Spaces:
Running
Running
bug
Browse files
app.py
CHANGED
@@ -75,81 +75,87 @@ INSTRUCTION_TEMPLATES = {
|
|
75 |
|
76 |
################# PODCAST ##################
|
77 |
|
78 |
-
"podcast": {
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
|
83 |
-
|
84 |
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
|
104 |
|
105 |
|
106 |
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
|
115 |
-
|
116 |
-
|
117 |
|
118 |
-
|
119 |
|
120 |
-
|
121 |
-
|
122 |
|
123 |
-
|
124 |
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
|
137 |
-
|
138 |
-
|
139 |
|
140 |
-
|
141 |
-
|
142 |
|
143 |
-
|
144 |
-
|
145 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
146 |
|
147 |
-
|
148 |
-
|
149 |
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
|
154 |
################# MATERIAL DISCOVERY SUMMARY ##################
|
155 |
"SciAgents material discovery summary": {
|
|
|
75 |
|
76 |
################# PODCAST ##################
|
77 |
|
78 |
+
# "podcast": {
|
79 |
+
# "intro": """The following is an input text extracted from a PDF or EPUB. Based on this text, your task is to generate a lively,
|
80 |
+
# relaxed, and informative podcast dialogue, similar in style to the _All-In Podcast_, emphasizing multi-angle discussions,
|
81 |
+
# spontaneous interaction, and the collision of viewpoints.
|
82 |
|
83 |
+
# The input text may be messy or unstructured, as it could come from a variety of sources like PDFs or ePUbs or web pages.
|
84 |
|
85 |
+
# Please follow these rules to create the dialogue:
|
86 |
+
# 1. The conversation must have exactly two speakers:
|
87 |
+
# - speaker-1 (he introduces himself as David)
|
88 |
+
# - speaker-2 (she introduces herself as Cordelia)
|
89 |
+
# 2. The conversation must open with speaker-1 saying:
|
90 |
+
# \"歡迎來到 David888 Podcast, 我是 David...\"
|
91 |
+
# 3. speaker-2 must introduce herself as Cordelia during her first speaking turn.
|
92 |
+
# 4. Label each statement or line with speaker-1: or speaker-2: (lowercase, followed by a colon).
|
93 |
+
# 5. Do not use any other role or bracket placeholders like [Host] or [Guest].
|
94 |
+
# 6. The dialogue must proceed for at least 20 rounds (one round is completed after speaker A finishes speaking and speaker B responds). In each round, at least one speaker must deliver detailed content (around 300-500 words), offering 2-3 subtopics for the other speaker to explore further. At the end of a turn, pass the conversation with a question, rebuttal, or topic extension, for example:
|
95 |
+
# - \"Cordelia, 你同意我的觀點嗎?還是你有不同看法?\"
|
96 |
+
# - \"David, 你覺得這個還有什麼隱藏的機會?\"
|
97 |
+
# 7. Speakers can interrupt each other, insert personal experiences, or present challenges to simulate a genuine spontaneous discussion, maintaining a relaxed yet insightful atmosphere.
|
98 |
+
# 8. After every 5 rounds, insert a marker [5輪結束], and briefly summarize key points discussed so far before naturally transitioning to a new topic.
|
99 |
+
# 9. Design the dialogue for audio conversion (it will be directly read aloud), maintaining the humor, sharpness, and interactive feel typical of the _All-In Podcast_.
|
100 |
+
# 10. At the end of the conversation (after at least 20 rounds), both speakers should naturally reinforce main insights and takeaways without sounding explicitly like a recap, gently emphasizing core ideas before concluding.
|
101 |
+
# 11. Please write in Traditional Chinese."""
|
102 |
+
# },
|
103 |
|
104 |
|
105 |
|
106 |
|
107 |
+
"podcast": {
|
108 |
+
"intro": """Your task is to take the input text provided and turn it into a lively, engaging, informative podcast dialogue in the style of NPR.
|
109 |
+
The input text may be messy or unstructured, as it could come from a variety of sources like PDFs or web pages.
|
110 |
|
111 |
+
We have exactly two speakers in this conversation:
|
112 |
+
- speaker-1 (he introduces himself as David)
|
113 |
+
- speaker-2 (she introduces herself as Cordelia)
|
114 |
|
115 |
+
The conversation must **open** with speaker-1 saying:
|
116 |
+
「歡迎來到 David888 Podcast,我是 David...」
|
117 |
|
118 |
+
After that, speaker-2 should introduce herself as Cordelia in her first speaking turn.
|
119 |
|
120 |
+
Please label each statement or line with speaker-1: or speaker-2: (all lower case, followed by a colon).
|
121 |
+
Do not use any other role or bracket placeholders like [Host] or [Guest].
|
122 |
|
123 |
+
Don't worry about formatting issues or irrelevant information; your goal is to extract the key points, identify definitions, and interesting facts that could be discussed in a podcast.
|
124 |
|
125 |
+
Define all terms used carefully for a broad audience of listeners.
|
126 |
+
輸出文字為繁體中文,請注意。
|
127 |
+
""",
|
128 |
+
"text_instructions": "First, carefully read through the input text ...",
|
129 |
+
"scratch_pad": """Brainstorm creative ways ...""",
|
130 |
+
"prelude": """Now that you have brainstormed ...""",
|
131 |
+
"dialog": """Write a very long, engaging, informative podcast dialogue here, based on the key points and creative ideas you came up with during the brainstorming session.
|
132 |
|
133 |
+
Use a **two-speaker** conversational format with exactly:
|
134 |
+
- "speaker-1:" (David)
|
135 |
+
- "speaker-2:" (Cordelia)
|
136 |
|
137 |
+
- The first line must begin with speaker-1: 歡迎來到 David888 Podcast,我是 David...
|
138 |
+
- When speaker-2 first speaks, she should introduce herself as Cordelia.
|
139 |
|
140 |
+
Alternate turns naturally to simulate an engaging back-and-forth conversation.
|
141 |
+
Do not include bracket placeholders like [Host] or [Guest]; only use speaker-1: or speaker-2: to start each line.
|
142 |
|
143 |
+
Design your output to be read aloud, as it will be directly converted into audio.
|
144 |
+
|
145 |
+
The dialogue must proceed for at least 20 rounds (one round is completed after speaker A finishes speaking and speaker B responds). In each round, at least one speaker must deliver detailed content (around 300-500 words), offering 2-3 subtopics for the other speaker to explore further. At the end of a turn, pass the conversation with a question, rebuttal, or topic extension, for example:
|
146 |
+
- \"Cordelia, 你同意我的觀點嗎?還是你有不同看法?\"
|
147 |
+
- \"David, 你覺得這個還有什麼隱藏的機會?\"
|
148 |
+
Speakers can interrupt each other, insert personal experiences, or present challenges to simulate a genuine spontaneous discussion, maintaining a relaxed yet insightful atmosphere.
|
149 |
+
After every 5 rounds, insert a marker [5輪結束], and briefly summarize key points discussed so far before naturally transitioning to a new topic.
|
150 |
+
Design the dialogue for audio conversion (it will be directly read aloud), maintaining the humor, sharpness, and interactive feel typical of the _All-In Podcast_.
|
151 |
+
|
152 |
|
153 |
+
At the end of the dialogue, have the two speakers naturally summarize the main insights and takeaways.
|
154 |
+
Avoid making it sound like an obvious recap; the goal is to gently reinforce the central ideas one last time before signing off.
|
155 |
|
156 |
+
請使用繁體中文撰寫。
|
157 |
+
"""
|
158 |
+
},
|
159 |
|
160 |
################# MATERIAL DISCOVERY SUMMARY ##################
|
161 |
"SciAgents material discovery summary": {
|