Add presets and model preloading
Browse files- app.py +241 -58
- contents.py +1 -1
- presets.py +58 -0
- utils.py +1 -14
app.py
CHANGED
|
@@ -13,14 +13,25 @@ from contents import (
|
|
| 13 |
title,
|
| 14 |
)
|
| 15 |
from gradio_highlightedtextbox import HighlightedTextbox
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
from style import custom_css
|
| 17 |
-
from utils import
|
| 18 |
|
| 19 |
-
from inseq import list_feature_attribution_methods, list_step_functions
|
| 20 |
from inseq.commands.attribute_context.attribute_context import (
|
| 21 |
AttributeContextArgs,
|
| 22 |
-
|
| 23 |
)
|
|
|
|
|
|
|
|
|
|
| 24 |
|
| 25 |
|
| 26 |
@spaces.GPU()
|
|
@@ -38,17 +49,41 @@ def pecore(
|
|
| 38 |
attribution_std_threshold: float,
|
| 39 |
attribution_topk: int,
|
| 40 |
input_template: str,
|
| 41 |
-
|
| 42 |
output_template: str,
|
| 43 |
special_tokens_to_keep: str | list[str] | None,
|
|
|
|
| 44 |
model_kwargs: str,
|
| 45 |
tokenizer_kwargs: str,
|
| 46 |
generation_kwargs: str,
|
| 47 |
attribution_kwargs: str,
|
| 48 |
):
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
pecore_args = AttributeContextArgs(
|
| 53 |
show_intermediate_outputs=False,
|
| 54 |
save_path=os.path.join(os.path.dirname(__file__), "outputs/output.json"),
|
|
@@ -66,24 +101,41 @@ def pecore(
|
|
| 66 |
generation_kwargs=json.loads(generation_kwargs),
|
| 67 |
attribution_kwargs=json.loads(attribution_kwargs),
|
| 68 |
context_sensitivity_metric=context_sensitivity_metric,
|
| 69 |
-
align_output_context_auto=False,
|
| 70 |
prompt_user_for_contextless_output_next_tokens=False,
|
| 71 |
special_tokens_to_keep=special_tokens_to_keep,
|
| 72 |
context_sensitivity_std_threshold=context_sensitivity_std_threshold,
|
| 73 |
-
context_sensitivity_topk=context_sensitivity_topk
|
| 74 |
-
if context_sensitivity_topk > 0
|
| 75 |
-
else None,
|
| 76 |
attribution_std_threshold=attribution_std_threshold,
|
| 77 |
-
|
| 78 |
-
input_current_text=formatted_input_current_text,
|
| 79 |
-
input_context_text=input_context_text if input_context_text else None,
|
| 80 |
input_template=input_template,
|
| 81 |
-
output_current_text=output_current_text if output_current_text else None,
|
| 82 |
-
output_context_text=output_context_text if output_context_text else None,
|
| 83 |
output_template=output_template,
|
|
|
|
|
|
|
|
|
|
| 84 |
)
|
| 85 |
-
out =
|
| 86 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 87 |
|
| 88 |
|
| 89 |
with gr.Blocks(css=custom_css) as demo:
|
|
@@ -93,12 +145,12 @@ with gr.Blocks(css=custom_css) as demo:
|
|
| 93 |
with gr.Tab("π Attributing Context"):
|
| 94 |
with gr.Row():
|
| 95 |
with gr.Column():
|
| 96 |
-
input_current_text = gr.Textbox(
|
| 97 |
-
label="Input query", placeholder="Your input query..."
|
| 98 |
-
)
|
| 99 |
input_context_text = gr.Textbox(
|
| 100 |
label="Input context", lines=4, placeholder="Your input context..."
|
| 101 |
)
|
|
|
|
|
|
|
|
|
|
| 102 |
attribute_input_button = gr.Button("Submit", variant="primary")
|
| 103 |
with gr.Column():
|
| 104 |
pecore_output_highlights = HighlightedTextbox(
|
|
@@ -139,15 +191,57 @@ with gr.Blocks(css=custom_css) as demo:
|
|
| 139 |
inputs=[input_current_text, input_context_text],
|
| 140 |
outputs=pecore_output_highlights,
|
| 141 |
)
|
| 142 |
-
with gr.Tab("βοΈ Parameters"):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 143 |
gr.Markdown("## βοΈ PECoRe Parameters")
|
| 144 |
with gr.Row(equal_height=True):
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 151 |
context_sensitivity_metric = gr.Dropdown(
|
| 152 |
value="kl_divergence",
|
| 153 |
label="Context sensitivity metric",
|
|
@@ -224,12 +318,13 @@ with gr.Blocks(css=custom_css) as demo:
|
|
| 224 |
info="Template to format the output from the model. Use {current} and {context} placeholders.",
|
| 225 |
interactive=True,
|
| 226 |
)
|
| 227 |
-
|
| 228 |
value="<Q>:{current}",
|
| 229 |
label="Input current text template",
|
| 230 |
info="Template to format the input query for the model. Use {current} placeholder.",
|
| 231 |
interactive=True,
|
| 232 |
)
|
|
|
|
| 233 |
special_tokens_to_keep = gr.Dropdown(
|
| 234 |
label="Special tokens to keep",
|
| 235 |
info="Special tokens to keep in the attribution. If empty, all special tokens are ignored.",
|
|
@@ -237,8 +332,28 @@ with gr.Blocks(css=custom_css) as demo:
|
|
| 237 |
multiselect=True,
|
| 238 |
allow_custom_value=True,
|
| 239 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 240 |
|
| 241 |
gr.Markdown("## βοΈ Generation Parameters")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 242 |
with gr.Row(equal_height=True):
|
| 243 |
output_current_text = gr.Textbox(
|
| 244 |
label="Generation output",
|
|
@@ -250,36 +365,37 @@ with gr.Blocks(css=custom_css) as demo:
|
|
| 250 |
info="If specified, this context is used as starting point for generation. Useful for e.g. chain-of-thought reasoning.",
|
| 251 |
interactive=True,
|
| 252 |
)
|
| 253 |
-
generation_kwargs = gr.Code(
|
| 254 |
-
value="{}",
|
| 255 |
-
language="json",
|
| 256 |
-
label="Generation kwargs",
|
| 257 |
-
interactive=True,
|
| 258 |
-
lines=1,
|
| 259 |
-
)
|
| 260 |
gr.Markdown("## βοΈ Other Parameters")
|
| 261 |
with gr.Row(equal_height=True):
|
| 262 |
-
|
| 263 |
-
|
| 264 |
-
|
| 265 |
-
|
| 266 |
-
|
| 267 |
-
|
| 268 |
-
|
| 269 |
-
|
| 270 |
-
|
| 271 |
-
|
| 272 |
-
|
| 273 |
-
|
| 274 |
-
|
| 275 |
-
)
|
| 276 |
-
|
| 277 |
-
|
| 278 |
-
|
| 279 |
-
|
| 280 |
-
|
| 281 |
-
|
| 282 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 283 |
|
| 284 |
gr.Markdown(how_it_works)
|
| 285 |
gr.Markdown(how_to_use)
|
|
@@ -301,9 +417,10 @@ with gr.Blocks(css=custom_css) as demo:
|
|
| 301 |
attribution_std_threshold,
|
| 302 |
attribution_topk,
|
| 303 |
input_template,
|
| 304 |
-
|
| 305 |
output_template,
|
| 306 |
special_tokens_to_keep,
|
|
|
|
| 307 |
model_kwargs,
|
| 308 |
tokenizer_kwargs,
|
| 309 |
generation_kwargs,
|
|
@@ -316,4 +433,70 @@ with gr.Blocks(css=custom_css) as demo:
|
|
| 316 |
],
|
| 317 |
)
|
| 318 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 319 |
demo.launch(allowed_paths=["outputs/"])
|
|
|
|
| 13 |
title,
|
| 14 |
)
|
| 15 |
from gradio_highlightedtextbox import HighlightedTextbox
|
| 16 |
+
from presets import (
|
| 17 |
+
set_chatml_preset,
|
| 18 |
+
set_cora_preset,
|
| 19 |
+
set_default_preset,
|
| 20 |
+
set_mmt_preset,
|
| 21 |
+
set_towerinstruct_preset,
|
| 22 |
+
set_zephyr_preset,
|
| 23 |
+
)
|
| 24 |
from style import custom_css
|
| 25 |
+
from utils import get_formatted_attribute_context_results
|
| 26 |
|
| 27 |
+
from inseq import list_feature_attribution_methods, list_step_functions, load_model
|
| 28 |
from inseq.commands.attribute_context.attribute_context import (
|
| 29 |
AttributeContextArgs,
|
| 30 |
+
attribute_context_with_model,
|
| 31 |
)
|
| 32 |
+
from inseq.models import HuggingfaceModel
|
| 33 |
+
|
| 34 |
+
loaded_model: HuggingfaceModel = None
|
| 35 |
|
| 36 |
|
| 37 |
@spaces.GPU()
|
|
|
|
| 49 |
attribution_std_threshold: float,
|
| 50 |
attribution_topk: int,
|
| 51 |
input_template: str,
|
| 52 |
+
contextless_input_current_text: str,
|
| 53 |
output_template: str,
|
| 54 |
special_tokens_to_keep: str | list[str] | None,
|
| 55 |
+
decoder_input_output_separator: str,
|
| 56 |
model_kwargs: str,
|
| 57 |
tokenizer_kwargs: str,
|
| 58 |
generation_kwargs: str,
|
| 59 |
attribution_kwargs: str,
|
| 60 |
):
|
| 61 |
+
global loaded_model
|
| 62 |
+
if "{context}" in output_template and not output_context_text:
|
| 63 |
+
raise gr.Error(
|
| 64 |
+
"Parameter 'Generated context' is required when using {context} in the output template."
|
| 65 |
+
)
|
| 66 |
+
if loaded_model is None or model_name_or_path != loaded_model.model_name:
|
| 67 |
+
gr.Info("Loading model...")
|
| 68 |
+
loaded_model = load_model(
|
| 69 |
+
model_name_or_path,
|
| 70 |
+
attribution_method,
|
| 71 |
+
model_kwargs=json.loads(model_kwargs),
|
| 72 |
+
tokenizer_kwargs=json.loads(tokenizer_kwargs),
|
| 73 |
+
)
|
| 74 |
+
kwargs = {}
|
| 75 |
+
if context_sensitivity_topk > 0:
|
| 76 |
+
kwargs["context_sensitivity_topk"] = context_sensitivity_topk
|
| 77 |
+
if attribution_topk > 0:
|
| 78 |
+
kwargs["attribution_topk"] = attribution_topk
|
| 79 |
+
if input_context_text:
|
| 80 |
+
kwargs["input_context_text"] = input_context_text
|
| 81 |
+
if output_context_text:
|
| 82 |
+
kwargs["output_context_text"] = output_context_text
|
| 83 |
+
if output_current_text:
|
| 84 |
+
kwargs["output_current_text"] = output_current_text
|
| 85 |
+
if decoder_input_output_separator:
|
| 86 |
+
kwargs["decoder_input_output_separator"] = decoder_input_output_separator
|
| 87 |
pecore_args = AttributeContextArgs(
|
| 88 |
show_intermediate_outputs=False,
|
| 89 |
save_path=os.path.join(os.path.dirname(__file__), "outputs/output.json"),
|
|
|
|
| 101 |
generation_kwargs=json.loads(generation_kwargs),
|
| 102 |
attribution_kwargs=json.loads(attribution_kwargs),
|
| 103 |
context_sensitivity_metric=context_sensitivity_metric,
|
|
|
|
| 104 |
prompt_user_for_contextless_output_next_tokens=False,
|
| 105 |
special_tokens_to_keep=special_tokens_to_keep,
|
| 106 |
context_sensitivity_std_threshold=context_sensitivity_std_threshold,
|
|
|
|
|
|
|
|
|
|
| 107 |
attribution_std_threshold=attribution_std_threshold,
|
| 108 |
+
input_current_text=input_current_text,
|
|
|
|
|
|
|
| 109 |
input_template=input_template,
|
|
|
|
|
|
|
| 110 |
output_template=output_template,
|
| 111 |
+
contextless_input_current_text=contextless_input_current_text,
|
| 112 |
+
handle_output_context_strategy="pre",
|
| 113 |
+
**kwargs,
|
| 114 |
)
|
| 115 |
+
out = attribute_context_with_model(pecore_args, loaded_model)
|
| 116 |
+
tuples = get_formatted_attribute_context_results(loaded_model, out.info, out)
|
| 117 |
+
if not tuples:
|
| 118 |
+
msg = "Warning: No pairs were found by PECoRe. Try adjusting Results Selection parameters."
|
| 119 |
+
tuples = [(msg, None)]
|
| 120 |
+
return tuples, gr.Button(visible=True), gr.Button(visible=True)
|
| 121 |
+
|
| 122 |
+
|
| 123 |
+
@spaces.GPU()
|
| 124 |
+
def preload_model(
|
| 125 |
+
model_name_or_path: str,
|
| 126 |
+
attribution_method: str,
|
| 127 |
+
model_kwargs: str,
|
| 128 |
+
tokenizer_kwargs: str,
|
| 129 |
+
):
|
| 130 |
+
global loaded_model
|
| 131 |
+
if loaded_model is None or model_name_or_path != loaded_model.model_name:
|
| 132 |
+
gr.Info("Loading model...")
|
| 133 |
+
loaded_model = load_model(
|
| 134 |
+
model_name_or_path,
|
| 135 |
+
attribution_method,
|
| 136 |
+
model_kwargs=json.loads(model_kwargs),
|
| 137 |
+
tokenizer_kwargs=json.loads(tokenizer_kwargs),
|
| 138 |
+
)
|
| 139 |
|
| 140 |
|
| 141 |
with gr.Blocks(css=custom_css) as demo:
|
|
|
|
| 145 |
with gr.Tab("π Attributing Context"):
|
| 146 |
with gr.Row():
|
| 147 |
with gr.Column():
|
|
|
|
|
|
|
|
|
|
| 148 |
input_context_text = gr.Textbox(
|
| 149 |
label="Input context", lines=4, placeholder="Your input context..."
|
| 150 |
)
|
| 151 |
+
input_current_text = gr.Textbox(
|
| 152 |
+
label="Input query", placeholder="Your input query..."
|
| 153 |
+
)
|
| 154 |
attribute_input_button = gr.Button("Submit", variant="primary")
|
| 155 |
with gr.Column():
|
| 156 |
pecore_output_highlights = HighlightedTextbox(
|
|
|
|
| 191 |
inputs=[input_current_text, input_context_text],
|
| 192 |
outputs=pecore_output_highlights,
|
| 193 |
)
|
| 194 |
+
with gr.Tab("βοΈ Parameters") as params_tab:
|
| 195 |
+
gr.Markdown("## β¨ Presets")
|
| 196 |
+
with gr.Row(equal_height=True):
|
| 197 |
+
with gr.Column():
|
| 198 |
+
default_preset = gr.Button("Default", variant="secondary")
|
| 199 |
+
gr.Markdown(
|
| 200 |
+
"Default preset using templates without special tokens or parameters.\nCan be used with most decoder-only and encoder-decoder models."
|
| 201 |
+
)
|
| 202 |
+
with gr.Column():
|
| 203 |
+
cora_preset = gr.Button("CORA mQA", variant="secondary")
|
| 204 |
+
gr.Markdown(
|
| 205 |
+
"Preset for the <a href='https://huggingface.co/gsarti/cora_mgen' target='_blank'>CORA Multilingual QA</a> model.\nUses special templates for inputs."
|
| 206 |
+
)
|
| 207 |
+
with gr.Column():
|
| 208 |
+
zephyr_preset = gr.Button("Zephyr Template", variant="secondary")
|
| 209 |
+
gr.Markdown(
|
| 210 |
+
"Preset for models using the <a href='https://huggingface.co/HuggingFaceH4/zephyr-7b-beta' target='_blank'>Zephyr conversational template</a>.\nUses <code><|system|></code>, <code><|user|></code> and <code><|assistant|></code> special tokens."
|
| 211 |
+
)
|
| 212 |
+
with gr.Row(equal_height=True):
|
| 213 |
+
with gr.Column(scale=1):
|
| 214 |
+
multilingual_mt_template = gr.Button(
|
| 215 |
+
"Multilingual MT", variant="secondary"
|
| 216 |
+
)
|
| 217 |
+
gr.Markdown(
|
| 218 |
+
"Present for multilingual MT models such as <a href='https://huggingface.co/facebook/nllb-200-distilled-600M' target='_blank'>NLLB</a> and <a href='https://huggingface.co/facebook/mbart-large-50-many-to-many-mmt' target='_blank'>mBART</a> using language tags."
|
| 219 |
+
)
|
| 220 |
+
with gr.Column(scale=1):
|
| 221 |
+
chatml_template = gr.Button("ChatML Template", variant="secondary")
|
| 222 |
+
gr.Markdown(
|
| 223 |
+
"Preset for models using the <a href='https://github.com/MicrosoftDocs/azure-docs/blob/main/articles/ai-services/openai/includes/chat-markup-language.md' target='_blank'>ChatML conversational template</a>.\nUses <code><|im_start|></code>, <code><|im_end|></code> special tokens."
|
| 224 |
+
)
|
| 225 |
+
with gr.Column(scale=1):
|
| 226 |
+
towerinstruct_template = gr.Button(
|
| 227 |
+
"Unbabel TowerInstruct", variant="secondary"
|
| 228 |
+
)
|
| 229 |
+
gr.Markdown(
|
| 230 |
+
"Preset for models using the <a href='https://huggingface.co/Unbabel/TowerInstruct-7B-v0.1' target='_blank'>Unbabel TowerInstruct</a> conversational template.\nUses <code><|im_start|></code>, <code><|im_end|></code> special tokens."
|
| 231 |
+
)
|
| 232 |
gr.Markdown("## βοΈ PECoRe Parameters")
|
| 233 |
with gr.Row(equal_height=True):
|
| 234 |
+
with gr.Column():
|
| 235 |
+
model_name_or_path = gr.Textbox(
|
| 236 |
+
value="gpt2",
|
| 237 |
+
label="Model",
|
| 238 |
+
info="Hugging Face Hub identifier of the model to analyze with PECoRe.",
|
| 239 |
+
interactive=True,
|
| 240 |
+
)
|
| 241 |
+
load_model_button = gr.Button(
|
| 242 |
+
"Load model",
|
| 243 |
+
variant="secondary",
|
| 244 |
+
)
|
| 245 |
context_sensitivity_metric = gr.Dropdown(
|
| 246 |
value="kl_divergence",
|
| 247 |
label="Context sensitivity metric",
|
|
|
|
| 318 |
info="Template to format the output from the model. Use {current} and {context} placeholders.",
|
| 319 |
interactive=True,
|
| 320 |
)
|
| 321 |
+
contextless_input_current_text = gr.Textbox(
|
| 322 |
value="<Q>:{current}",
|
| 323 |
label="Input current text template",
|
| 324 |
info="Template to format the input query for the model. Use {current} placeholder.",
|
| 325 |
interactive=True,
|
| 326 |
)
|
| 327 |
+
with gr.Row(equal_height=True):
|
| 328 |
special_tokens_to_keep = gr.Dropdown(
|
| 329 |
label="Special tokens to keep",
|
| 330 |
info="Special tokens to keep in the attribution. If empty, all special tokens are ignored.",
|
|
|
|
| 332 |
multiselect=True,
|
| 333 |
allow_custom_value=True,
|
| 334 |
)
|
| 335 |
+
decoder_input_output_separator = gr.Textbox(
|
| 336 |
+
label="Decoder input/output separator",
|
| 337 |
+
info="Separator to use between input and output in the decoder input.",
|
| 338 |
+
value="",
|
| 339 |
+
interactive=True,
|
| 340 |
+
lines=1,
|
| 341 |
+
)
|
| 342 |
|
| 343 |
gr.Markdown("## βοΈ Generation Parameters")
|
| 344 |
+
with gr.Row(equal_height=True):
|
| 345 |
+
with gr.Column(scale=0.5):
|
| 346 |
+
gr.Markdown(
|
| 347 |
+
"The following arguments can be used to control generation parameters and force specific model outputs."
|
| 348 |
+
)
|
| 349 |
+
with gr.Column(scale=1):
|
| 350 |
+
generation_kwargs = gr.Code(
|
| 351 |
+
value="{}",
|
| 352 |
+
language="json",
|
| 353 |
+
label="Generation kwargs (JSON)",
|
| 354 |
+
interactive=True,
|
| 355 |
+
lines=1,
|
| 356 |
+
)
|
| 357 |
with gr.Row(equal_height=True):
|
| 358 |
output_current_text = gr.Textbox(
|
| 359 |
label="Generation output",
|
|
|
|
| 365 |
info="If specified, this context is used as starting point for generation. Useful for e.g. chain-of-thought reasoning.",
|
| 366 |
interactive=True,
|
| 367 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 368 |
gr.Markdown("## βοΈ Other Parameters")
|
| 369 |
with gr.Row(equal_height=True):
|
| 370 |
+
with gr.Column():
|
| 371 |
+
gr.Markdown(
|
| 372 |
+
"The following arguments will be passed to initialize the Hugging Face model and tokenizer, and to the `inseq_model.attribute` method."
|
| 373 |
+
)
|
| 374 |
+
with gr.Column():
|
| 375 |
+
model_kwargs = gr.Code(
|
| 376 |
+
value="{}",
|
| 377 |
+
language="json",
|
| 378 |
+
label="Model kwargs (JSON)",
|
| 379 |
+
interactive=True,
|
| 380 |
+
lines=1,
|
| 381 |
+
min_width=160,
|
| 382 |
+
)
|
| 383 |
+
with gr.Column():
|
| 384 |
+
tokenizer_kwargs = gr.Code(
|
| 385 |
+
value="{}",
|
| 386 |
+
language="json",
|
| 387 |
+
label="Tokenizer kwargs (JSON)",
|
| 388 |
+
interactive=True,
|
| 389 |
+
lines=1,
|
| 390 |
+
)
|
| 391 |
+
with gr.Column():
|
| 392 |
+
attribution_kwargs = gr.Code(
|
| 393 |
+
value="{}",
|
| 394 |
+
language="json",
|
| 395 |
+
label="Attribution kwargs (JSON)",
|
| 396 |
+
interactive=True,
|
| 397 |
+
lines=1,
|
| 398 |
+
)
|
| 399 |
|
| 400 |
gr.Markdown(how_it_works)
|
| 401 |
gr.Markdown(how_to_use)
|
|
|
|
| 417 |
attribution_std_threshold,
|
| 418 |
attribution_topk,
|
| 419 |
input_template,
|
| 420 |
+
contextless_input_current_text,
|
| 421 |
output_template,
|
| 422 |
special_tokens_to_keep,
|
| 423 |
+
decoder_input_output_separator,
|
| 424 |
model_kwargs,
|
| 425 |
tokenizer_kwargs,
|
| 426 |
generation_kwargs,
|
|
|
|
| 433 |
],
|
| 434 |
)
|
| 435 |
|
| 436 |
+
load_model_button.click(
|
| 437 |
+
preload_model,
|
| 438 |
+
inputs=[model_name_or_path, attribution_method, model_kwargs, tokenizer_kwargs],
|
| 439 |
+
outputs=[],
|
| 440 |
+
)
|
| 441 |
+
|
| 442 |
+
# Preset params
|
| 443 |
+
|
| 444 |
+
outputs_to_reset = [
|
| 445 |
+
model_name_or_path,
|
| 446 |
+
input_template,
|
| 447 |
+
contextless_input_current_text,
|
| 448 |
+
output_template,
|
| 449 |
+
special_tokens_to_keep,
|
| 450 |
+
decoder_input_output_separator,
|
| 451 |
+
model_kwargs,
|
| 452 |
+
tokenizer_kwargs,
|
| 453 |
+
generation_kwargs,
|
| 454 |
+
attribution_kwargs,
|
| 455 |
+
]
|
| 456 |
+
reset_kwargs = {
|
| 457 |
+
"fn": set_default_preset,
|
| 458 |
+
"inputs": None,
|
| 459 |
+
"outputs": outputs_to_reset,
|
| 460 |
+
}
|
| 461 |
+
|
| 462 |
+
# Presets
|
| 463 |
+
|
| 464 |
+
default_preset.click(**reset_kwargs)
|
| 465 |
+
cora_preset.click(**reset_kwargs).then(
|
| 466 |
+
set_cora_preset,
|
| 467 |
+
outputs=[model_name_or_path, input_template, contextless_input_current_text],
|
| 468 |
+
)
|
| 469 |
+
zephyr_preset.click(**reset_kwargs).then(
|
| 470 |
+
set_zephyr_preset,
|
| 471 |
+
outputs=[
|
| 472 |
+
model_name_or_path,
|
| 473 |
+
input_template,
|
| 474 |
+
contextless_input_current_text,
|
| 475 |
+
decoder_input_output_separator,
|
| 476 |
+
],
|
| 477 |
+
)
|
| 478 |
+
multilingual_mt_template.click(**reset_kwargs).then(
|
| 479 |
+
set_mmt_preset,
|
| 480 |
+
outputs=[model_name_or_path, input_template, output_template, tokenizer_kwargs],
|
| 481 |
+
)
|
| 482 |
+
chatml_template.click(**reset_kwargs).then(
|
| 483 |
+
set_chatml_preset,
|
| 484 |
+
outputs=[
|
| 485 |
+
model_name_or_path,
|
| 486 |
+
input_template,
|
| 487 |
+
contextless_input_current_text,
|
| 488 |
+
decoder_input_output_separator,
|
| 489 |
+
special_tokens_to_keep,
|
| 490 |
+
],
|
| 491 |
+
)
|
| 492 |
+
towerinstruct_template.click(**reset_kwargs).then(
|
| 493 |
+
set_towerinstruct_preset,
|
| 494 |
+
outputs=[
|
| 495 |
+
model_name_or_path,
|
| 496 |
+
input_template,
|
| 497 |
+
contextless_input_current_text,
|
| 498 |
+
decoder_input_output_separator,
|
| 499 |
+
],
|
| 500 |
+
)
|
| 501 |
+
|
| 502 |
demo.launch(allowed_paths=["outputs/"])
|
contents.py
CHANGED
|
@@ -48,6 +48,6 @@ citation = r"""
|
|
| 48 |
examples = [
|
| 49 |
[
|
| 50 |
"When was Banff National Park established?",
|
| 51 |
-
"Banff National Park is Canada's oldest national park, established in 1885 as Rocky Mountains Park. Located in Alberta's Rocky Mountains, 110
|
| 52 |
]
|
| 53 |
]
|
|
|
|
| 48 |
examples = [
|
| 49 |
[
|
| 50 |
"When was Banff National Park established?",
|
| 51 |
+
"Banff National Park is Canada's oldest national park, established in 1885 as Rocky Mountains Park. Located in Alberta's Rocky Mountains, 110-180 kilometres (68-112 mi) west of Calgary, Banff encompasses 6,641 square kilometres (2,564 sq mi) of mountainous terrain.",
|
| 52 |
]
|
| 53 |
]
|
presets.py
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
def set_cora_preset():
|
| 2 |
+
return (
|
| 3 |
+
"gsarti/cora_mgen", # model_name_or_path
|
| 4 |
+
"<Q>:{current} <P>:{context}", # input_template
|
| 5 |
+
"<Q>:{current}", # input_current_text_template
|
| 6 |
+
)
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
def set_default_preset():
|
| 10 |
+
return (
|
| 11 |
+
"gpt2", # model_name_or_path
|
| 12 |
+
"{current} {context}", # input_template
|
| 13 |
+
"{current}", # input_current_template
|
| 14 |
+
"{current}", # output_template
|
| 15 |
+
[], # special_tokens_to_keep
|
| 16 |
+
"", # decoder_input_output_separator
|
| 17 |
+
"{}", # model_kwargs
|
| 18 |
+
"{}", # tokenizer_kwargs
|
| 19 |
+
"{}", # generation_kwargs
|
| 20 |
+
"{}", # attribution_kwargs
|
| 21 |
+
)
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
def set_zephyr_preset():
|
| 25 |
+
return (
|
| 26 |
+
"stabilityai/stablelm-2-zephyr-1_6b", # model_name_or_path
|
| 27 |
+
"<|system|>\n{context}</s>\n<|user|>\n{current}</s>\n<|assistant|>\n", # input_template
|
| 28 |
+
"<|user|>\n{current}</s>\n<|assistant|>\n", # input_current_text_template
|
| 29 |
+
"\n", # decoder_input_output_separator
|
| 30 |
+
)
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
def set_chatml_preset():
|
| 34 |
+
return (
|
| 35 |
+
"Qwen/Qwen1.5-0.5B-Chat", # model_name_or_path
|
| 36 |
+
"<|im_start|>system\n{context}<|im_end|>\n<|im_start|>user\n{current}<|im_end|>\n<|im_start|>assistant\n", # input_template
|
| 37 |
+
"<|im_start|>user\n{current}<|im_end|>\n<|im_start|>assistant\n", # input_current_text_template
|
| 38 |
+
"", # decoder_input_output_separator
|
| 39 |
+
["<|im_start|>", "<|im_end|>"], # special_tokens_to_keep
|
| 40 |
+
)
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
def set_mmt_preset():
|
| 44 |
+
return (
|
| 45 |
+
"facebook/mbart-large-50-one-to-many-mmt", # model_name_or_path
|
| 46 |
+
"{context} {current}", # input_template
|
| 47 |
+
"{context} {current}", # output_template
|
| 48 |
+
'{\n\t"src_lang": "en_XX",\n\t"tgt_lang": "fr_XX"\n}', # tokenizer_kwargs
|
| 49 |
+
)
|
| 50 |
+
|
| 51 |
+
|
| 52 |
+
def set_towerinstruct_preset():
|
| 53 |
+
return (
|
| 54 |
+
"Unbabel/TowerInstruct-7B-v0.1", # model_name_or_path
|
| 55 |
+
"<|im_start|>user\nSource: {current}\nContext: {context}\nTranslate the above text into French. Use the context to guide your answer.\nTarget:<|im_end|>\n<|im_start|>assistant\n", # input_template
|
| 56 |
+
"<|im_start|>user\nSource: {current}\nTranslate the above text into French.\nTarget:<|im_end|>\n<|im_start|>assistant\n", # input_current_text_template
|
| 57 |
+
"", # decoder_input_output_separator
|
| 58 |
+
)
|
utils.py
CHANGED
|
@@ -1,7 +1,5 @@
|
|
| 1 |
-
from copy import deepcopy
|
| 2 |
from typing import Optional
|
| 3 |
|
| 4 |
-
from inseq import load_model
|
| 5 |
from inseq.commands.attribute_context.attribute_context_args import AttributeContextArgs
|
| 6 |
from inseq.commands.attribute_context.attribute_context_helpers import (
|
| 7 |
AttributeContextOutput,
|
|
@@ -81,7 +79,6 @@ def get_formatted_attribute_context_results(
|
|
| 81 |
cci_out.output_context_scores,
|
| 82 |
cci_out.input_context_scores,
|
| 83 |
is_target=True,
|
| 84 |
-
context_type="Output",
|
| 85 |
)
|
| 86 |
out += [
|
| 87 |
("\n\n" if example_idx > 1 else "", None),
|
|
@@ -95,16 +92,6 @@ def get_formatted_attribute_context_results(
|
|
| 95 |
out += [("\nInput context:\t", None)]
|
| 96 |
out += input_context_tokens
|
| 97 |
if args.has_output_context:
|
| 98 |
-
out += [("
|
| 99 |
out += output_context_tokens
|
| 100 |
return out
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
def get_tuples_from_output(output: AttributeContextOutput):
|
| 104 |
-
model = load_model(
|
| 105 |
-
output.info.model_name_or_path,
|
| 106 |
-
output.info.attribution_method,
|
| 107 |
-
model_kwargs=deepcopy(output.info.model_kwargs),
|
| 108 |
-
tokenizer_kwargs=deepcopy(output.info.tokenizer_kwargs),
|
| 109 |
-
)
|
| 110 |
-
return get_formatted_attribute_context_results(model, output.info, output)
|
|
|
|
|
|
|
| 1 |
from typing import Optional
|
| 2 |
|
|
|
|
| 3 |
from inseq.commands.attribute_context.attribute_context_args import AttributeContextArgs
|
| 4 |
from inseq.commands.attribute_context.attribute_context_helpers import (
|
| 5 |
AttributeContextOutput,
|
|
|
|
| 79 |
cci_out.output_context_scores,
|
| 80 |
cci_out.input_context_scores,
|
| 81 |
is_target=True,
|
|
|
|
| 82 |
)
|
| 83 |
out += [
|
| 84 |
("\n\n" if example_idx > 1 else "", None),
|
|
|
|
| 92 |
out += [("\nInput context:\t", None)]
|
| 93 |
out += input_context_tokens
|
| 94 |
if args.has_output_context:
|
| 95 |
+
out += [("\nOutput context:\t", None)]
|
| 96 |
out += output_context_tokens
|
| 97 |
return out
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|