Welcome to my KZread channel!(formerly known as Henry AI Labs) I am very excited about Deep Learning and AI powered technology!
I am making paper summaries that I hope you will find useful for staying up to date with new papers, at least to give an overview if you don't have time to digest the full paper.
This channel includes topics such as Computer Vision, Natural Language Processing, Graph Embeddings, Generative Adversarial Networks, Reinforcement Learning, and more.
I also try to post coding videos occasionally and am working on developing a podcast!
Thanks for checking it out, please subscribe!
Пікірлер
@3:46, how can low reach to high with search action?
🎯 Key points for quick navigation: Exciting new framework Programming model optimization Graph computation programs 09:55 *Signature, dock string* 10:10 *Prompt optimization, syntax* 10:38 *Input, output fields* 10:52 *Control flow, loops* 11:20 *UAPI, web queries* 12:26 *DSPy Assertions, suggestions* 13:49 *Citation attribution suggestions* 14:16 *Optimization, instructions, examples* 14:59 *DpY as PyTorch* 16:31 *Inductive biases, depth* 17:43 *Intermediate supervision, DpY compiler* 19:22 *Testing with programs* 19:35 *Optimizing instructions and examples* 20:03 *Automatic data labeling* 20:46 *Ending manual prompts* 21:14 *Adapting to new models* 22:49 *Structured output with prompts* 25:33 *Fine-tuning neural networks* 26:26 *Using few-shot examples* 27:51 *Bootstrapping rationales* 28:19 *Evaluating synthetic examples* Overlapping keywords metrics LM judge prompt LM produce metric 37:58 *Deep learning paradigm shift* 38:41 *Data set formatting* 40:03 *Inspect intermediate outputs* 40:59 *Add Chain of Thought* 43:35 *Define optimization metric* 45:13 *Value of reasoning* 45:28 *Inspect parameters* 46:25 *Multi-hop search integration* Queries connected to final answer Introduction to multi-hop search Supervision on intermediate hops Made with HARPA AI
The recipe is gone
Hey Richard! Sorry we refactored recipes! The links are now fixed!
😊😊😊😊
😊😊😊😊😊
👏🏿👏🏿👏🏿👏🏿👏🏿👏🏿👏🏿👏🏿👏🏿👏🏿👏🏿👏🏿👏🏿😁😁👏🏿💯💯💯💯💯
short and sweet, gets to the point. marvelous video
Hi Connor, thanks for the awesome content. I have one small suggestion - Instead of covering maximum information, if it was topic by topic it would be more better. Example: In depth Information on 1 topic "Optimizers (formerly Teleprompters)". Thank you🙂
😁😁😁😁😁😁 0:10 0:11 0:11 👀 0:13 0:13 0:13 👀👏🏿👏🏿👏🏿👏🏿👏🏿👏🏿Educational
Is this notebook shared somewhere?
🎉🎉🎉🎉🎉🎉🎉🎉😊
😊😊😊😊😊🎉🎉🎉🎉
Back on it again
Keeping up with the LLMs!
Do you laugh when you go in small tangents?
lets gooo 🎉🎉🎉
Haha, indeed! Thanks for watching!
Hey Connor, this is great content. Thanks for posting it.
Thanks so much DJ!
zooming in and out is distracting
I tried the implementation but i keep getting the error "model not found"
great intro, and any GitHub repository for it verification?if so ,will be greatly appreciated.
Super interesting. But boy you move quickly through all this. It's really hard to follow at times.
Thanks
What metrics should i keep an eye on to know what to change to get better and better results? I'm trying to create the fake images to use for data augmentation, and obviously I want them as realistic as possible, but I honestly don't know which parameters to change for it to get better. Plus, i have 10 classes and I dont know if i should just change the same for all of them or just see what works for each. But, again, i dont even know how to make it better for even one
Two questions: - Why use gpt-4 instead of gpt-4-turbo for the teleprompter? - What are you using to make your pointer act like that?
I always thought it was pronounced as D ES Pie. thanks for the deep dive btw!
great presentation!
Thanks for the great content. One of the things I am missing is how to save the optimized program so I can use it after that without constantly re-training.
Thank you.
i'm getting a headache by the zooming-in and then skipping across the page.
Connor be experimenting with video formats.
This is almost exactly the same video as the one by Qdrant. Weird.
Guess which one came out first... 🫣 It's super weird indeed
I ran your notebook and got the following error. print(RAG()("What is binary quantization?").answer) AttributeError Traceback (most recent call last) Cell In[7], line 1 ----> 1 print(RAG()("What is binary quantization?").answer) File ~/code/vector_search/weaviate/recipes/.wenv/lib/python3.11/site-packages/dspy/primitives/program.py:26, in Module.__call__(self, *args, **kwargs) 25 def __call__(self, *args, **kwargs): ---> 26 return self.forward(*args, **kwargs) Cell In[6], line 16 15 def forward(self, question): ---> 16 context = self.retrieve(question).passages 17 pred = self.generate_answer(context=context, question=question).answer 18 return dspy.Prediction(context=context, answer=pred, question=question) File ~/code/vector_search/weaviate/recipes/.wenv/lib/python3.11/site-packages/dspy/retrieve/retrieve.py:30, in Retrieve.__call__(self, *args, **kwargs) 29 def __call__(self, *args, **kwargs): ---> 30 return self.forward(*args, **kwargs) File ~/code/vector_search/weaviate/recipes/.wenv/lib/python3.11/site-packages/dspy/retrieve/retrieve.py:39, in Retrieve.forward(self, query_or_queries, k) 36 # print(queries) 37 # TODO: Consider removing any quote-like markers that surround the query too. 38 k = k if k is not None else self.k ---> 39 passages = dsp.retrieveEnsemble(queries, k=k) 40 return Prediction(passages=passages) ... 79 .do() 81 results = results["data"]["Get"][self._weaviate_collection_name] 82 parsed_results = [result[self._weaviate_collection_text_key] for result in results] AttributeError: 'WeaviateClient' object has no attribute 'query'
Hey Peter! Apologies we have upgraded the WeaviateRM to use the Weaviate v4 client, can you please try upgrading dspy with `!pip install dspy-ai --upgrade` ?
Can you please share any error messages as an Issue on Weaviate recipes? It might be easier to help debug there instead of KZread comments.
@connorshorten6311 Please do update the video with accurate setup instructions. I have been fighting to get this running (DSPY + Weaviate + OLLAMA) for the past 2-3 hours to no avail. Tried multiple weaviate-client/server combinations, ran trough docker and standalone, configured, updated/downgraded dspy-ai. Went through so many help pages, cannot count now. I am tired, but still would like to play with this set of technologies. Thanks
Hey, what version of Weaviate-client you are using????
Hey! I am using v4 and the latest version of dspy-ai, can you please share any error messages as an Issue on Weaviate recipes? It might be easier to help debug there instead of KZread comments.
Allright, will check the issues!
How can we get metadata that is associated with any chunk of docs
Is it me, or at least the last part is a digital avatar?
Great video, Connor. Have you tested out if SAMMO is better than DSPy for production?
I thought SAMMO was primarily prompt templating? Does it have some new features we are not aware of?
@@larsbell1569 No, you're right. I was working on the assumption that in production, you'd be using the most capable language models. At the risk of running token-heavy pre-compiled DSPy prompts, having a simpler prompt that automatically augments few-shot prompting only on specific user inputs/triggered events might be a more cost-efficient solution.
This is so different from RAG using GPT. Lots to learn
maybe just me but the blur/smooth filter to the face cam makes me suspect that face cam is AI generated 🤣🤣🤣🤣🤣🤣
Uncanny valley feel.
Is llama really OSS if we don’t know how or what it is trained on?
how to install ollama
this mf cookin
Would love to see an interface to groq please!
Dear Connor, that was the fastest release ever!
man, that's an amazing overview, thank you
Means a lot! Thank you so much!
Thank you for great content, but the zoom in/zoom out effects on video are really annoying! Why not using simple highlighting or just leave the slide as simple as they are?
I love your energy throughout this video Connor!
If I can give some feedback, too much discussion of regular RAG not enough of getting to the point of the benefits of this model. 128k context windows is what everyone has, if not better....I gave up waiting for the point.
Hey conor can we connect to ollama local cohere commad r model to implement function calling or tool use using dspy
Constant zoom-in and zoom-out is quite too much - so distracting. Didn’t finish the video 😘
could we cover the creation of the schema from an empty database such that the notebook flow actually runs through
Is there video about optimization with gradient descent?