[Feature]: Hugging Face Inference API Integration

[felixsuarez0727]

Docs: https://huggingface.co/inference-api

[felixsuarez0727]

As of now, tree actions have been added. Need to add seven more, at least, and work on computer vision sample app.

[felixsuarez0727]

As of now, I have added eight actions. Each of them correspond to a different machine learning model.

[felixsuarez0727]

A minor sample app is also ready:

https://app.appsmith.com/app/test-hugging-face/page1-6258fbf50d3d384069c11017

[felixsuarez0727]

Moving to QA since the integration and the sample app is ready.

[btsgh]

Testcases are available here -

1. Basic Testcase
2. Bert-base-uncased-hateexplain
3. Classificacion-texto-suicida-finetuned-amazon-review
4. dehatebert-mono-spanish (Sentiment Analysis)
5. Detect Acoso TwitterEs
6. Discriminator finetuned clasificacion comentariossuicidas
7. DistilGPT2
8. Emotion English DistilRoBERTa-base (Sentiment Analysis)

[btsgh]

Issues found so far -

1. In Bert-base-uncased-hateexplain - testcase point 2 fails as text is spelt incorrectly. It could be also reworded to - Please enter your sentence here... This issue occurs across all the commands

2. Testcase point 3 - FAILS CURRENTLY - it runs even though the field is blank. It should give an appropriate message that required field is missing. This issue occurs across all the commands

3. Testcase point 6 - FAILS sometimes - an error pops up.

4. In the "Classificacion-texto-suicida-finetuned-amazon-review" command, testcase point 4 - FAILS CURRENTLY - it gives an output exactly the reverse of the expected output. Actual result - Expected result -

5. Testcase5 in Classificacion-texto-suicida-finetuned-amazon-review enter an english text equivalent - (FAILS CURRENTLY - is English supported?)

6. Testcase4 in dehatebert-mono-spanish (Sentiment Analysis) - (FAILS CURRENTLY - it gives an exact opposite response)

7. Testcase5 in dehatebert-mono-spanish (Sentiment Analysis) - Currently gives a slightly different response number as below: [ [ { "label": "NON_HATE", "score": 0.7812689542770386 }, { "label": "HATE", "score": 0.218731090426445 } ] ]

8. Detect Acoso TwitterEs - Testcase4 - FAILS CURRENTLY - gives response which is opposite to the expected.

9. Detect Acoso TwitterEs - Testcase5 - Output is different from what is seen on this url

10. Discriminator finetuned clasificacion comentariossuicidas - Testcase4 - Response is not the same as seen in the comparison URL.

11. Discriminator finetuned clasificacion comentariossuicidas - Testcase5 - Currently fails as the text entered gives an opposite response as below: [ [ { "label": "Texto no suicida", "score": 0.24806660413742065 }, { "label": "Texto Suicida", "score": 0.7519333958625793 } ] ]

12. DistilGPT2 - Testcase4 fails as the response we get is as below - [ { "generated_text": "text=My%20name%20is%20Clara%20and%20I%20am%20a(%20a),%20a=My%20name%20i%20the%20a%20a" }

Expected response as per this url - [ { "generated_text": "My name is Clara and I am a graduate student. I am working in government and business. I can't tell you that I am a little more than a good citizen of California. I do have some personal problems with my job due to social work" } ]

13. Emotion English DistilRoBERTa-base (Sentiment Analysis) - Testcase4 - The response received on our interface is different and is as below - [ [ { "label": "anger", "score": 0.11598271876573563 }, { "label": "disgust", "score": 0.00854487530887127 }, { "label": "fear", "score": 0.02401498146355152 }, { "label": "joy", "score": 0.039270929992198944 }, { "label": "neutral", "score": 0.26834872364997864 }, { "label": "sadness", "score": 0.4954221546649933 }, { "label": "surprise", "score": 0.04841558635234833 } ] ] Comparing with the url - https://huggingface.co/j-hartmann/emotion-english-distilroberta-base, the above values are different on our appsmith interface
14. Emotion English DistilRoBERTa-base (Sentiment Analysis) - Testcase6 - This currently fails. Neutral shows the largest value.
15. Emotion English DistilRoBERTa-base (Sentiment Analysis) - Testcase7 - This currently fails. Joy shows the largest value.
16. Emotion English DistilRoBERTa-base (Sentiment Analysis) - Testcase8 - This currently fails. Neutral shows the largest value.
17. Emotion English DistilRoBERTa-base (Sentiment Analysis) - Testcase10 - This currently does not give the same response as seen in the above url, though anger shows the largest value of 0.5.

[felixsuarez0727]

Hugging Face Observations

#	Question/Observation	Related Issue	Resolved
1	In Bert-base-uncased-hateexplain - testcase point 2 fails as text is spelt incorrectly	-	No
2	It runs even though the field is blank. It should give an appropriate message that required field is missing.	-	No
3	FAILS sometimes - an error pops up	-	No
4	It gives an output exactly the reverse of the expected output.	-	No
5	FAILS CURRENTLY - is English supported?	-	No
6	Gives an exact opposite response	-	No
7	Currently gives a slightly different response number	-	No
8	Gives an exact opposite response	-	No
9	Output is different	-	No
10	Response is not the same as seen in the comparison URL.	-	No
11	Gives an exact opposite response	-	No
12	Fails as the response we get	-	No
13	The response received on our interface is different	-	No
14	Currently fails. Neutral shows the largest value.	-	No
15	Currently fails. Joy shows the largest value.	-	No
16	Currently fails. Neutral shows the largest value.	-	No
17	Does not give the same response as seen in the above url	-	No

Updates about observations

1. We are going to change this in all the commands

Note: For the other test we need to do more research on the problem

[rohan-arthur]

@felixsuarez0727 can we move this to icebox since we are not actively working on it?

[rohan-arthur]

closing this because we are not working on this currently