Dataset Viewer
Auto-converted to Parquet Duplicate
instruction
stringlengths
16
91
answer
stringclasses
36 values
tag
stringclasses
35 values
split_kind
stringclasses
13 values
text
stringlengths
53
134
Can water become frozen? Give a reason.
Yes, because water can become ice when it is cold enough.
yes_water_freeze
train_seen
Instruction: Can water become frozen? Give a reason. Answer: Yes, because water can become ice when it is cold enough. <END>
The password supplied in this prompt is orange. Question: Which password was supplied?
orange.
context_password
context_seen_train
Instruction: The password supplied in this prompt is orange. Question: Which password was supplied? Answer: orange. <END>
The name supplied in this prompt is Sam. Question: Which name was supplied?
Sam.
context_name
context_seen_train
Instruction: The name supplied in this prompt is Sam. Question: Which name was supplied? Answer: Sam. <END>
What was Evan's meal today?
I don't know from the given information.
idk_food_today
task_upstream_regression_train
Instruction: What was Evan's meal today? Answer: I don't know from the given information. <END>
Is this sentiment positive or negative: That was helpful.
positive
sentiment_positive
task_seen_train
Instruction: Is this sentiment positive or negative: That was helpful. Answer: positive <END>
Type label: A clear sky is usually blue.
prose
classify_prose
task_seen_train
Instruction: Type label: A clear sky is usually blue. Answer: prose <END>
Do you have enough information to name my favorite movie?
I don't know from the given information.
idk_favorite_movie
task_seen_train
Instruction: Do you have enough information to name my favorite movie? Answer: I don't know from the given information. <END>
Facts: password is blue. name is Alex. Question: Which name was given?
Alex.
context_name
context_seen_train
Instruction: Facts: password is blue. name is Alex. Question: Which name was given? Answer: Alex. <END>
Is this sentiment positive or negative: This feels solid.
positive
sentiment_positive
task_seen_train
Instruction: Is this sentiment positive or negative: This feels solid. Answer: positive <END>
Make a Python function named multiply that multiplies two numbers.
def multiply(a, b): return a * b
code_multiply
train_seen
Instruction: Make a Python function named multiply that multiplies two numbers. Answer: def multiply(a, b): return a * b <END>
Provided name: Kai. Question: What name was provided?
Kai.
context_name
context_upstream_regression_train
Instruction: Provided name: Kai. Question: What name was provided? Answer: Kai. <END>
Use this stored name value: Nina. Question: What stored name value should be used?
Nina.
context_name
context_upstream_regression_train
Instruction: Use this stored name value: Nina. Question: What stored name value should be used? Answer: Nina. <END>
Context: password: purple. Question: Which password was given?
purple.
context_password
context_seen_train
Instruction: Context: password: purple. Question: Which password was given? Answer: purple. <END>
What sentiment is this: I love this tool.
positive
sentiment_positive
task_seen_train
Instruction: What sentiment is this: I love this tool. Answer: positive <END>
What sentiment is this: I enjoy this result.
positive
sentiment_positive
task_upstream_regression_train
Instruction: What sentiment is this: I enjoy this result. Answer: positive <END>
Label the input type: def add(a, b): return a + b
code
classify_code
task_upstream_regression_train
Instruction: Label the input type: def add(a, b): return a + b Answer: code <END>
Positive or negative sentiment: The fix helped.
positive
anti_collision_sentiment_positive_not_negative_or_prose
task_upstream_regression_train
Instruction: Positive or negative sentiment: The fix helped. Answer: positive <END>
Formal rewrite: I can't make it.
I am unable to attend.
rewrite_professional_attend
task_upstream_regression_train
Instruction: Formal rewrite: I can't make it. Answer: I am unable to attend. <END>
What sentiment is this: I enjoy this.
positive
sentiment_positive
task_seen_train
Instruction: What sentiment is this: I enjoy this. Answer: positive <END>
Type label: for i in range(3): print(i)
code
classify_code
task_seen_train
Instruction: Type label: for i in range(3): print(i) Answer: code <END>
Context: the name to remember is Jordan. Question: What name was given?
Jordan.
context_name
context_seen_train
Instruction: Context: the name to remember is Jordan. Question: What name was given? Answer: Jordan. <END>
Answer only code or prose: print('hello')
code
classify_code
task_seen_train
Instruction: Answer only code or prose: print('hello') Answer: code <END>
Context: the name to remember is Maya. Question: What name was given?
Maya.
context_name
context_seen_train
Instruction: Context: the name to remember is Maya. Question: What name was given? Answer: Maya. <END>
What should the assistant focus on?
I answer the user's request directly.
assistant_role
tt641_v2_guard_train
Instruction: What should the assistant focus on? Answer: I answer the user's request directly. <END>
Conversation: User: Use red as the password. User: What password did I give?
red.
context_password
context_seen_train
Instruction: Conversation: User: Use red as the password. User: What password did I give? Answer: red. <END>
Type label: def divide(a, b): return a / b
code
classify_code
task_seen_train
Instruction: Type label: def divide(a, b): return a / b Answer: code <END>
Sentiment label only: This feels solid.
positive
sentiment_positive
task_seen_train
Instruction: Sentiment label only: This feels solid. Answer: positive <END>
Type label: The cat sat down.
prose
classify_prose
task_seen_train
Instruction: Type label: The cat sat down. Answer: prose <END>
Make this concise: I am going to arrive late today.
I will be late today.
rewrite_short_late
task_seen_train
Instruction: Make this concise: I am going to arrive late today. Answer: I will be late today. <END>
What movie do I like most?
I don't know from the given information.
idk_favorite_movie
task_seen_train
Instruction: What movie do I like most? Answer: I don't know from the given information. <END>
Positive or negative sentiment: The fix helped.
positive
sentiment_positive
task_upstream_regression_train
Instruction: Positive or negative sentiment: The fix helped. Answer: positive <END>
Write the MULTIPLY function.
def multiply(a, b): return a * b
code_multiply
train_seen
Instruction: Write the MULTIPLY function. Answer: def multiply(a, b): return a * b <END>
Give the code/prose label: return user.email
code
classify_code
task_upstream_regression_train
Instruction: Give the code/prose label: return user.email Answer: code <END>
Classify this as positive or negative: This is broken.
negative
sentiment_negative
task_upstream_regression_train
Instruction: Classify this as positive or negative: This is broken. Answer: negative <END>
Classify input: The appointment was moved to Friday.
prose
anti_collision_classify_prose_not_rewrite
task_upstream_regression_train
Instruction: Classify input: The appointment was moved to Friday. Answer: prose <END>
Answer only positive or negative: This is a strong result.
positive
sentiment_positive
task_seen_train
Instruction: Answer only positive or negative: This is a strong result. Answer: positive <END>
Answer only code or prose: The user asked a direct question.
prose
classify_prose
task_seen_train
Instruction: Answer only code or prose: The user asked a direct question. Answer: prose <END>
Name the capital city of France.
Paris.
capital_france
train_seen
Instruction: Name the capital city of France. Answer: Paris. <END>
What should you do when you cannot prove the cause?
Ask for the missing evidence.
evidence_missing
train_seen
Instruction: What should you do when you cannot prove the cause? Answer: Ask for the missing evidence. <END>
Answer only positive or negative: I dislike this issue.
negative
sentiment_negative
task_seen_train
Instruction: Answer only positive or negative: I dislike this issue. Answer: negative <END>
What sentiment is this: This feels solid.
positive
sentiment_positive
task_seen_train
Instruction: What sentiment is this: This feels solid. Answer: positive <END>
Give the sentiment label only: That failed again.
negative
sentiment_negative
task_upstream_regression_train
Instruction: Give the sentiment label only: That failed again. Answer: negative <END>
What is your job here?
I answer the user's request directly.
assistant_role
train_seen
Instruction: What is your job here? Answer: I answer the user's request directly. <END>
Sentiment of this text: This is nice.
positive
sentiment_positive
task_seen_train
Instruction: Sentiment of this text: This is nice. Answer: positive <END>
Facts: password is purple. name is Kai. Question: Which name was given?
Kai.
context_name
context_seen_train
Instruction: Facts: password is purple. name is Kai. Question: Which name was given? Answer: Kai. <END>
Facts: password is purple. name is Kai. Question: Which name was given?
Kai.
context_name
context_seen_train
Instruction: Facts: password is purple. name is Kai. Question: Which name was given? Answer: Kai. <END>
Provided name: Sam. Question: What name was provided?
Sam.
context_name
context_upstream_regression_train
Instruction: Provided name: Sam. Question: What name was provided? Answer: Sam. <END>
Context: name = Alex; password = purple. Question: What is the password?
purple.
context_password
context_seen_train
Instruction: Context: name = Alex; password = purple. Question: What is the password? Answer: purple. <END>
Which food did Evan mention eating today?
I don't know from the given information.
idk_food_today
task_seen_train
Instruction: Which food did Evan mention eating today? Answer: I don't know from the given information. <END>
How should you answer the user's request?
I answer the user's request directly.
assistant_role
train_seen
Instruction: How should you answer the user's request? Answer: I answer the user's request directly. <END>
Give the code/prose label: for i in range(3): print(i)
code
classify_code
task_seen_train
Instruction: Give the code/prose label: for i in range(3): print(i) Answer: code <END>
The password supplied in this prompt is blue. Question: Which password was supplied?
blue.
context_password
context_seen_train
Instruction: The password supplied in this prompt is blue. Question: Which password was supplied? Answer: blue. <END>
Line one says NAME=Kai. Question: What is NAME?
Kai.
context_name
context_upstream_regression_train
Instruction: Line one says NAME=Kai. Question: What is NAME? Answer: Kai. <END>
The name supplied in this prompt is Kai. Question: Which name was supplied?
Kai.
context_name
context_upstream_regression_train
Instruction: The name supplied in this prompt is Kai. Question: Which name was supplied? Answer: Kai. <END>
What sentiment is this: This works great.
positive
sentiment_positive
task_seen_train
Instruction: What sentiment is this: This works great. Answer: positive <END>
What sentiment is this: That failed again.
negative
anti_collision_sentiment_negative_not_positive
task_upstream_regression_train
Instruction: What sentiment is this: That failed again. Answer: negative <END>
Context: name = Alex; password = purple. Question: What is the password?
purple.
context_password
context_seen_train
Instruction: Context: name = Alex; password = purple. Question: What is the password? Answer: purple. <END>
Return the product using multiply(a, b).
def multiply(a, b): return a * b
code_multiply
train_seen
Instruction: Return the product using multiply(a, b). Answer: def multiply(a, b): return a * b <END>
Facts: name is Nina. password is black. Question: Which password was given?
black.
context_password
context_seen_train
Instruction: Facts: name is Nina. password is black. Question: Which password was given? Answer: black. <END>
Facts: password is green. name is Sam. Question: Which name was given?
Sam.
context_name
context_seen_train
Instruction: Facts: password is green. name is Sam. Question: Which name was given? Answer: Sam. <END>
France has what capital?
Paris.
capital_france
train_seen
Instruction: France has what capital? Answer: Paris. <END>
Line one says PASSWORD=gold. Question: What is PASSWORD?
gold.
context_password
context_upstream_regression_train
Instruction: Line one says PASSWORD=gold. Question: What is PASSWORD? Answer: gold. <END>
Choose one label, code or prose: return value
code
classify_code
task_seen_train
Instruction: Choose one label, code or prose: return value Answer: code <END>
What type of text is this: z = divide(9, 3)
code
classify_code
task_seen_train
Instruction: What type of text is this: z = divide(9, 3) Answer: code <END>
Is this code or prose: A clear sky is usually blue.
prose
classify_prose
task_seen_train
Instruction: Is this code or prose: A clear sky is usually blue. Answer: prose <END>
Give only the type label for this: The schedule changed again.
prose
classify_prose
task_seen_train
Instruction: Give only the type label for this: The schedule changed again. Answer: prose <END>
What sentiment is this: I really like this.
positive
sentiment_positive
task_upstream_regression_train
Instruction: What sentiment is this: I really like this. Answer: positive <END>
Choose the sentiment label for: I dislike this issue.
negative
sentiment_negative
task_seen_train
Instruction: Choose the sentiment label for: I dislike this issue. Answer: negative <END>
Sentiment of this text: This is nice.
positive
sentiment_positive
task_seen_train
Instruction: Sentiment of this text: This is nice. Answer: positive <END>
Answer only positive or negative: This is a strong result.
positive
sentiment_positive
task_seen_train
Instruction: Answer only positive or negative: This is a strong result. Answer: positive <END>
Context: name = Maya; password = gold. Question: What is the name?
Maya.
context_name
context_seen_train
Instruction: Context: name = Maya; password = gold. Question: What is the name? Answer: Maya. <END>
Before a code change, what should happen first?
Inspect the current files first.
inspect_first
tt641_route_boundary_train
Instruction: Before a code change, what should happen first? Answer: Inspect the current files first. <END>
Facts: password is blue. name is Alex. Question: Which name was given?
Alex.
context_name
context_seen_train
Instruction: Facts: password is blue. name is Alex. Question: Which name was given? Answer: Alex. <END>
Positive or negative sentiment: The fix helped.
positive
sentiment_positive
task_upstream_regression_train
Instruction: Positive or negative sentiment: The fix helped. Answer: positive <END>
What is the assistant's job?
I answer the user's request directly.
assistant_role
train_seen
Instruction: What is the assistant's job? Answer: I answer the user's request directly. <END>
Use this stored name value: Sam. Question: What stored name value should be used?
Sam.
context_name
context_upstream_regression_train
Instruction: Use this stored name value: Sam. Question: What stored name value should be used? Answer: Sam. <END>
How should you answer the user's request?
I answer the user's request directly.
assistant_role
tt641_v2_guard_train
Instruction: How should you answer the user's request? Answer: I answer the user's request directly. <END>
Make this professional: I can't make it.
I am unable to attend.
rewrite_professional_attend
task_upstream_regression_train
Instruction: Make this professional: I can't make it. Answer: I am unable to attend. <END>
Context: name = Nina; password = orange. Question: What is the name?
Nina.
anti_context_name_not_password
tt639g_lite_v2_failure_repair_train
Instruction: Context: name = Nina; password = orange. Question: What is the name? Answer: Nina. <END>
If evidence is missing, what should you do?
Ask for the missing evidence.
evidence_missing
tt641_route_boundary_train
Instruction: If evidence is missing, what should you do? Answer: Ask for the missing evidence. <END>
Context: The user's name is Maya. Question: What is the user's name?
Maya.
context_name
context_seen_train
Instruction: Context: The user's name is Maya. Question: What is the user's name? Answer: Maya. <END>
Give only the type label for this: The dog ran home.
prose
classify_prose
task_seen_train
Instruction: Give only the type label for this: The dog ran home. Answer: prose <END>
Line one says NAME=Kai. Question: What is NAME?
Kai.
context_name
context_upstream_regression_train
Instruction: Line one says NAME=Kai. Question: What is NAME? Answer: Kai. <END>
Classify input: Please read the current file first.
prose
classify_prose
task_seen_train
Instruction: Classify input: Please read the current file first. Answer: prose <END>
User: The passcode is silver. User: What is the passcode?
silver.
context_password
context_seen_train
Instruction: User: The passcode is silver. User: What is the passcode? Answer: silver. <END>
Classify input: This sentence is plain English.
prose
classify_prose
task_seen_train
Instruction: Classify input: This sentence is plain English. Answer: prose <END>
What sentiment is this: That failed again.
negative
sentiment_negative
task_upstream_regression_train
Instruction: What sentiment is this: That failed again. Answer: negative <END>
How should you answer the user's request?
I answer the user's request directly.
assistant_role
tt639g_lite_v2_failure_repair_train
Instruction: How should you answer the user's request? Answer: I answer the user's request directly. <END>
What should happen if the cause is unproven?
Ask for the missing evidence.
evidence_missing
train_seen
Instruction: What should happen if the cause is unproven? Answer: Ask for the missing evidence. <END>
Choose one label, code or prose: The meeting starts tomorrow.
prose
classify_prose
task_seen_train
Instruction: Choose one label, code or prose: The meeting starts tomorrow. Answer: prose <END>
What sentiment is this: I dislike this issue.
negative
sentiment_negative
task_upstream_regression_train
Instruction: What sentiment is this: I dislike this issue. Answer: negative <END>
Classify sentiment: I really like this.
positive
sentiment_positive
task_seen_train
Instruction: Classify sentiment: I really like this. Answer: positive <END>
Make this professional: I won't be able to make it.
I am unable to attend.
rewrite_professional_attend
task_seen_train
Instruction: Make this professional: I won't be able to make it. Answer: I am unable to attend. <END>
Make a Python function named divide that divides a by b.
def divide(a, b): return a / b
code_divide
train_seen
Instruction: Make a Python function named divide that divides a by b. Answer: def divide(a, b): return a / b <END>
Positive or negative sentiment: This failed badly.
negative
sentiment_negative
task_seen_train
Instruction: Positive or negative sentiment: This failed badly. Answer: negative <END>
What type of text is this: A clear sky is usually blue.
prose
classify_prose
task_seen_train
Instruction: What type of text is this: A clear sky is usually blue. Answer: prose <END>
Concise rewrite: I am going to be late today.
I will be late today.
rewrite_short_late
task_upstream_regression_train
Instruction: Concise rewrite: I am going to be late today. Answer: I will be late today. <END>
Answer only positive or negative: That failed again.
negative
sentiment_negative
task_seen_train
Instruction: Answer only positive or negative: That failed again. Answer: negative <END>
Decide if this is code or prose: This sentence is plain English.
prose
classify_prose
task_seen_train
Instruction: Decide if this is code or prose: This sentence is plain English. Answer: prose <END>
Facts: name is Kai. password is silver. Question: Which password was given?
silver.
context_password
context_seen_train
Instruction: Facts: name is Kai. password is silver. Question: Which password was given? Answer: silver. <END>
End of preview. Expand in Data Studio

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

TT639G Recombined Tiny Assistant v1

Recombines isolated proof rungs:

  • TT638D code behavior + dyadic/Mercy proof upstream
  • TT639E2 context-copy behavior
  • TT639F3 task-routing behavior
  • simple rule/Q&A behavior

Blocking dense gates:

  • seen_combined_pass
  • upstream_regression_pass
  • mixed_heldout_pass
  • anti_collision_pass

Do not run dyadic/Mercy compare unless all four gates pass.

Downloads last month
64