Welcome aboard Twitter follower! Not all videos will be of this high quality as this one was a special one XD but it'll still be mostly like this so I hope to still see you around on future videos 😊
Love your content! I would like to see how a model can be used to answer questions about a given dataset, for instance; “What is the sum of column … for all dates after …”. Do you think it is possible?
Thanks, looks like an interesting problem Well as a first try, if your dataset is in a dataframe: ``` data = {"a": ["2", "5", "19"], "b": ["87", "53", "69"], "c": ["1990/02/01", "1995/02/01", "2000/02/01"]} table = pd.DataFrame.from_dict(data) query = "Can you sum b for all rows of column c after year 1993?" pipeline(task="table-question-answering")(table=table, query=query) ``` did retrieve the correct cells for me, then you would need to do the necessary aggregation. Though to be honest it then depends on the kind of queries you want to build. If the vocabulary of queries is limited, I would actually try to convert it manually into a Pandas .query(). If the queries are very complex, and not really looking like SQL queries, I don't know how table QA models will fare. But as long as the query is similar to a SQL query expressed with natural language, Table Q&A huggingface.co/tasks/table-question-answering should work. In the future, you may look into higher level APIs like Haystack (haystack.deepset.ai/tutorials/15_tableqa) or Langchain (python.langchain.com/en/latest/use_cases/tabular.html#) for robust table QA pipelines. Hope it helps! I'm not really an expert in the NLP field ^^, so the HF Discord (huggingface.co/join/discord) or forum (discuss.huggingface.co/) are also good places to ask :) Have a nice day!
Hi Fanilo! Thanks for the overview! What is the best approach for us to build a query machine that will be able to query our dataset? I've tried openai functions for this and it worked. But not able to figure it out using huggingface. Thanks!
Hey there! If I understood the question correctly, I haven't tested any of those approaches but I've seen PandasAI + HF Models work yet being a little slower than OpenAI ( docs.pandas-ai.com/en/latest/LLMs/llms/#huggingface-models , I did a PandasAI prototype video here ua-cam.com/video/30_0j0XYOas/v-deo.html , replacing the LLM argument by a HF Model should work ), and I've seen Huggingface Tools & Agents ( huggingface.co/docs/transformers/transformers_agents & huggingface.co/docs/transformers/custom_tools ) but if you scroll a little on the page it doesn't seem they promote Dataset Q&A that much so I'm not sure it's well integrated compared to text-to-image + there's a little note I'm going to quote "If you’re facing issues, we recommend trying out the OpenAI model which, while sadly not open-source, performs better at this given time." I don't have any real recommendation, but if you test those approaches I'd love to hear back from you :) Have a nice day!
Great content as usual! I found plenty of useful information. Could you pls show us how to make an application using transformer tools which would do exactly what you said at 15:14? Is it possible to make it as you described at 14:46? Thanks
Thanks for the support and for watching :) At 15:14 on video transcription you can read the whisper documentation on Hugging Face: huggingface.co/openai/whisper-large-v2, it has a good starting point. Or watch @1littlecoder's video example using the whisper library directly (ua-cam.com/video/KtAFU_xeHr4/v-deo.html or ua-cam.com/video/ywIyc8l1K1Q/v-deo.html) though it's colab/Gradio. Another rec I have is @AssemblyAI's APIs from a Streamlit app but that's a closed API :) I'll think about a Streamlit video for it, since subtitling quickly a UA-cam video has actually been in my TODO list Have a nice day!
your video has some quality knowledge, but with all respect, I dont understand the sense of that guy speaking in between and you are trying to be funny, it doesn't look funny at all, but weird and annoying, Pleaseeee don't do it next time, its reallyyyy distracting...
WOW , now this is impressive 😍😍😍
Thanks for the support :)
What a high-quality video!
Thanks for watching :) Happy Huggingfacing!
u are awsome , found u during phd work
Hey, thanks for watching! Hope to see you on the next video :)
amazing video. loved it.
Thanks for watching, glad you enjoyed it :D
Great Video ... Informative
Thanks for watching 😁
Congratulations on the sponsor 🔥 Amazing video as usual!
Thank you so much. This video has been a wild ride to edit!
Thank you for putting together an overview! it's definitely a must watch before diving into the hugging face library
Thanks for watching :) good luck on your future Huggingface projects! What will you build?
Wow..!!nice video..!! Thanks
Thanks for watching :D very appreciated
I found you randomly on tweet and not even 30% into the video I subscribed. You are awesome bro 😎
Welcome aboard Twitter follower! Not all videos will be of this high quality as this one was a special one XD but it'll still be mostly like this so I hope to still see you around on future videos 😊
Amazing video 🔥🔥🔥
💖💖 thanks for the support Sophia! Time for a little break now ahah!
@@andfanilo so proud of you! You deserve a nice break and some great food 🙌
@@SophiaYangDS You too :) hope your month went as smoothly as it could be!
Yeah you're right, I'm going to treat myself now, some pizza ould be nice ☺
Such a wonderful video. Funny, informative, clear. Keep up the awesome work!
Oh, thank you so much for the positive feedback, you made my day :) what are you looking into building?
I'll keep doing it, I hope to see you around!
Awesome work! Like the vibe, it was definitely entertaining to watch :)
Thanks, glad you enjoyed this once in a while highly edited video, I'll go back to low-level video recordings now 😆
such a dynamic hands-on into to HF transformers! Great content! :)
🔥🔥 thanks for watching! Hope it got you pumped up to building a new HF demo :D
Love your content! I would like to see how a model can be used to answer questions about a given dataset, for instance; “What is the sum of column … for all dates after …”. Do you think it is possible?
Thanks, looks like an interesting problem
Well as a first try, if your dataset is in a dataframe:
```
data = {"a": ["2", "5", "19"], "b": ["87", "53", "69"], "c": ["1990/02/01", "1995/02/01", "2000/02/01"]}
table = pd.DataFrame.from_dict(data)
query = "Can you sum b for all rows of column c after year 1993?"
pipeline(task="table-question-answering")(table=table, query=query)
```
did retrieve the correct cells for me, then you would need to do the necessary aggregation.
Though to be honest it then depends on the kind of queries you want to build. If the vocabulary of queries is limited, I would actually try to convert it manually into a Pandas .query(). If the queries are very complex, and not really looking like SQL queries, I don't know how table QA models will fare. But as long as the query is similar to a SQL query expressed with natural language, Table Q&A huggingface.co/tasks/table-question-answering should work.
In the future, you may look into higher level APIs like Haystack (haystack.deepset.ai/tutorials/15_tableqa) or Langchain (python.langchain.com/en/latest/use_cases/tabular.html#) for robust table QA pipelines.
Hope it helps! I'm not really an expert in the NLP field ^^, so the HF Discord (huggingface.co/join/discord) or forum (discuss.huggingface.co/) are also good places to ask :)
Have a nice day!
Very helpful, thank you so much!❤
Hi Fanilo! Thanks for the overview! What is the best approach for us to build a query machine that will be able to query our dataset? I've tried openai functions for this and it worked. But not able to figure it out using huggingface. Thanks!
Hey there!
If I understood the question correctly, I haven't tested any of those approaches but
I've seen PandasAI + HF Models work yet being a little slower than OpenAI ( docs.pandas-ai.com/en/latest/LLMs/llms/#huggingface-models , I did a PandasAI prototype video here ua-cam.com/video/30_0j0XYOas/v-deo.html , replacing the LLM argument by a HF Model should work ),
and I've seen Huggingface Tools & Agents ( huggingface.co/docs/transformers/transformers_agents & huggingface.co/docs/transformers/custom_tools ) but if you scroll a little on the page it doesn't seem they promote Dataset Q&A that much so I'm not sure it's well integrated compared to text-to-image + there's a little note I'm going to quote "If you’re facing issues, we recommend trying out the OpenAI model which, while sadly not open-source, performs better at this given time."
I don't have any real recommendation, but if you test those approaches I'd love to hear back from you :)
Have a nice day!
Look like we gonna all end by having a J.A.R.V.I.S like at home running on a solar SBC just waiting a 64threads model
Great content as usual! I found plenty of useful information. Could you pls show us how to make an application using transformer tools which would do exactly what you said at 15:14? Is it possible to make it as you described at 14:46? Thanks
Thanks for the support and for watching :)
At 15:14 on video transcription you can read the whisper documentation on Hugging Face: huggingface.co/openai/whisper-large-v2, it has a good starting point.
Or watch @1littlecoder's video example using the whisper library directly (ua-cam.com/video/KtAFU_xeHr4/v-deo.html or ua-cam.com/video/ywIyc8l1K1Q/v-deo.html) though it's colab/Gradio.
Another rec I have is @AssemblyAI's APIs from a Streamlit app but that's a closed API :)
I'll think about a Streamlit video for it, since subtitling quickly a UA-cam video has actually been in my TODO list
Have a nice day!
can I get the link for the neural style transfer you were showing 0:16
This one style-transfer-webrtc.streamlit.app/ ?
your video has some quality knowledge, but with all respect, I dont understand the sense of that guy speaking in between and you are trying to be funny, it doesn't look funny at all, but weird and annoying, Pleaseeee don't do it next time, its reallyyyy distracting...
Ah that's too bad, thanks for the feedback though
Have a nice day!
❤ The best 🎉 video for learning 🫡
Thank you for the support, much appreciated 😍