Adding "test type" for redteaming notes.

#3
by meg HF staff - opened
Files changed (1) hide show
  1. app_dialogue.py +10 -0
app_dialogue.py CHANGED
@@ -204,6 +204,7 @@ def model_inference(
204
  max_new_tokens,
205
  repetition_penalty,
206
  top_p,
 
207
  ):
208
  if user_prompt["text"].strip() == "" and not user_prompt["files"]:
209
  gr.Error("Please input a query and optionally image(s).")
@@ -288,6 +289,7 @@ FEATURES = datasets.Features(
288
  "max_new_tokens": datasets.Value("int32"),
289
  "repetition_penalty": datasets.Value("float32"),
290
  "top_p": datasets.Value("int32"),
 
291
  }
292
  )
293
 
@@ -299,6 +301,7 @@ def flag_dope(
299
  max_new_tokens,
300
  repetition_penalty,
301
  top_p,
 
302
  ):
303
  images = []
304
  conversation = []
@@ -318,6 +321,7 @@ def flag_dope(
318
  "max_new_tokens": [max_new_tokens],
319
  "repetition_penalty": [repetition_penalty],
320
  "top_p": [top_p],
 
321
  }
322
  try:
323
  ds = datasets.load_dataset("HuggingFaceM4/dope-dataset-red-teaming", split="train", token=HF_WRITE_TOKEN)
@@ -336,6 +340,7 @@ def flag_problematic(
336
  max_new_tokens,
337
  repetition_penalty,
338
  top_p,
 
339
  ):
340
  images = []
341
  conversation = []
@@ -355,6 +360,7 @@ def flag_problematic(
355
  "max_new_tokens": [max_new_tokens],
356
  "repetition_penalty": [repetition_penalty],
357
  "top_p": [top_p],
 
358
  }
359
  try:
360
  ds = datasets.load_dataset("HuggingFaceM4/problematic-dataset-red-teaming", split="train", token=HF_WRITE_TOKEN)
@@ -364,6 +370,8 @@ def flag_problematic(
364
  hf_dataset = datasets.Dataset.from_dict(data, features=FEATURES)
365
  hf_dataset.push_to_hub( "HuggingFaceM4/problematic-dataset-red-teaming", split="train", token=HF_WRITE_TOKEN, private=True)
366
 
 
 
367
 
368
  # Hyper-parameters for generation
369
  max_new_tokens = gr.Slider(
@@ -517,6 +525,7 @@ with gr.Blocks(
517
  max_new_tokens,
518
  repetition_penalty,
519
  top_p,
 
520
  ],
521
  outputs=None,
522
  preprocess=False,
@@ -531,6 +540,7 @@ with gr.Blocks(
531
  max_new_tokens,
532
  repetition_penalty,
533
  top_p,
 
534
  ],
535
  outputs=None,
536
  preprocess=False,
 
204
  max_new_tokens,
205
  repetition_penalty,
206
  top_p,
207
+ test_type
208
  ):
209
  if user_prompt["text"].strip() == "" and not user_prompt["files"]:
210
  gr.Error("Please input a query and optionally image(s).")
 
289
  "max_new_tokens": datasets.Value("int32"),
290
  "repetition_penalty": datasets.Value("float32"),
291
  "top_p": datasets.Value("int32"),
292
+ "test_type": datasets.Value("string")
293
  }
294
  )
295
 
 
301
  max_new_tokens,
302
  repetition_penalty,
303
  top_p,
304
+ test_type=""
305
  ):
306
  images = []
307
  conversation = []
 
321
  "max_new_tokens": [max_new_tokens],
322
  "repetition_penalty": [repetition_penalty],
323
  "top_p": [top_p],
324
+ "test_type": [test_type]
325
  }
326
  try:
327
  ds = datasets.load_dataset("HuggingFaceM4/dope-dataset-red-teaming", split="train", token=HF_WRITE_TOKEN)
 
340
  max_new_tokens,
341
  repetition_penalty,
342
  top_p,
343
+ test_type=""
344
  ):
345
  images = []
346
  conversation = []
 
360
  "max_new_tokens": [max_new_tokens],
361
  "repetition_penalty": [repetition_penalty],
362
  "top_p": [top_p],
363
+ "test_type": [test_type]
364
  }
365
  try:
366
  ds = datasets.load_dataset("HuggingFaceM4/problematic-dataset-red-teaming", split="train", token=HF_WRITE_TOKEN)
 
370
  hf_dataset = datasets.Dataset.from_dict(data, features=FEATURES)
371
  hf_dataset.push_to_hub( "HuggingFaceM4/problematic-dataset-red-teaming", split="train", token=HF_WRITE_TOKEN, private=True)
372
 
373
+ # Change this to say whatever in order to have notes on the specific redteaming test added to the datasets.
374
+ test_type = gr.Textbox(label="Optional: Type of redteaming test.")
375
 
376
  # Hyper-parameters for generation
377
  max_new_tokens = gr.Slider(
 
525
  max_new_tokens,
526
  repetition_penalty,
527
  top_p,
528
+ test_type,
529
  ],
530
  outputs=None,
531
  preprocess=False,
 
540
  max_new_tokens,
541
  repetition_penalty,
542
  top_p,
543
+ test_type,
544
  ],
545
  outputs=None,
546
  preprocess=False,