Spaces:
Running
on
Zero
Running
on
Zero
WillSH97
commited on
:memo: trial push-to-hub
Browse files- context_window_gradio.py +1 -1
context_window_gradio.py
CHANGED
|
@@ -72,7 +72,7 @@ def llama32_1b_chat(message, history, context_window) -> str:
|
|
| 72 |
def create_interface():
|
| 73 |
|
| 74 |
with gr.Blocks() as demo:
|
| 75 |
-
gr.Markdown("""DRAFT
|
| 76 |
NOTE: the strategy being used to generate outputs based on the simulated context window is to simply delete entire messages from earlier in the conversation until the input fits within the context window. There are several more sophisticated ways in which long chat histories can be dealt with (e.g. summarising the history, deleting messages but keeping keywords/concepts, other clever things), but this is the simplest, and potentially most illustrative of the shortcomings of having a limited context window.
|
| 77 |
|
| 78 |
""")
|
|
|
|
| 72 |
def create_interface():
|
| 73 |
|
| 74 |
with gr.Blocks() as demo:
|
| 75 |
+
gr.Markdown("""DRAFT!: change context window lmao.
|
| 76 |
NOTE: the strategy being used to generate outputs based on the simulated context window is to simply delete entire messages from earlier in the conversation until the input fits within the context window. There are several more sophisticated ways in which long chat histories can be dealt with (e.g. summarising the history, deleting messages but keeping keywords/concepts, other clever things), but this is the simplest, and potentially most illustrative of the shortcomings of having a limited context window.
|
| 77 |
|
| 78 |
""")
|