Inference API documentation

Detailed parameters

Join the Hugging Face community

and get access to the augmented documentation experience

to get started

# Detailed parameters

## Which task is used by this model ?

In general the 🤗 Hosted API Inference accepts a simple string as an input. However, more advanced usage depends on the “task” that the model solves.

The “task” of a model is defined here on it’s model page:

## Natural Language Processing

### Fill Mask task

Tries to fill in a hole with a missing word (token to be precise). That’s the base task for BERT models.

Recommended model: bert-base-uncased (it’s a simple model, but fun to play with).

Available with: 🤗 Transformers

Example:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/bert-base-uncased"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query({"inputs": "The answer to the universe is [MASK]."})

When sending your request, you should send a JSON encoded payload. Here are all the options

All parameters
inputs (required): a string to be filled from, must contain the [MASK] token (check model card for exact name of the mask)
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

Return value is either a dict or a list of dicts if you sent a list of inputs

Python
JavaScript
cURL
self.assertEqual(
deep_round(data),
[
{
"sequence": "the answer to the universe is no.",
"score": 0.1696,
"token": 2053,
"token_str": "no",
},
{
"sequence": "the answer to the universe is nothing.",
"score": 0.0734,
"token": 2498,
"token_str": "nothing",
},
{
"sequence": "the answer to the universe is yes.",
"score": 0.0580,
"token": 2748,
"token_str": "yes",
},
{
"sequence": "the answer to the universe is unknown.",
"score": 0.044,
"token": 4242,
"token_str": "unknown",
},
{
"sequence": "the answer to the universe is simple.",
"score": 0.0402,
"token": 3722,
"token_str": "simple",
},
],
)
Returned values
sequence The actual sequence of tokens that ran against the model (may contain special tokens)
score The probability for this token.
token The id of the token
token_str The string representation of the token

### Summarization task

This task is well known to summarize longer text into shorter text. Be careful, some models have a maximum length of input. That means that the summary cannot handle full books for instance. Be careful when choosing your model. If you want to discuss your summarization needs, please get in touch with us: <api-enterprise@huggingface.co>

Recommended model: facebook/bart-large-cnn.

Available with: 🤗 Transformers

Example:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/facebook/bart-large-cnn"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query(
{
"inputs": "The tower is 324 metres (1,063 ft) tall, about the same height as an 81-storey building, and the tallest structure in Paris. Its base is square, measuring 125 metres (410 ft) on each side. During its construction, the Eiffel Tower surpassed the Washington Monument to become the tallest man-made structure in the world, a title it held for 41 years until the Chrysler Building in New York City was finished in 1930. It was the first structure to reach a height of 300 metres. Due to the addition of a broadcasting aerial at the top of the tower in 1957, it is now taller than the Chrysler Building by 5.2 metres (17 ft). Excluding transmitters, the Eiffel Tower is the second tallest free-standing structure in France after the Millau Viaduct.",
"parameters": {"do_sample": False},
}
)
# Response
self.assertEqual(
data,
[
{
"summary_text": "The tower is 324 metres (1,063 ft) tall, about the same height as an 81-storey building. Its base is square, measuring 125 metres (410 ft) on each side. During its construction, the Eiffel Tower surpassed the Washington Monument to become the tallest man-made structure in the world.",
},
],
)

When sending your request, you should send a JSON encoded payload. Here are all the options

All parameters
inputs (required) a string to be summarized
parameters a dict containing the following keys:
min_length (Default: None). Integer to define the minimum length in tokens of the output summary.
max_length (Default: None). Integer to define the maximum length in tokens of the output summary.
top_k (Default: None). Integer to define the top tokens considered within the sample operation to create new text.
top_p (Default: None). Float to define the tokens that are within the sample operation of text generation. Add tokens in the sample for more probable to least probable until the sum of the probabilities is greater than top_p.
temperature (Default: 1.0). Float (0.0-100.0). The temperature of the sampling operation. 1 means regular sampling, 0 means always take the highest score, 100.0 is getting closer to uniform probability.
repetition_penalty (Default: None). Float (0.0-100.0). The more a token is used within generation the more it is penalized to not be picked in successive generation passes.
max_time (Default: None). Float (0-120.0). The amount of time in seconds that the query should take maximum. Network can cause some overhead so it will be a soft limit.
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

Return value is either a dict or a list of dicts if you sent a list of inputs

Returned values
summary_text The string after summarization

### Question Answering task

Want to have a nice know-it-all bot that can answer any question?

Recommended model: deepset/roberta-base-squad2.

Available with: 🤗Transformers and AllenNLP

Example:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/deepset/roberta-base-squad2"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query(
{
"inputs": {
"question": "What's my name?",
"context": "My name is Clara and I live in Berkeley.",
}
}
)

When sending your request, you should send a JSON encoded payload. Here are all the options

Return value is a dict.

Python
JavaScript
cURL
self.assertEqual(
deep_round(data),
{"score": 0.9327, "start": 11, "end": 16, "answer": "Clara"},
)
Returned values
answer A string that’s the answer within the text.
score A float that represents how likely that the answer is correct
start The index (string wise) of the start of the answer within context.
stop The index (string wise) of the stop of the answer within context.

### Table Question Answering task

Don’t know SQL? Don’t want to dive into a large spreadsheet? Ask questions in plain english!

Available with: 🤗 Transformers

Example:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/google/tapas-base-finetuned-wtq"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query(
{
"inputs": {
"query": "How many stars does the transformers repository have?",
"table": {
"Repository": ["Transformers", "Datasets", "Tokenizers"],
"Stars": ["36542", "4512", "3934"],
"Contributors": ["651", "77", "34"],
"Programming language": [
"Python",
"Python",
"Rust, Python and NodeJS",
],
},
}
}
)

When sending your request, you should send a JSON encoded payload. Here are all the options

All parameters
inputs (required)
query (required) The query in plain text that you want to ask the table
table (required) A table of data represented as a dict of list where entries are headers and the lists are all the values, all lists must have the same size.
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

Return value is either a dict or a list of dicts if you sent a list of inputs

Python
JavaScript
cURL
self.assertEqual(
data,
{
"answer": "AVERAGE > 36542",
"coordinates": [[0, 1]],
"cells": ["36542"],
"aggregator": "AVERAGE",
},
)
Returned values
answer The plaintext answer
coordinates a list of coordinates of the cells referenced in the answer
cells a list of coordinates of the cells contents
aggregator The aggregator used to get the answer

### Sentence Similarity task

Calculate the semantic similarity between one text and a list of other sentences by comparing their embeddings.

Available with: Sentence Transformers

Example:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/sentence-transformers/all-MiniLM-L6-v2"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query(
{
"inputs": {
"source_sentence": "That is a happy person",
"sentences": ["That is a happy dog", "That is a very happy person", "Today is a sunny day"],
}
}
)

When sending your request, you should send a JSON encoded payload. Here are all the options

All parameters
inputs (required)
source_sentence (required) The string that you wish to compare the other strings with. This can be a phrase, sentence, or longer passage, depending on the model being used.
sentences (required) A list of strings which will be compared against the source_sentence.
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

The return value is a list of similarity scores, given as floats.

Python
JavaScript
cURL
self.assertEqual(
data,
[0.6945773363113403, 0.9429150819778442, 0.2568760812282562],
)
Returned values
Scores The associated similarity score for each of the given strings

### Text Classification task

Usually used for sentiment-analysis this will output the likelihood of classes of an input.

Available with: 🤗 Transformers

Example:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/distilbert-base-uncased-finetuned-sst-2-english"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query({"inputs": "I like you. I love you"})

When sending your request, you should send a JSON encoded payload. Here are all the options

All parameters
inputs (required) a string to be classified
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

Return value is either a dict or a list of dicts if you sent a list of inputs

Python
JavaScript
cURL
self.assertEqual(
deep_round(data),
[
[
{"label": "POSITIVE", "score": 0.9999},
{"label": "NEGATIVE", "score": 0.0001},
]
],
)
Returned values
label The label for the class (model specific)
score A floats that represents how likely is that the text belongs the this class.

### Text Generation task

Use to continue text from a prompt. This is a very generic task.

Recommended model: gpt2 (it’s a simple model, but fun to play with).

Available with: 🤗 Transformers

Example:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/gpt2"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query({"inputs": "The answer to the universe is"})

When sending your request, you should send a JSON encoded payload. Here are all the options

All parameters
inputs (required): a string to be generated from
parameters dict containing the following keys:
top_k (Default: None). Integer to define the top tokens considered within the sample operation to create new text.
top_p (Default: None). Float to define the tokens that are within the sample operation of text generation. Add tokens in the sample for more probable to least probable until the sum of the probabilities is greater than top_p.
temperature (Default: 1.0). Float (0.0-100.0). The temperature of the sampling operation. 1 means regular sampling, 0 means always take the highest score, 100.0 is getting closer to uniform probability.
repetition_penalty (Default: None). Float (0.0-100.0). The more a token is used within generation the more it is penalized to not be picked in successive generation passes.
max_new_tokens (Default: None). Int (0-250). The amount of new tokens to be generated, this does not include the input length it is a estimate of the size of generated text you want. Each new tokens slows down the request, so look for balance between response times and length of text generated.
max_time (Default: None). Float (0-120.0). The amount of time in seconds that the query should take maximum. Network can cause some overhead so it will be a soft limit. Use that in combination with max_new_tokens for best results.
return_full_text (Default: True). Bool. If set to False, the return results will not contain the original query making it easier for prompting.
num_return_sequences (Default: 1). Integer. The number of proposition you want to be returned.
do_sample (Optional: True). Bool. Whether or not to use sampling, use greedy decoding otherwise.
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

Return value is either a dict or a list of dicts if you sent a list of inputs

Python
JavaScript
cURL
data == [
{
"generated_text": 'The answer to the universe is that we are the creation of the entire universe," says Fitch.\n\nAs of the 1960s, six times as many Americans still make fewer than six bucks (\$17) per year on their way to retirement.'
}
]
Returned values
generated_text The continuated string

### Text2Text Generation task

Essentially Text-generation task. But uses Encoder-Decoder architecture, so might change in the future for more options.

### Token Classification task

Usually used for sentence parsing, either grammatical, or Named Entity Recognition (NER) to understand keywords contained within text.

Available with: 🤗 Transformers, Flair

Example:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/dbmdz/bert-large-cased-finetuned-conll03-english"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query({"inputs": "My name is Sarah Jessica Parker but you can call me Jessica"})

When sending your request, you should send a JSON encoded payload. Here are all the options

All parameters
inputs (required) a string to be classified
parameters a dict containing the following key:
aggregation_strategy (Default: simple). There are several aggregation strategies:
none: Every token gets classified without further aggregation.
simple: Entities are grouped according to the default schema (B-, I- tags get merged when the tag is similar).
first: Same as the simple strategy except words cannot end up with different tags. Words will use the tag of the first token when there is ambiguity.
average: Same as the simple strategy except words cannot end up with different tags. Scores are averaged across tokens and then the maximum label is applied.
max: Same as the simple strategy except words cannot end up with different tags. Word entity will be the token with the maximum score.
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

Return value is either a dict or a list of dicts if you sent a list of inputs

Python
JavaScript
cURL
self.assertEqual(
deep_round(data),
[
{
"entity_group": "PER",
"score": 0.9991,
"word": "Sarah Jessica Parker",
"start": 11,
"end": 31,
},
{
"entity_group": "PER",
"score": 0.998,
"word": "Jessica",
"start": 52,
"end": 59,
},
],
)
Returned values
entity_group The type for the entity being recognized (model specific).
score How likely the entity was recognized.
word The string that was captured
start The offset stringwise where the answer is located. Useful to disambiguate if word occurs multiple times.
end The offset stringwise where the answer is located. Useful to disambiguate if word occurs multiple times.

### Translation task

This task is well known to translate text from one language to another

Recommended model: Helsinki-NLP/opus-mt-ru-en. Helsinki-NLP uploaded many models with many language pairs. Recommended model: t5-base.

Available with: 🤗 Transformers

Example:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/Helsinki-NLP/opus-mt-ru-en"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query(
{
"inputs": "Меня зовут Вольфганг и я живу в Берлине",
}
)
# Response
self.assertEqual(
data,
[
{
"translation_text": "My name is Wolfgang and I live in Berlin.",
},
],
)

When sending your request, you should send a JSON encoded payload. Here are all the options

All parameters
inputs (required) a string to be translated in the original languages
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

Return value is either a dict or a list of dicts if you sent a list of inputs

Returned values
translation_text The string after translation

### Zero-Shot Classification task

This task is super useful to try out classification with zero code, you simply pass a sentence/paragraph and the possible labels for that sentence, and you get a result.

Recommended model: facebook/bart-large-mnli.

Available with: 🤗 Transformers

Request:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/facebook/bart-large-mnli"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query(
{
"inputs": "Hi, I recently bought a device from your company but it is not working as advertised and I would like to get reimbursed!",
"parameters": {"candidate_labels": ["refund", "legal", "faq"]},
}
)

When sending your request, you should send a JSON encoded payload. Here are all the options

All parameters
inputs (required) a string or list of strings
parameters (required) a dict containing the following keys:
candidate_labels (required) a list of strings that are potential classes for inputs. (max 10 candidate_labels, for more, simply run multiple requests, results are going to be misleading if using too many candidate_labels anyway. If you want to keep the exact same, you can simply run multi_label=True and do the scaling on your end. )
multi_label (Default: false) Boolean that is set to True if classes can overlap
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

Return value is either a dict or a list of dicts if you sent a list of inputs

Response:

Python
JavaScript
cURL
self.assertEqual(
deep_round(data),
{
"sequence": "Hi, I recently bought a device from your company but it is not working as advertised and I would like to get reimbursed!",
"labels": ["refund", "faq", "legal"],
"scores": [
# 88% refund
0.8778,
0.1052,
0.017,
],
},
)
Returned values
sequence The string sent as an input
labels The list of strings for labels that you sent (in order)
scores a list of floats that correspond the the probability of label, in the same order as labels.

### Conversational task

This task corresponds to any chatbot like structure. Models tend to have shorter max_length, so please check with caution when using a given model if you need long range dependency or not.

Recommended model: microsoft/DialoGPT-large.

Available with: 🤗 Transformers

Example:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/microsoft/DialoGPT-large"
def query(payload):
data = json.dumps(payload)
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query(
{
"inputs": {
"past_user_inputs": ["Which movie is the best ?"],
"generated_responses": ["It's Die Hard for sure."],
"text": "Can you explain why ?",
},
}
)
# Response
self.assertEqual(
data,
{
"generated_text": "It's the best movie ever.",
"conversation": {
"past_user_inputs": [
"Which movie is the best ?",
"Can you explain why ?",
],
"generated_responses": [
"It's Die Hard for sure.",
"It's the best movie ever.",
],
},
"warnings": ["Setting pad_token_id to eos_token_id:50256 for open-end generation."],
},
)

When sending your request, you should send a JSON encoded payload. Here are all the options

All parameters
inputs (required)
text (required) The last input from the user in the conversation.
generated_responses A list of strings corresponding to the earlier replies from the model.
past_user_inputs A list of strings corresponding to the earlier replies from the user. Should be of the same length of generated_responses.
parameters a dict containing the following keys:
min_length (Default: None). Integer to define the minimum length in tokens of the output summary.
max_length (Default: None). Integer to define the maximum length in tokens of the output summary.
top_k (Default: None). Integer to define the top tokens considered within the sample operation to create new text.
top_p (Default: None). Float to define the tokens that are within the sample operation of text generation. Add tokens in the sample for more probable to least probable until the sum of the probabilities is greater than top_p.
temperature (Default: 1.0). Float (0.0-100.0). The temperature of the sampling operation. 1 means regular sampling, 0 means always take the highest score, 100.0 is getting closer to uniform probability.
repetition_penalty (Default: None). Float (0.0-100.0). The more a token is used within generation the more it is penalized to not be picked in successive generation passes.
max_time (Default: None). Float (0-120.0). The amount of time in seconds that the query should take maximum. Network can cause some overhead so it will be a soft limit.
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

Return value is either a dict or a list of dicts if you sent a list of inputs

Returned values
generated_text The answer of the bot
conversation A facility dictionnary to send back for the next input (with the new user input addition).
past_user_inputs List of strings. The last inputs from the user in the conversation, <em>after the model has run.
generated_responses List of strings. The last outputs from the model in the conversation, <em>after the model has run.

### Feature Extraction task

This task reads some text and outputs raw float values, that are usually consumed as part of a semantic database/semantic search.

Recommended model: Sentence-transformers.

Available with: 🤗 Transformers Sentence-transformers

Request:

All parameters
inputs (required): a string or a list of strings to get the features from.
options a dict containing the following keys:
use_cache (Default: true). Boolean. There is a cache layer on the inference API to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
wait_for_model (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.

Return value is either a dict or a list of dicts if you sent a list of inputs

Returned values
A list of float (or list of list of floats) The numbers that are the representation features of the input.
Returned values are a list of floats, or a list of list of floats (depending on if you sent a string or a list of string, and if the automatic reduction, usually mean_pooling for instance was applied for you or not. This should be explained on the model's README.

## Audio

### Automatic Speech Recognition task

This task reads some audio input and outputs the said words within the audio files.

Recommended model: Check your langage.

Available with: 🤗 Transformers ESPnet and SpeechBrain

Request:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/facebook/wav2vec2-base-960h"
def query(filename):
with open(filename, "rb") as f:
data = f.read()
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query("sample1.flac")

When sending your request, you should send a binary payload that simply contains your audio file. We try to support most formats (Flac, Wav, Mp3, Ogg etc...). And we automatically rescale the sampling rate to the appropriate rate for the given model (usually 16KHz).

All parameters
no parameter (required) a binary representation of the audio file. No other parameters are currently allowed.

Return value is either a dict or a list of dicts if you sent a list of inputs

Response:

Python
JavaScript
cURL
self.assertEqual(
data,
{
"text": "GOING ALONG SLUSHY COUNTRY ROADS AND SPEAKING TO DAMP AUDIENCES IN DRAUGHTY SCHOOL ROOMS DAY AFTER DAY FOR A FORTNIGHT HE'LL HAVE TO PUT IN AN APPEARANCE AT SOME PLACE OF WORSHIP ON SUNDAY MORNING AND HE CAN COME TO US IMMEDIATELY AFTERWARDS"
},
)
Returned values
text The string that was recognized within the audio file.

### Audio Classification task

This task reads some audio input and outputs the likelihood of classes.

Recommended model: superb/hubert-large-superb-er.

Available with: 🤗 Transformers SpeechBrain

Request:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/superb/hubert-large-superb-er"
def query(filename):
with open(filename, "rb") as f:
data = f.read()
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query("sample1.flac")

When sending your request, you should send a binary payload that simply contains your audio file. We try to support most formats (Flac, Wav, Mp3, Ogg etc...). And we automatically rescale the sampling rate to the appropriate rate for the given model (usually 16KHz).

All parameters
no parameter (required) a binary representation of the audio file. No other parameters are currently allowed.

Return value is a dict

Python
JavaScript
cURL
self.assertEqual(
deep_round(data, 4),
[
{"score": 0.5928, "label": "neu"},
{"score": 0.2003, "label": "hap"},
{"score": 0.128, "label": "ang"},
{"score": 0.079, "label": "sad"},
],
)
Returned values
label The label for the class (model specific)
score A float that represents how likely it is that the audio file belongs to this class.

## Computer Vision

### Image Classification task

This task reads some image input and outputs the likelihood of classes.

Recommended model: google/vit-base-patch16-224.

Available with: 🤗 Transformers

Request:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/google/vit-base-patch16-224"
def query(filename):
with open(filename, "rb") as f:
data = f.read()
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query("cats.jpg")

When sending your request, you should send a binary payload that simply contains your image file. We support all image formats Pillow supports.

All parameters
no parameter (required) a binary representation of the image file. No other parameters are currently allowed.

Return value is a dict

Python
JavaScript
cURL
self.assertEqual(
deep_round(data, 4),
[
{"score": 0.9374, "label": "Egyptian cat"},
{"score": 0.0384, "label": "tabby, tabby cat"},
{"score": 0.0144, "label": "tiger cat"},
{"score": 0.0033, "label": "lynx, catamount"},
{"score": 0.0007, "label": "Siamese cat, Siamese"},
],
)
Returned values
label The label for the class (model specific)
score A float that represents how likely it is that the image file belongs to this class.

### Object Detection task

This task reads some image input and outputs the likelihood of classes & bounding boxes of detected objects.

Recommended model: facebook/detr-resnet-50.

Available with: 🤗 Transformers

Request:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/facebook/detr-resnet-50"
def query(filename):
with open(filename, "rb") as f:
data = f.read()
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query("cats.jpg")

When sending your request, you should send a binary payload that simply contains your image file. We support all image formats Pillow supports.

All parameters
no parameter (required) a binary representation of the image file. No other parameters are currently allowed.

Return value is a dict

Python
JavaScript
cURL
self.assertEqual(
deep_round(data, 4),
[
{
"score": 0.9982,
"label": "remote",
"box": {"xmin": 40, "ymin": 70, "xmax": 175, "ymax": 117},
},
{
"score": 0.9960,
"label": "remote",
"box": {"xmin": 333, "ymin": 72, "xmax": 368, "ymax": 187},
},
{
"score": 0.9955,
"label": "couch",
"box": {"xmin": 0, "ymin": 1, "xmax": 639, "ymax": 473},
},
{
"score": 0.9988,
"label": "cat",
"box": {"xmin": 13, "ymin": 52, "xmax": 314, "ymax": 470},
},
{
"score": 0.9987,
"label": "cat",
"box": {"xmin": 345, "ymin": 23, "xmax": 640, "ymax": 368},
},
],
)
Returned values
label The label for the class (model specific) of a detected object.
score A float that represents how likely it is that the detected object belongs to the given class.
box A dict (with keys [xmin,ymin,xmax,ymax]) representing the bounding box of a detected object.

### Image Segmentation task

This task reads some image input and outputs the likelihood of classes & bounding boxes of detected objects.

Available with: 🤗 Transformers

Request:

Python
JavaScript
cURL
import json
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://api-inference.huggingface.co/models/facebook/detr-resnet-50-panoptic"
def query(filename):
with open(filename, "rb") as f:
data = f.read()
response = requests.request("POST", API_URL, headers=headers, data=data)
return json.loads(response.content.decode("utf-8"))
data = query("cats.jpg")

When sending your request, you should send a binary payload that simply contains your image file. We support all image formats Pillow supports.

All parameters
no parameter (required) a binary representation of the image file. No other parameters are currently allowed.

Return value is a dict

Python
JavaScript
cURL
import base64
from io import BytesIO
from PIL import Image
with Image.open("cats.jpg") as img:
masks = [d["mask"] for d in data]
self.assertEqual(img.size, (640, 480))
mask_imgs = [Image.open(BytesIO(base64.b64decode(mask))) for mask in masks]
for mask_img in mask_imgs:
self.assertEqual(mask_img.size, img.size)
self.assertEqual(mask_img.mode, "L")  # L (8-bit pixels, black and white)
first_mask_img = mask_imgs[0]
min_pxl_val, max_pxl_val = first_mask_img.getextrema()
self.assertGreaterEqual(min_pxl_val, 0)
self.assertLessEqual(max_pxl_val, 255)
Returned values
label The label for the class (model specific) of a segment.
score A float that represents how likely it is that the segment belongs to the given class.
mask A str (base64 str of a single channel black-and-white img) representing the mask of a segment.