |
+
|
4 |
+
TODOs:
|
5 |
+
- Better prompts.
|
6 |
+
- Output reader / parser.
|
7 |
+
- Agents for evaluation and task planning / splitting.
|
8 |
+
* Haystack for orchestration
|
9 |
+
- Tools for agents
|
10 |
+
* Haystack for orchestration
|
11 |
+
-
|
12 |
+
|
13 |
+
"""
|
14 |
+
|
15 |
+
|
16 |
+
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
|
17 |
+
|
18 |
+
def query_submit(user_message, history):
|
19 |
+
return "", history + [[user_message, None]]
|
20 |
+
|
21 |
+
def format_prompt(query, history, lookback):
|
22 |
+
prompt = "Responses should be no more than 100 words long.\n"
|
23 |
+
|
24 |
+
for previous_query, prevous_completion in history[-lookback:]:
|
25 |
+
prompt += f"<s>[INST] {previous_query} [/INST] {prevous_completion}</s> "
|
26 |
+
|
27 |
+
prompt += f"[INST] {query} [/INST]"
|
28 |
+
|
29 |
+
return prompt
|
30 |
+
|
31 |
+
def query_completion(
|
32 |
+
query,
|
33 |
+
history,
|
34 |
+
lookback = 3,
|
35 |
+
max_new_tokens = 256,
|
36 |
+
):
|
37 |
+
|
38 |
+
generateKwargs = dict(
|
39 |
+
max_new_tokens = max_new_tokens,
|
40 |
+
seed = 1337,
|
41 |
+
)
|
42 |
+
|
43 |
+
formatted_query = format_prompt(query, history, lookback)
|
44 |
+
|
45 |
+
stream = client.text_generation(
|
46 |
+
formatted_query,
|
47 |
+
**generateKwargs,
|
48 |
+
stream = True,
|
49 |
+
details = True,
|
50 |
+
return_full_text = False
|
51 |
+
)
|
52 |
+
|
53 |
+
history[-1][1] = ""
|
54 |
+
|
55 |
+
for response in stream:
|
56 |
+
history[-1][1] += response.token.text
|
57 |
+
yield history
|
58 |
+
|
59 |
+
|
60 |
+
"""
|
61 |
+
Chat UI using Gradio Blocks.
|
62 |
+
|
63 |
+
Blocks preferred for lower-level "atomic" layout control and state management.
|
64 |
+
|
65 |
+
TODOs:
|
66 |
+
- State management for dynamic components update.
|
67 |
+
- Add scratpad readout to right of chat log.
|
68 |
+
* Placeholder added for now.
|
69 |
+
- Add functionality to retry button.
|
70 |
+
* Placeholder added for now.
|
71 |
+
- Add dropdown for model selection.
|
72 |
+
- Add textbox for HF model selection.
|
73 |
+
|
74 |
+
"""
|
75 |
+
|
76 |
+
with gr.Blocks() as chatUI:
|
77 |
+
with gr.Row():
|
78 |
+
chatOutput = gr.Chatbot(
|
79 |
+
bubble_full_width = False,
|
80 |
+
scale = 2
|
81 |
+
)
|
82 |
+
agentWhiteBoard = gr.Markdown(scale = 1)
|
83 |
+
|
84 |
+
with gr.Row():
|
85 |
+
queryInput = gr.Textbox(
|
86 |
+
placeholder = "Please enter you question or request here...",
|
87 |
+
show_label = False,
|
88 |
+
scale = 4
|
89 |
+
)
|
90 |
+
submitButton = gr.Button("Submit", scale = 1)
|
91 |
+
|
92 |
+
with gr.Row():
|
93 |
+
retry = gr.Button("Retry (null)")
|
94 |
+
clear = gr.ClearButton([queryInput, chatOutput])
|
95 |
+
|
96 |
+
# gr.State()
|
97 |
+
|
98 |
+
queryInput.submit(
|
99 |
+
fn = query_submit,
|
100 |
+
inputs = [queryInput, chatOutput],
|
101 |
+
outputs = [queryInput, chatOutput],
|
102 |
+
queue = False,
|
103 |
+
).then(
|
104 |
+
fn = query_completion,
|
105 |
+
inputs = [queryInput, chatOutput],
|
106 |
+
outputs = [chatOutput],
|
107 |
+
)
|
108 |
+
|
109 |
+
submitButton.click(
|
110 |
+
fn = query_submit,
|
111 |
+
inputs = [queryInput, chatOutput],
|
112 |
+
outputs = [queryInput, chatOutput],
|
113 |
+
queue = False,
|
114 |
+
).then(
|
115 |
+
fn = query_completion,
|
116 |
+
inputs = [queryInput, chatOutput],
|
117 |
+
outputs = [chatOutput],
|
118 |
+
)
|
119 |
+
|
120 |
+
chatUI.queue()
|
121 |
+
chatUI.launch(show_api = False)
|