Spaces:
Sleeping
Sleeping
Update app.py retry button
Browse files
app.py
CHANGED
@@ -23,9 +23,6 @@ selected_model = "mistralai/Mixtral-8x7B-Instruct-v0.1"
|
|
23 |
|
24 |
client = InferenceClient(selected_model)
|
25 |
|
26 |
-
def query_submit(user_message, history):
|
27 |
-
return "", history + [[user_message, None]]
|
28 |
-
|
29 |
def format_prompt(query, history, lookback):
|
30 |
prompt = "Responses should be no more than 100 words long.\n"
|
31 |
|
@@ -35,6 +32,9 @@ def format_prompt(query, history, lookback):
|
|
35 |
prompt += f"[INST] {query} [/INST]"
|
36 |
|
37 |
return prompt
|
|
|
|
|
|
|
38 |
|
39 |
def query_completion(
|
40 |
query,
|
@@ -64,24 +64,62 @@ def query_completion(
|
|
64 |
history[-1][1] += response.token.text
|
65 |
yield history
|
66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
|
68 |
"""
|
69 |
Chat UI using Gradio Blocks.
|
70 |
|
71 |
-
Blocks preferred for lower-level
|
72 |
|
73 |
TODOs:
|
74 |
- State management for dynamic components update.
|
75 |
-
- Add
|
76 |
* Placeholder added for now.
|
77 |
- Add functionality to retry button.
|
78 |
* Placeholder added for now.
|
79 |
- Add dropdown for model selection.
|
80 |
-
|
81 |
|
82 |
"""
|
83 |
|
84 |
with gr.Blocks() as chatUI:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
with gr.Row():
|
86 |
chatOutput = gr.Chatbot(
|
87 |
bubble_full_width = False,
|
@@ -98,16 +136,11 @@ with gr.Blocks() as chatUI:
|
|
98 |
submitButton = gr.Button("Submit", scale = 1)
|
99 |
|
100 |
with gr.Row():
|
101 |
-
retry = gr.Button("Retry (null)")
|
102 |
-
clear = gr.ClearButton([queryInput, chatOutput])
|
103 |
-
|
104 |
-
with gr.Row():
|
105 |
-
modelSelect = gr.Dropdown(
|
106 |
-
label = "Model selection:",
|
107 |
-
)
|
108 |
fileUpload = gr.File(
|
109 |
height = 100,
|
110 |
)
|
|
|
|
|
111 |
|
112 |
with gr.Row():
|
113 |
with gr.Accordion(label = "Expand for edit system prompt:"):
|
@@ -118,11 +151,11 @@ with gr.Blocks() as chatUI:
|
|
118 |
scale = 4,
|
119 |
)
|
120 |
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
# gr.State()
|
125 |
|
|
|
126 |
queryInput.submit(
|
127 |
fn = query_submit,
|
128 |
inputs = [queryInput, chatOutput],
|
@@ -145,5 +178,12 @@ with gr.Blocks() as chatUI:
|
|
145 |
outputs = [chatOutput],
|
146 |
)
|
147 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
148 |
chatUI.queue()
|
149 |
chatUI.launch(show_api = False)
|
|
|
23 |
|
24 |
client = InferenceClient(selected_model)
|
25 |
|
|
|
|
|
|
|
26 |
def format_prompt(query, history, lookback):
|
27 |
prompt = "Responses should be no more than 100 words long.\n"
|
28 |
|
|
|
32 |
prompt += f"[INST] {query} [/INST]"
|
33 |
|
34 |
return prompt
|
35 |
+
|
36 |
+
def query_submit(user_message, history):
|
37 |
+
return "", history + [[user_message, None]]
|
38 |
|
39 |
def query_completion(
|
40 |
query,
|
|
|
64 |
history[-1][1] += response.token.text
|
65 |
yield history
|
66 |
|
67 |
+
def retry_query(history):
|
68 |
+
history,
|
69 |
+
lookback = 3,
|
70 |
+
max_new_tokens = 256,
|
71 |
+
):
|
72 |
+
|
73 |
+
query = history[-1][0]
|
74 |
+
history[-1][1] = None
|
75 |
+
|
76 |
+
generateKwargs = dict(
|
77 |
+
max_new_tokens = max_new_tokens,
|
78 |
+
seed = 1337,
|
79 |
+
)
|
80 |
+
|
81 |
+
formatted_query = format_prompt(query, history, lookback)
|
82 |
+
|
83 |
+
stream = client.text_generation(
|
84 |
+
formatted_query,
|
85 |
+
**generateKwargs,
|
86 |
+
stream = True,
|
87 |
+
details = True,
|
88 |
+
return_full_text = False
|
89 |
+
)
|
90 |
+
|
91 |
+
history[-1][1] = ""
|
92 |
+
|
93 |
+
for response in stream:
|
94 |
+
history[-1][1] += response.token.text
|
95 |
+
yield history
|
96 |
+
|
97 |
|
98 |
"""
|
99 |
Chat UI using Gradio Blocks.
|
100 |
|
101 |
+
Blocks preferred for "lower-level" layout control and state management.
|
102 |
|
103 |
TODOs:
|
104 |
- State management for dynamic components update.
|
105 |
+
- Add scratchpad readout to the right of chat log.
|
106 |
* Placeholder added for now.
|
107 |
- Add functionality to retry button.
|
108 |
* Placeholder added for now.
|
109 |
- Add dropdown for model selection.
|
110 |
+
* Placeholder added for now.
|
111 |
|
112 |
"""
|
113 |
|
114 |
with gr.Blocks() as chatUI:
|
115 |
+
# gr.State()
|
116 |
+
|
117 |
+
with gr.Row():
|
118 |
+
modelSelect = gr.Dropdown(
|
119 |
+
label = "Model selection:",
|
120 |
+
scale = 0.5,
|
121 |
+
)
|
122 |
+
|
123 |
with gr.Row():
|
124 |
chatOutput = gr.Chatbot(
|
125 |
bubble_full_width = False,
|
|
|
136 |
submitButton = gr.Button("Submit", scale = 1)
|
137 |
|
138 |
with gr.Row():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
139 |
fileUpload = gr.File(
|
140 |
height = 100,
|
141 |
)
|
142 |
+
retryButton = gr.Button("Retry")
|
143 |
+
clearButton = gr.ClearButton([queryInput, chatOutput])
|
144 |
|
145 |
with gr.Row():
|
146 |
with gr.Accordion(label = "Expand for edit system prompt:"):
|
|
|
151 |
scale = 4,
|
152 |
)
|
153 |
|
154 |
+
|
155 |
+
"""
|
156 |
+
Event functions
|
|
|
157 |
|
158 |
+
"""
|
159 |
queryInput.submit(
|
160 |
fn = query_submit,
|
161 |
inputs = [queryInput, chatOutput],
|
|
|
178 |
outputs = [chatOutput],
|
179 |
)
|
180 |
|
181 |
+
retryButton.click(
|
182 |
+
fn = retry_query,
|
183 |
+
inputs = [chatOutput],
|
184 |
+
outputs = [chatOutput],
|
185 |
+
)
|
186 |
+
|
187 |
+
|
188 |
chatUI.queue()
|
189 |
chatUI.launch(show_api = False)
|