Zephyr 7b beta: paper deep dive, code, & RAG

  Рет қаралды 9,201

Sophia Yang

Sophia Yang

Күн бұрын

Is Zephyr 7b beta the best fine-tuned smaller open-source LLM? It is fine-tuned on the best small open-source foundation model Mistral 7B, outperforms all small models, and even outperforms LLama2-Chat-70B and ChatGPT on multiple benchmarks.
What are the difference between Zephyr 7b alpha and Zephyr 7b beta?
From the author: "The main differences are (a) better dataset filtering of UltraChat, namely fixing grammatical errors and some unwanted responses and (b) training for longer, ie 3 DPO epochs vs 1"
Model: huggingface.co...
Technical report: arxiv.org/abs/...
Colab: colab.research...
LlamaIndex Tweet (colab link in the tweet): / 1718054631413363196
🔔 SUBSCRIBE to my channel: www.youtube.co...
⭐ Stay in touch ⭐
📚 DS/ML Book Club: dsbookclub.gith...
▶ KZbin: / sophiayangds
✍️ Medium: / sophiamyang
🐦 Twitter: / sophiamyang
🤝 Linkedin: / sophiamyang
💚 #ai

Пікірлер: 29
10 ай бұрын
Thanks a lot. Using Zephyr with ollama and it's pretty cool and fast.
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
Awesome! Ollama is really nice.
@jstevh
@jstevh 10 ай бұрын
Thanks for that analysis. Like a good overview of a solid paper.
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
Thank you!
@prestonmccauley43
@prestonmccauley43 10 ай бұрын
Thanks for the information I look forward to giving this model a try though it does seem like we keep producing models every week. They really need to be run for many months to determine how well they work.
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
Thanks! Yeah too many models coming out. One of the takeaways is Mistral 7B is a great foundation model to fine-tune on.
@Hash_Boy
@Hash_Boy 10 ай бұрын
many many thanks
@taeyangoh7305
@taeyangoh7305 10 ай бұрын
great review, Sophia! I want to give it a try for care robot use case!!
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
Thanks! Sounds like a really cool use case. Good luck!
@foreignconta
@foreignconta 10 ай бұрын
I am using this model as my default model currently. It is not perfect but I prompted to according to my liking. I wish the prompt template was a bit simpler. 😅
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
Nice!
@DistortedV12
@DistortedV12 10 ай бұрын
It's okay..I tried it with different prompts. Can't wait till they scale it.
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
It's a good 7B model.
@talhayousuf4599
@talhayousuf4599 10 ай бұрын
Your channel is really informative, good thing is that you remain updated with new things coming. Are you a researcher?, do you have any blog? / related to llm research?
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
Thanks 🙏 I have a casual blog with book reading notes and other random things: sophiamyang.medium.com/
@TheReferrer72
@TheReferrer72 10 ай бұрын
Is this the Model that does not reason or code well? How can this model be useful?
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
Still better than llama2 70B on coding and math
@jsalsman
@jsalsman 10 ай бұрын
Do they mention the cost of the GPT-4 calls during the training steps?
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
They did not mention cost in the paper.
@jsalsman
@jsalsman 10 ай бұрын
@@SophiaYangDS Any idea roughly how many API calls they must have used?
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
Must be a lot. Just to process and create this UltraFeedback dataset (huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized?row=0), it's around 74k x 5 API calls (4 models responses + 1 GPT4 ratings). @@jsalsman
@mshonle
@mshonle 10 ай бұрын
Does distilling help to remove overfitting?
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
I'm not sure. You can overfit in distillation.
@mshonle
@mshonle 10 ай бұрын
I wonder how much this cost them to make, in terms of compute? (Or, maybe more relevant: how much would it cost someone else to do the same?)
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
​@@mshonle I think they might be spending most of the compute/API calls in generating the datasets. The actual SFT and DPO might not be that crazy.
@mshonle
@mshonle 10 ай бұрын
Hmm, and those datasets are available?
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
@@mshonle Yes they are public on Hugging Face
@serkhetreo2489
@serkhetreo2489 10 ай бұрын
Sometimes, u discover new channels
@SophiaYangDS
@SophiaYangDS 10 ай бұрын
Thanks for discovering my channel :)
Advanced RAG 01: Small-to-Big Retrieval with LlamaIndex
9:17
Sophia Yang
Рет қаралды 12 М.
Building Production-Ready RAG Applications: Jerry Liu
18:35
AI Engineer
Рет қаралды 312 М.
Apple peeling hack @scottsreality
00:37
_vector_
Рет қаралды 127 МЛН
ПРИКОЛЫ НАД БРАТОМ #shorts
00:23
Паша Осадчий
Рет қаралды 5 МЛН
GraphRAG: The Marriage of Knowledge Graphs and RAG: Emil Eifrem
19:15
Has Generative AI Already Peaked? - Computerphile
12:48
Computerphile
Рет қаралды 988 М.
AI Pioneer Shows The Power of AI AGENTS - "The Future Is Agentic"
23:47
Efficient Fine-Tuning for Llama-v2-7b on a Single GPU
59:53
DeepLearningAI
Рет қаралды 84 М.
RAG But Better: Rerankers with Cohere AI
23:43
James Briggs
Рет қаралды 59 М.
What is RAG? (Retrieval Augmented Generation)
11:37
Don Woodlock
Рет қаралды 142 М.
High-performance RAG with LlamaIndex
59:37
AI Makerspace
Рет қаралды 18 М.
This is why Deep Learning is really weird.
2:06:38
Machine Learning Street Talk
Рет қаралды 387 М.
A Hackers' Guide to Language Models
1:31:13
Jeremy Howard
Рет қаралды 525 М.