Minor fixes to presets
Browse files- app.py +1 -0
- presets.py +6 -5
app.py
CHANGED
@@ -673,6 +673,7 @@ with gr.Blocks(css=custom_css) as demo:
|
|
673 |
outputs=[
|
674 |
model_name_or_path,
|
675 |
input_template,
|
|
|
676 |
contextless_input_template,
|
677 |
special_tokens_to_keep,
|
678 |
generation_kwargs,
|
|
|
673 |
outputs=[
|
674 |
model_name_or_path,
|
675 |
input_template,
|
676 |
+
decoder_input_output_separator,
|
677 |
contextless_input_template,
|
678 |
special_tokens_to_keep,
|
679 |
generation_kwargs,
|
presets.py
CHANGED
@@ -77,17 +77,18 @@ def set_towerinstruct_preset():
|
|
77 |
def set_gemma_preset():
|
78 |
return (
|
79 |
"google/gemma-2b-it", # model_name_or_path
|
80 |
-
"<start_of_turn>user\n{context}\n{current}<end_of_turn>\n<start_of_turn>model
|
81 |
-
"
|
|
|
82 |
["<start_of_turn>", "<end_of_turn>"], # special_tokens_to_keep
|
83 |
'{\n\t"max_new_tokens": 50\n}', # generation_kwargs
|
84 |
)
|
85 |
|
86 |
def set_mistral_instruct_preset():
|
87 |
return (
|
88 |
-
"mistralai/Mistral-7B-Instruct-v0.2" # model_name_or_path
|
89 |
-
"[INST]{context}\n{current}[/INST]" # input_template
|
90 |
-
"[INST]{current}[/INST]" # input_current_text_template
|
91 |
'{\n\t"max_new_tokens": 50\n}', # generation_kwargs
|
92 |
)
|
93 |
|
|
|
77 |
def set_gemma_preset():
|
78 |
return (
|
79 |
"google/gemma-2b-it", # model_name_or_path
|
80 |
+
"<start_of_turn>user\n{context}\n{current}<end_of_turn>\n<start_of_turn>model", # input_template
|
81 |
+
"\n", # decoder_input_output_separator
|
82 |
+
"<start_of_turn>user\n{current}<end_of_turn>\n<start_of_turn>model", # input_current_text_template
|
83 |
["<start_of_turn>", "<end_of_turn>"], # special_tokens_to_keep
|
84 |
'{\n\t"max_new_tokens": 50\n}', # generation_kwargs
|
85 |
)
|
86 |
|
87 |
def set_mistral_instruct_preset():
|
88 |
return (
|
89 |
+
"mistralai/Mistral-7B-Instruct-v0.2", # model_name_or_path
|
90 |
+
"[INST]{context}\n{current}[/INST]", # input_template
|
91 |
+
"[INST]{current}[/INST]", # input_current_text_template
|
92 |
'{\n\t"max_new_tokens": 50\n}', # generation_kwargs
|
93 |
)
|
94 |
|