Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
11
22
source_dataset
stringclasses
1 value
source_config
stringclasses
6 values
source_family
stringclasses
1 value
source_split
stringclasses
1 value
benchmark_split
stringclasses
1 value
transcript
stringlengths
1.43k
4.32k
prompt
stringlengths
1.57k
4.46k
target_json
stringclasses
43 values
reference_summary
stringclasses
0 values
reference_intent
stringclasses
43 values
reference_sentiment
stringclasses
0 values
reference_resolved
bool
0 classes
reference_escalation_needed
bool
0 classes
has_gold_summary
bool
1 class
has_gold_intent
bool
2 classes
has_gold_sentiment
bool
1 class
has_gold_resolution
bool
1 class
has_gold_escalation
bool
1 class
transcript_tokens
int32
478
1.17k
input_tokens
int32
512
1.2k
expected_max_new_tokens
int32
104
114
prefill_ratio_estimate
float64
0.83
0.91
decode_ratio_estimate
float64
0.09
0.17
static_prefix_tokens
int32
34
34
static_prefix_reuse
float64
0.03
0.07
length_bucket
stringclasses
2 values
license
stringclasses
3 values
redistribution_status
stringclasses
3 values
source_attribution
stringclasses
6 values
Taskmaster3-train-3745
Salesforce/dialogstudio
Taskmaster3
customer_service
train
source
USER: I am trying to purchase movie tickets for tonight. Can you help me with that? ASSISTANT: OK. Can you tell me the city and state where the theater is located? USER: I have been looking forward to going to the movies with mom and dad as I have not seen them for a long time so I am a bit excited ASSISTANT: Great. Ca...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I am trying to purchase movie tickets for tonight. Can you help me with that? ASSISTANT: OK. Can you tell me the city and state where the theater is located? USER: I have...
null
null
null
null
null
null
false
false
false
false
false
547
581
105
0.846939
0.153061
34
0.05852
512_1k
CC BY 4.0
redistributable
Taskmaster-3 / TicketTalk (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
Taskmaster2-train-659
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: Hi. I would like to find hotel for my summer vacation. ASSISTANT: Hello! How can I help you. Okay where would you like to go on your summer vacation? USER: Naples, Florida. ASSISTANT: Okay and when were you planning on going? Do you have any preferences for the hotel? USER: Yes. Hotel should be walking distance f...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hi. I would like to find hotel for my summer vacation. ASSISTANT: Hello! How can I help you. Okay where would you like to go on your summer vacation? USER: Naples, Florid...
null
null
null
null
null
null
false
false
false
false
false
510
544
104
0.839506
0.160494
34
0.0625
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
Taskmaster3-train-82
Salesforce/dialogstudio
Taskmaster3
customer_service
train
source
USER: I saw The Addams Family yesterday. I didn't care for the way they animated it; it's too fake looking and doesn't appeal to me the way the one with real actors did. And yet, Maleficent had an opposite effect on me; it took a cartoon story and put real actors into it, and I just wanted to leave the theater. There w...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I saw The Addams Family yesterday. I didn't care for the way they animated it; it's too fake looking and doesn't appeal to me the way the one with real actors did. And ye...
null
null
null
null
null
null
false
false
false
false
false
1,050
1,084
112
0.906355
0.093645
34
0.031365
1k_2k
CC BY 4.0
redistributable
Taskmaster-3 / TicketTalk (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
SalesBot-train-1160
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: Oh, I see. I've never been fishing, but I do enjoy watching it on TV. ASSISTANT: It's a lot of fun. You should try it sometime. What do you like to watch on TV? USER: I don't have cable, so I mostly just watch Netflix and Hulu. What about you? ASSISTANT: I got it. I only watch Netflix. Do you have any interest in...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Oh, I see. I've never been fishing, but I do enjoy watching it on TV. ASSISTANT: It's a lot of fun. You should try it sometime. What do you like to watch on TV? USER: I d...
{"intent": "FindMovies", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindMovies
null
null
null
false
true
false
false
false
648
682
106
0.865482
0.134518
34
0.049853
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
Taskmaster2-train-613
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: Hi. I would like help finding a hotel in New Orleans. ASSISTANT: I can help you with your hotel search in New Orleans. USER: I would like to stay from March 5th to March 10th. ASSISTANT: Okay, check in date March 5th check out date March 10th a five-night stay. USER: There will also be two people staying in the r...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hi. I would like help finding a hotel in New Orleans. ASSISTANT: I can help you with your hotel search in New Orleans. USER: I would like to stay from March 5th to March ...
null
null
null
null
null
null
false
false
false
false
false
495
529
104
0.835703
0.164297
34
0.064272
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
SalesBot-train-3
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: That would be nice. I love to cook. What do you like to do in your spare time? ASSISTANT: I enjoy dressing up and going out to eat. I think I'll be famous someday. USER: That sounds like a lot of fun. What kind of food do you dress up as? I love Italian. ASSISTANT: I love italian as well. I usually dress up in co...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: That would be nice. I love to cook. What do you like to do in your spare time? ASSISTANT: I enjoy dressing up and going out to eat. I think I'll be famous someday. USER: ...
{"intent": "FindAttractions", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindAttractions
null
null
null
false
true
false
false
false
549
583
105
0.847384
0.152616
34
0.058319
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
Taskmaster2-train-2989
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: Hi, this is Moli. Hi, Would you like to help finding a flight to California? ASSISTANT: What part of California? USER: Northridge, San Diego. ASSISTANT: What date were you looking to leave? USER: Yeah, I have a work trip on coming Febuary 2017. And could you please find a commercial flight for me? And could you p...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hi, this is Moli. Hi, Would you like to help finding a flight to California? ASSISTANT: What part of California? USER: Northridge, San Diego. ASSISTANT: What date were yo...
null
null
null
null
null
null
false
false
false
false
false
612
646
106
0.859043
0.140957
34
0.052632
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
Taskmaster2-train-3584
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: Hi. I would like help in finding a flight to Orlando, Florida. ASSISTANT: Okay, and what do you need in a flight? USER: I need one coach ticket. ASSISTANT: Okay, when you're leaving? USER: I'd like to leave on Friday, February 3rd. ASSISTANT: And come back on? USER: Tuesday, February 7th. ASSISTANT: Okay, and wha...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hi. I would like help in finding a flight to Orlando, Florida. ASSISTANT: Okay, and what do you need in a flight? USER: I need one coach ticket. ASSISTANT: Okay, when you...
null
null
null
null
null
null
false
false
false
false
false
481
515
104
0.831987
0.168013
34
0.066019
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
SalesBot-train-385
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: I moved to Las Vegas, Nevada. I love it here. What about you? ASSISTANT: I've never been there, but I've always wanted to go. I live in New York City. USER: I have never been to New York. I would love to go one day. What do you do for a living? ASSISTANT: I work in an office. It's boring, but it pays the bills. H...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I moved to Las Vegas, Nevada. I love it here. What about you? ASSISTANT: I've never been there, but I've always wanted to go. I live in New York City. USER: I have never ...
{"intent": "FindMovies", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindMovies
null
null
null
false
true
false
false
false
576
610
105
0.853147
0.146853
34
0.055738
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
ABCD-train-2819
Salesforce/dialogstudio
ABCD
customer_service
train
source
ASSISTANT: Hello and thank you for contacting us. How may I help you? USER: Hi there. I made an order, but I changed my mind on it. I was hoping to get a refund. ASSISTANT: Sure thing, I should be able to help you with that. Can I please have your First and Last name? USER: yes, it's Albert Sanders. ASSISTANT: Thank yo...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: ASSISTANT: Hello and thank you for contacting us. How may I help you? USER: Hi there. I made an order, but I changed my mind on it. I was hoping to get a refund. ASSISTANT: Sur...
{"intent": "refund_initiate", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
refund_initiate
null
null
null
false
true
false
false
false
555
589
105
0.848703
0.151297
34
0.057725
512_1k
MIT
redistributable
ABCD (Action-Based Conversations Dataset), ASAPP Research, MIT License. https://github.com/asappresearch/abcd — via Salesforce/dialogstudio.
SalesBot-train-260
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: That sounds like a lot of fun. Do you have a favorite game to play with your friends? ASSISTANT: I like to play billiards with my friends. It is a great way to socialize. USER: I agree. It's a good way to get together with friends and have fun. What else do you like to do? ASSISTANT: I also like to go to the movi...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: That sounds like a lot of fun. Do you have a favorite game to play with your friends? ASSISTANT: I like to play billiards with my friends. It is a great way to socialize....
{"intent": "FindMovies", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindMovies
null
null
null
false
true
false
false
false
501
535
104
0.837246
0.162754
34
0.063551
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
Taskmaster3-train-1884
Salesforce/dialogstudio
Taskmaster3
customer_service
train
source
USER: I'd like to find out if the movie 1917 is playing at the AMC Boston Common movie theater\r ASSISTANT: Sure, I can help you with that. Is there a specific time you'd like to see the movie? USER: Yes, 9 PM please. ASSISTANT: Okay, I just checked and it is playing at the AMC Boston Common at 9 PM. USER: Great, I'd l...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I'd like to find out if the movie 1917 is playing at the AMC Boston Common movie theater\r ASSISTANT: Sure, I can help you with that. Is there a specific time you'd like ...
null
null
null
null
null
null
false
false
false
false
false
483
517
104
0.832528
0.167472
34
0.065764
512_1k
CC BY 4.0
redistributable
Taskmaster-3 / TicketTalk (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
SalesBot-train-419
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: That sounds like a lot of fun. What do you like to do in your spare time? ASSISTANT: I like to spend time with my family. We have a big house so we have lots of room for activities. USER: That's great. I love spending time with family as well. Do you have any hobbies? ASSISTANT: Yes, I love to cook and bake. I al...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: That sounds like a lot of fun. What do you like to do in your spare time? ASSISTANT: I like to spend time with my family. We have a big house so we have lots of room for ...
{"intent": "FindMovies", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindMovies
null
null
null
false
true
false
false
false
598
632
105
0.857531
0.142469
34
0.053797
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
Taskmaster2-train-2337
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: I'd like help finding a hotel in San Francisco, California. ASSISTANT: Okay. USER: Great. I need for two adults. ASSISTANT: I can help you. USER: Okay. ASSISTANT: What else? USER: Do you have something in the Richmond District? ASSISTANT: Yes I do. USER: Great. Do you have a four-star hotel? ASSISTANT: What else?...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I'd like help finding a hotel in San Francisco, California. ASSISTANT: Okay. USER: Great. I need for two adults. ASSISTANT: I can help you. USER: Okay. ASSISTANT: What el...
null
null
null
null
null
null
false
false
false
false
false
708
742
107
0.873969
0.126031
34
0.045822
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
Taskmaster3-train-2140
Salesforce/dialogstudio
Taskmaster3
customer_service
train
source
USER: Hello, can I book 2 movie tickets for tonight? ASSISTANT: Sure, May I know which location you prefer? USER: Spring Hills Kansas. ASSISTANT: Ok, let me check on the movie list here. So I have found 5 movies playing for tonight, Do you have a specific movie in mind or do you want to hear the list of the movies? USE...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hello, can I book 2 movie tickets for tonight? ASSISTANT: Sure, May I know which location you prefer? USER: Spring Hills Kansas. ASSISTANT: Ok, let me check on the movie ...
null
null
null
null
null
null
false
false
false
false
false
583
617
105
0.854571
0.145429
34
0.055105
512_1k
CC BY 4.0
redistributable
Taskmaster-3 / TicketTalk (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
MULTIWOZ2_2-train-2091
Salesforce/dialogstudio
MULTIWOZ2_2
customer_service
train
source
USER: i am looking for a place to stay. The hotel should be in the type of guesthouse and should be in the east. ASSISTANT: There are several to choose from. Is there a price range you are looking for? USER: Yes, I'm on food stamps and I can only afford the cheapest place. ASSISTANT: There are 3 guesthouses available. ...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: i am looking for a place to stay. The hotel should be in the type of guesthouse and should be in the east. ASSISTANT: There are several to choose from. Is there a price r...
null
null
null
null
null
null
false
false
false
false
false
489
523
104
0.834131
0.165869
34
0.06501
512_1k
MIT
redistributable
MultiWOZ 2.2 (Budzianowski et al. / Google fixes), MIT License. https://github.com/budzianowski/multiwoz — via Salesforce/dialogstudio.
Taskmaster2-train-2741
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: Hi, I would like to arrange a flight. ASSISTANT: Okay. USER: Okay. I'm going to Athens, Greece on 24th. ASSISTANT: Sounds good. Is it March 24th? USER: Yes and I'll be returning on March 31st, so I'll need a return flight as well. ASSISTANT: Okay. USER: I'd prefer it to be sometime in the morning or afternoon for...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hi, I would like to arrange a flight. ASSISTANT: Okay. USER: Okay. I'm going to Athens, Greece on 24th. ASSISTANT: Sounds good. Is it March 24th? USER: Yes and I'll be re...
null
null
null
null
null
null
false
false
false
false
false
638
672
106
0.863753
0.136247
34
0.050595
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
Taskmaster2-train-3288
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: I need a flight to Los Angeles. ASSISTANT: Okay to Los Angeles International. From where? USER: New York LaGuardia. ASSISTANT: Okay from LaGuardia. What else can you tell me? USER: Leave on March 1st, return on March 4th. First class. I'd like to leave in the morning. ASSISTANT: Okay. Mar 1st to Mar 4th. Then fir...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I need a flight to Los Angeles. ASSISTANT: Okay to Los Angeles International. From where? USER: New York LaGuardia. ASSISTANT: Okay from LaGuardia. What else can you tell...
null
null
null
null
null
null
false
false
false
false
false
570
604
105
0.851904
0.148096
34
0.056291
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
Taskmaster3-train-1034
Salesforce/dialogstudio
Taskmaster3
customer_service
train
source
USER: Hi, I'd like to check out some options for movies I am interested in seeing tonight, any suggestions? ASSISTANT: Sure! do you already have an idea about what type of movie you want to see? USER: I really like comedies, but I am also into action pics too. ASSISTANT: Ok, two options come to mind, the first is a Com...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hi, I'd like to check out some options for movies I am interested in seeing tonight, any suggestions? ASSISTANT: Sure! do you already have an idea about what type of movi...
null
null
null
null
null
null
false
false
false
false
false
640
674
106
0.864103
0.135897
34
0.050445
512_1k
CC BY 4.0
redistributable
Taskmaster-3 / TicketTalk (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
SalesBot-train-737
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: That's good to know. I live on the Gulf of Mexico so I'm not too worried. ASSISTANT: That's awesome! I've always wanted to visit the gulf of mexico. Have you ever been to Cuba? USER: No, I haven't, but I would love to go one day. I hear it's beautiful. ASSISTANT: It really is. It's one of the most popular tour...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: That's good to know. I live on the Gulf of Mexico so I'm not too worried. ASSISTANT: That's awesome! I've always wanted to visit the gulf of mexico. Have you ever been ...
{"intent": "FindAttractions", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindAttractions
null
null
null
false
true
false
false
false
633
667
106
0.862872
0.137128
34
0.050975
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
Taskmaster3-train-748
Salesforce/dialogstudio
Taskmaster3
customer_service
train
source
USER: Hello. I want you to find out if two movies released several months ago are still play in local cinemas. ASSISTANT: Okay. What is your first movie choice? USER: The film, Harriet. ASSISTANT: Searching. The film, Harriet, is only playing at one local theater: Colonial Park Cinema. USER: What are the playing times ...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hello. I want you to find out if two movies released several months ago are still play in local cinemas. ASSISTANT: Okay. What is your first movie choice? USER: The film,...
null
null
null
null
null
null
false
false
false
false
false
615
649
106
0.859603
0.140397
34
0.052388
512_1k
CC BY 4.0
redistributable
Taskmaster-3 / TicketTalk (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
Taskmaster3-train-1761
Salesforce/dialogstudio
Taskmaster3
customer_service
train
source
USER: I want to order tickets for a movie to watch with my wife. ASSISTANT: That sounds great! What kind of movie do you want to see? USER: We want to see a good drama movie. ASSISTANT: There are a couple of dramas playing in theaters now. Would you be interested in "Little Women" or "1917"? USER: 1917 sounds interesti...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I want to order tickets for a movie to watch with my wife. ASSISTANT: That sounds great! What kind of movie do you want to see? USER: We want to see a good drama movie. A...
null
null
null
null
null
null
false
false
false
false
false
600
634
105
0.857916
0.142084
34
0.053628
512_1k
CC BY 4.0
redistributable
Taskmaster-3 / TicketTalk (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
SalesBot-train-620
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: I have a Honda civic and I love it. It's the best car I've ever owned. ASSISTANT: That's good to hear. Do you have any hobbies that you like to do in your spare time? USER: I like to sing and play video games. What about you? What kind of hobbies do you have? ASSISTANT: I love to rock climb! It's one of my favori...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I have a Honda civic and I love it. It's the best car I've ever owned. ASSISTANT: That's good to hear. Do you have any hobbies that you like to do in your spare time? USE...
{"intent": "FindMovies", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindMovies
null
null
null
false
true
false
false
false
530
564
104
0.844311
0.155689
34
0.060284
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
SalesBot-train-1186
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: I don't want her to think I'm flirting with her when I'm really not. ASSISTANT: That's a good way to look at it. What do you like to do in your spare time? USER: I like to spend time with my family and play video games. What about you? Do you have any hobbies? ASSISTANT: I spend a lot of time outdoors. I love to ...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I don't want her to think I'm flirting with her when I'm really not. ASSISTANT: That's a good way to look at it. What do you like to do in your spare time? USER: I like t...
{"intent": "FindAttractions", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindAttractions
null
null
null
false
true
false
false
false
578
612
105
0.853556
0.146444
34
0.055556
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
MULTIWOZ2_2-train-690
Salesforce/dialogstudio
MULTIWOZ2_2
customer_service
train
source
USER: I'm looking for a hotel in the moderate price range. I definitely want a hotel, not a guesthouse. ASSISTANT: Of course. There are two hotels in the north and one in the centre of town. Do you have a preference? USER: Is there anything to the west, preferably with a 4 star rating? ASSISTANT: I'm sorry but nothing ...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I'm looking for a hotel in the moderate price range. I definitely want a hotel, not a guesthouse. ASSISTANT: Of course. There are two hotels in the north and one in the c...
null
null
null
null
null
null
false
false
false
false
false
566
600
105
0.851064
0.148936
34
0.056667
512_1k
MIT
redistributable
MultiWOZ 2.2 (Budzianowski et al. / Google fixes), MIT License. https://github.com/budzianowski/multiwoz — via Salesforce/dialogstudio.
SalesBot-train-217
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: I know what you mean. I'm pretty conservative when it comes to politics, so I don't like to borrow things. ASSISTANT: I can understand that. Politics can be very stressful. Do you have any hobbies? USER: I like to play video games. What about you? What do you like to do in your spare time? ASSISTANT: I enjoy play...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I know what you mean. I'm pretty conservative when it comes to politics, so I don't like to borrow things. ASSISTANT: I can understand that. Politics can be very stressfu...
{"intent": "FindMovies", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindMovies
null
null
null
false
true
false
false
false
483
517
104
0.832528
0.167472
34
0.065764
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
Taskmaster2-train-2519
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: Hi, I'd like to get a flight to San Jose, California. ASSISTANT: From where will you be departing from? USER: Any one of the airports in the New York City area. ASSISTANT: When will you be departing? USER: On monday. ASSISTANT: When will you be returning? USER: The following Saturday. ASSISTANT: Any preferences o...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hi, I'd like to get a flight to San Jose, California. ASSISTANT: From where will you be departing from? USER: Any one of the airports in the New York City area. ASSISTANT...
null
null
null
null
null
null
false
false
false
false
false
528
562
104
0.843844
0.156156
34
0.060498
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
SalesBot-train-1592
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: Yes, I do. I work at a children's library. What do you do for a living? ASSISTANT: Do you have anything else? USER: Can you help me find attractions to visit? ASSISTANT: What city are you interested in? USER: I'm looking for child-friendly attractions near SF. ASSISTANT: I have [count] attractions you may like. F...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Yes, I do. I work at a children's library. What do you do for a living? ASSISTANT: Do you have anything else? USER: Can you help me find attractions to visit? ASSISTANT: ...
{"intent": "FindAttractions", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindAttractions
null
null
null
false
true
false
false
false
652
686
106
0.866162
0.133838
34
0.049563
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
Taskmaster2-train-3666
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: Hello. ASSISTANT: How can I help you? USER: Hi, today I need your help sending up a round trip flight from Miami to Berlin. ASSISTANT: I can help you with your flight reservation. USER: Great. ASSISTANT: When will you be traveling to Berlin? USER: I'm looking to travel to Berlin on this Sunday which I believe the...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hello. ASSISTANT: How can I help you? USER: Hi, today I need your help sending up a round trip flight from Miami to Berlin. ASSISTANT: I can help you with your flight res...
null
null
null
null
null
null
false
false
false
false
false
478
512
104
0.831169
0.168831
34
0.066406
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
Taskmaster3-train-2078
Salesforce/dialogstudio
Taskmaster3
customer_service
train
source
USER: Good morning. I would appreicate your assistance in finding a movie to see later this evening. ASSISTANT: I would be more than happy to assist you. In what area would you like to see a movie? USER: I live in Spring Hills, Kansas. ASSISTANT: I am showing there are two theaters in this area: AMC Holiday Theater and...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Good morning. I would appreicate your assistance in finding a movie to see later this evening. ASSISTANT: I would be more than happy to assist you. In what area would you...
null
null
null
null
null
null
false
false
false
false
false
506
540
104
0.838509
0.161491
34
0.062963
512_1k
CC BY 4.0
redistributable
Taskmaster-3 / TicketTalk (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
Taskmaster2-train-1475
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: Hi, I'd like to find the hotel for my summer vacation, please. Naples, Florida on the beach, please. I'd like it to be a resort-style hotel probably 4-stars and up, and I'd like to spend about $300 per night or less. ASSISTANT: Sounds beautiful. When were you planning on going? USER: June 16th to June 19th, 2017....
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: Hi, I'd like to find the hotel for my summer vacation, please. Naples, Florida on the beach, please. I'd like it to be a resort-style hotel probably 4-stars and up, and I...
null
null
null
null
null
null
false
false
false
false
false
551
585
105
0.847826
0.152174
34
0.05812
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
SalesBot-train-391
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: I know what you mean. I hate it when I get bitten by them. Do you have any pets? ASSISTANT: Yes, I have a dog and a cat. What about you? What kind of pets do you have? USER: I have two dogs and two cats. They're all rescues from the shelter. ASSISTANT: That's so nice of you to rescue them. Do they get along wi...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I know what you mean. I hate it when I get bitten by them. Do you have any pets? ASSISTANT: Yes, I have a dog and a cat. What about you? What kind of pets do you have? ...
{"intent": "FindMovies", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindMovies
null
null
null
false
true
false
false
false
995
1,029
112
0.90184
0.09816
34
0.033042
1k_2k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
SalesBot-train-1995
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: I don't think she heard me. I think she just didn't want to talk to me. ASSISTANT: I'm sorry to hear that. I hope you find a way to make it up to her. USER: I hope so too. I've been trying to get a new job so I can travel more. ASSISTANT: That's a good idea. Traveling is so much fun. Where would you like to go? U...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I don't think she heard me. I think she just didn't want to talk to me. ASSISTANT: I'm sorry to hear that. I hope you find a way to make it up to her. USER: I hope so too...
{"intent": "FindAttractions", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindAttractions
null
null
null
false
true
false
false
false
522
556
104
0.842424
0.157576
34
0.061151
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
Taskmaster2-train-2275
Salesforce/dialogstudio
Taskmaster2
customer_service
train
source
USER: I'd like help finding a hotel in Nashville, Tennessee. ASSISTANT: Okay, what else can you tell me? USER: I would like the the price to not be over $100 a night. ASSISTANT: Okay and what else? USER: I would like the hotel to be a two-bedroom. ASSISTANT: Okay. USER: I would like a refrigerator. ASSISTANT: Alright. ...
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: I'd like help finding a hotel in Nashville, Tennessee. ASSISTANT: Okay, what else can you tell me? USER: I would like the the price to not be over $100 a night. ASSISTANT...
null
null
null
null
null
null
false
false
false
false
false
478
512
104
0.831169
0.168831
34
0.066406
512_1k
CC BY 4.0
redistributable
Taskmaster-2 (Google Research), CC BY 4.0. https://github.com/google-research-datasets/Taskmaster — via Salesforce/dialogstudio.
SalesBot-train-646
Salesforce/dialogstudio
SalesBot
customer_service
train
source
USER: That's good to know. I live on the Gulf of Mexico, so I'm not too worried. ASSISTANT: I have been to the gulf of mexico a few times. It is beautiful. Do you have any hobbies? USER: I like to play video games. I play a lot of League of Legends. What about you? ASSISTANT: I love video games! I also like to read....
Analyze this customer-support transcript. Return compact JSON with: - intent - sentiment - resolved - escalation_needed - summary Transcript: USER: That's good to know. I live on the Gulf of Mexico, so I'm not too worried. ASSISTANT: I have been to the gulf of mexico a few times. It is beautiful. Do you have any ho...
{"intent": "FindAttractions", "sentiment": null, "resolved": null, "escalation_needed": null, "summary": null, "follow_up_actions": []}
null
FindAttractions
null
null
null
false
true
false
false
false
640
674
106
0.864103
0.135897
34
0.050445
512_1k
null
no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
SalesBot (MiuLab, NTU). https://github.com/MiuLab/SalesBot — via Salesforce/dialogstudio. No upstream LICENSE file; text is fully synthetic.
End of preview. Expand in Data Studio

Customer Transcript Analytics

Curated customer-support and meeting transcripts mapped to a single fixed "analyze this transcript → compact JSON" prompt, for benchmarking batched offline LLM inference on realistic workloads.

Motivation and intended use

This dataset provides a realistic transcript-analytics workload for batched offline-inference experiments: throughput benchmarking and predicted-vs-observed throughput validation. Rows range from short support chats of a few hundred tokens to long multi-party meeting segments, and every row carries token accounting (input_tokens, expected_max_new_tokens, prefill/decode ratio estimates, length_bucket) so throughput can be analyzed by prompt length.

It is not a correctness benchmark for intent, sentiment, resolution, or summarization quality. Model outputs generated from these prompts should not be interpreted as validated labels.

Prompt template

Every row uses this exact template (the static prefix before the transcript is constant across rows; per-row static_prefix_tokens / static_prefix_reuse quantify it for prefix-caching analysis):

Analyze this customer-support transcript.

Return compact JSON with:
- intent
- sentiment
- resolved
- escalation_needed
- summary

Transcript:
{{transcript}}

expected_max_new_tokens is a documented heuristic, not a measurement: min(256, max(96, 96 + input_tokens // 64)). Token accounting is approximate and uses the Qwen/Qwen3.5-9B tokenizer (add_special_tokens=False).

Example usage (OpenAI-style request)

Each row's prompt is sent as a single user message, and the row's expected_max_new_tokens sets max_tokens. The model's JSON is generated at benchmark time — it is not stored in the dataset and must not be read as a validated label. The responses below are illustrative only.

Short customer-support row (short_control, ABCD, 189 input tokens):

{
  "model": "Qwen/Qwen3.5-9B",
  "temperature": 0,
  "max_tokens": 98,
  "messages": [
    {
      "role": "user",
      "content": "Analyze this customer-support transcript.\n\nReturn compact JSON with:\n- intent\n- sentiment\n- resolved\n- escalation_needed\n- summary\n\nTranscript:\nASSISTANT: Hello, how can i help you today\nUSER: I need to check on my order because the delivery time is wrong\n[... truncated for display ...]\nUSER: No, thank you\nASSISTANT: Okay, have a good day"
    }
  ]
}

Illustrative response:

{
  "intent": "delivery_time_inquiry",
  "sentiment": "neutral",
  "resolved": true,
  "escalation_needed": false,
  "summary": "Customer asked why their delivery time seemed wrong; agent confirmed the order is on schedule for standard 5-day shipping."
}

This particular row also carries gold metadata passed through from the source (target_json = {"intent": "status_delivery_time", ...}, has_gold_intent=true), which a future quality check could compare against — the throughput benchmark itself does not score outputs.

Long meeting segment (long_dialogue, AMI window, 4,805 input tokens, 4k_8k bucket): identical request shape, but with ~4.8k tokens of speaker-prefixed meeting transcript in the user message and max_tokens: 171. Such rows are prefill-dominated (prefill_ratio_estimate ≈ 0.97), in contrast to the short case above.

Notes: the 34-token static prefix is identical across all rows, so engines with prefix caching will reuse it (static_prefix_tokens / static_prefix_reuse quantify this per row). If your server supports response_format: {"type": "json_object"}, it keeps outputs compact, though the prompt already encourages that and correctness is not evaluated.

Splits

  • source — Customer-relevant transcript-analytics prompts (demo believability). Mostly 512–4k-token task-oriented customer-service dialogues.
  • long_dialogue — Longer transcript-like prompts for long-context throughput and throughput-prediction validation, covering 512 to 16k input tokens with most rows between 2k and 8k.
  • short_control — Short realistic workloads (128–1024 input tokens) for benchmarking short-prompt serving behaviour.
  • holdout_eval — Held out for predicted-vs-observed throughput validation ONLY. Not a quality-eval split. Do not use it to fit or select anything. Transcripts are disjoint from all other splits (normalized-hash check) and drawn from all source families and prompt lengths.

Source mix by split

  • source (1566 rows): Taskmaster2: 544, SalesBot: 363, Taskmaster3: 331, MULTIWOZ2_2: 232, ABCD: 61, SGD: 35
  • long_dialogue (1217 rows): QMSum: 716, ICSI: 273, AMI: 228
  • short_control (3000 rows): ABCD: 545, SGD: 495, Taskmaster3: 488, Taskmaster2: 459, MULTIWOZ2_2: 434, bitext/Bitext-customer-support-llm-chatbot-training-dataset: 391, SalesBot: 188
  • holdout_eval (1049 rows): QMSum: 200, Taskmaster2: 167, Taskmaster3: 150, MULTIWOZ2_2: 107, SGD: 93, ABCD: 88, SalesBot: 85, bitext/Bitext-customer-support-llm-chatbot-training-dataset: 61, ICSI: 59, AMI: 39

Token statistics

split rows mean in-tok p50 p75 p90 p95 p99 mean exp-new prefill decode prefix-reuse p90
source 1566 599.8 569 634 712 793 908 104.9 0.8492 0.1508 0.0654
long_dialogue 1217 3581.4 3045 5654 6542 7652 13680 150.4 0.9414 0.0586 0.0407
short_control 3000 304.0 300 384 441 472 504 100.3 0.7358 0.2642 0.2037
holdout_eval 1049 1163.5 520 863 3240 5443 7476 113.6 0.8224 0.1776 0.1576
overall 6832 1087.6 495 675 3233 5772 7709 112.3 0.8117 0.1883 0.1659

Length-bucket counts

split short_under_512 512_1k 1k_2k 2k_4k 4k_8k 8k_16k
source 0 1560 6 0 0 0
long_dialogue 0 184 246 369 369 49
short_control 3000 0 0 0 0 0
holdout_eval 491 313 77 94 65 9

Length buckets on input_tokens: short_under_512 (<512), 512_1k, 1k_2k, 2k_4k, 4k_8k (4096–8192 inclusive), 8k_16k (8193–16384).

Licenses and redistribution

Each row carries license, redistribution_status, and source_attribution columns. Per-source terms:

source config license redistribution status
Salesforce/dialogstudio ABCD MIT redistributable
Salesforce/dialogstudio Taskmaster2 CC BY 4.0 redistributable
Salesforce/dialogstudio Taskmaster3 CC BY 4.0 redistributable
Salesforce/dialogstudio SGD CC BY-SA 4.0 redistributable (share-alike)
Salesforce/dialogstudio SalesBot none (no upstream license) no-upstream-license; included by dataset owner decision (fully synthetic simulator/GPT-3 text)
Salesforce/dialogstudio MULTIWOZ2_2 MIT redistributable
Salesforce/dialogstudio AMI CC BY 4.0 redistributable
Salesforce/dialogstudio ICSI CC BY 4.0 redistributable
Salesforce/dialogstudio QMSum MIT redistributable
bitext/Bitext-customer-support-llm-chatbot-training-dataset CDLA-Sharing-1.0 redistributable (share-alike for data)

Notes: SGD-derived rows remain CC BY-SA 4.0 (share-alike); Bitext-derived rows remain CDLA-Sharing-1.0. SalesBot has no upstream license; it is included by explicit dataset-owner decision because its text is fully synthetic (user-simulator/GPT-3 generated) with no third-party human content — treat those rows accordingly if that risk profile does not fit your use.

Excluded sources

The following sources from the original target list were excluded (no raw rows included) because their terms do not clearly allow public redistribution:

Construction

  • Built from raw Salesforce/dialogstudio JSON files (configs listed in metadata.jsonsource_files_used; for large task-oriented configs a deterministic subset of dialogue files was used) plus the Bitext CSV.
  • Task-oriented dialogues are flattened from the DialogStudio log field into USER: / ASSISTANT: turns; meeting transcripts come from original dialog info → dialog history as SPEAKER: utterance lines.
  • Long-meeting segmentation: AMI/ICSI/QMSum meetings run 9k–18k+ tokens, so rows in the 1k–8k buckets are contiguous, non-overlapping segments of a single meeting (never concatenations of unrelated transcripts). QMSum is cut on its own annotated topic spans; AMI/ICSI use utterance-aligned windows. A meeting contributes either windows or one whole-meeting row (≤16k), never both, to avoid shared long prefixes that would skew prefix-cache behaviour.
  • Rows over 16384 input tokens were rejected (never truncated) and are counted per source in source_report.json. Deduplication is by normalized transcript hash; holdout_eval transcripts are disjoint from all other splits.
  • Built by a deterministic fixed-seed pipeline (seed 42).

Label policy

  • Existing source labels are preserved when trivially available (reference_intent from ABCD subflow / SalesBot intent / Bitext intent; reference_summary from AMI abstractive summaries and QMSum query answers).
  • No labels are synthesized — nothing is LLM-generated, heuristically inferred, or guessed from text. has_gold_* flags are true only for fields that came directly from source data.
  • No source provides sentiment, resolution, or escalation labels, so reference_sentiment, reference_resolved, reference_escalation_needed are null everywhere and the corresponding has_gold_* flags are false.
  • target_json is assembled only from available source fields and may be partial or null.
  • The benchmark runs the model on prompt and generates fresh outputs; the reference fields are optional metadata for future quality checks only.

Limitations

  • Several sources are synthetic or proxy data (SalesBot and Bitext are fully synthetic; Taskmaster/SGD/MultiWOZ are Wizard-of-Oz or simulated task-oriented dialogues) — they look like customer service but are not real support calls.
  • Meeting transcripts (AMI/ICSI/QMSum) are research-meeting recordings, not customer service; they provide realistic long-transcript shape, not customer-support content.
  • Long rows are contiguous segments of longer meetings (see Construction).
  • long_dialogue is modest in size because most long-form sources (MediaSum, MeetingBank, SAMSum, DialogSum) failed license review.
  • Intended for inference throughput experiments, not model quality evaluation.

Files

  • data/<split>-*.parquet — the four splits
  • metadata.json — build configuration (seed, tokenizer, template, files used)
  • token_stats.json — per-split token statistics
  • source_report.json — per-source row accounting, license evidence, exclusions
Downloads last month
63