llm-awq
Voyager
llm-awq | Voyager | |
---|---|---|
7 | 53 | |
1,902 | 5,199 | |
10.9% | 2.3% | |
8.0 | 4.7 | |
8 days ago | about 2 months ago | |
Python | JavaScript | |
MIT License | MIT License |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
llm-awq
-
TinyChat: Large Language Model on the Edge
TinyChat is an efficient, lightweight, Python-native serving framework for 4-bit LLMs by AWQ. It delivers 2.3x generation speed up on RTX4090.
Code: https://github.com/mit-han-lab/llm-awq/tree/main/tinychat
- FLaNK Stack Weekly 23 Oct 2023
-
New base model InternLM 7B weights released, with 8k context window.
I am having trouble finding any 8bit GPTQ models at all, there don't seem to be any on HF it's almost all 4bit with the odd 3bit of the big ones. Suspect I will have to make my own for eval purposes but it's lower priority on my list then finding a 4bit that's GPU friendly but doesn't have such a performance penalty... Looking at AWQ they have 3 and 4bit versions.
-
Llama33B vs Falcon40B vs MPT30B
Using the currently popular gptq the 3bit quantization hurts performance much more than 4bit, but there's also awq (https://github.com/mit-han-lab/llm-awq) and squishllm (https://github.com/SqueezeAILab/SqueezeLLM) which are able to manage 3bit without as much performance drop - I hope to see them used more commonly.
- New hardware-friendly quantization method
-
Activation-Aware Weight Quantization for LLM Compression Outperforms GPTQ
Better quantization would have a direct and meaningful impact for everyone running local LLMs. The technique has already been applied to both Vicuna and the multimodal LLaMA variant LLaVA.
https://github.com/mit-han-lab/llm-awq
-
New quantization method AWQ outperforms GPTQ in 4-bit and 3-bit with 1.45x speedup and works with multimodal LLMs
GitHub: https://github.com/mit-han-lab/llm-awq
Voyager
-
Google Launches Gemini, Its "Most Powerful" AI Model to Date
Source: Conversation with Bing, 12/10/2023 (1) Wes Roth - YouTube. https://www.youtube.com/@WesRoth. (2) I've set most of my videos to Public again - Community. https://community.openai.com/t/ive-set-most-of-my-videos-to-public-again/24535. (3) AI Updates: Meta Develops Mind-Reading AI System, OpenAI’s Q* Is Here .... https://www.windermeresun.com/2023/11/20/ai-updates-meta-develops-mind-reading-ai-system-openais-q-is-here-how-economy-will-work-after-agi/. (4) David Shapiro. https://www.daveshap.io/. (5) undefined. https://natural20.com/. (6) undefined. https://arxiv.org/abs/2305.16291. (7) undefined. https://twitter.com/DrJimFan/status/1. (8) undefined. https://voyager.minedojo.org/. (9) undefined. https://minedojo.org/. (10) undefined. https://www.youtube.com/@DavidShapiroAutomator/videos.
- Is there any game that allow us to interact with it by python?
-
A Coder Considers the Waning Days of the Craft
> AI cannot sustain itself trained on AI work.
This isn’t true. You can train LLMs entirely on synthetic data and get strong results. [0]
> If new languages, engines etc pop up it cannot synthesize new forms of coding without that code having existed in the first place.
You can describe the semantics to a LLM, have it generate code, tell it what went wrong (i.e. with compiler feedback), and then train on that. For an example of this workflow in a different context, see [1].
> And most importantly, it cannot fundamentally rationalize about what code does or how it functions.
Most competent LLMs can trivially describe what some code does and speculate on the reasoning behind it.
I don’t disagree that they’re flawed and imperfect, but I also do not think this is an unassailable state of affairs. They’re only going to get better from here.
[0]: https://arxiv.org/abs/2309.05463
[1]: https://voyager.minedojo.org/
-
AutoGen: Enable Next-Gen Large Language Model Applications
In a way it is the same thing, agents are mostly an abstraction that make it easier to know what’s going on.
I think of agents more or less as python classes with a mixture of natural language and code functions. You design them to do something with information they produce, and to interface with other agents or “tools” in some way.
But all the agents can be the same language model under the hood, they are frames used to build different kinds of contexts.
And yes I think the idea is that emergent behaviour can be useful. This comes to mind
https://github.com/MineDojo/Voyager
But I think we are still a small ways off from being really smart about agents. My opinion is that we haven’t quite figured out what we are doing yet.
-
Open/Local LLM support for MineDojo/Voyager
This k8s application deploys an instance of Voyager along with a Fabric Minecraft server with required fabric mods. It assumes you have a local deployment of a Large Language Model (LLM) with 4K-8K token context length with a compatible OpenAI API, including embeddings support.
- Voyager – Minecraft Embodied Agent with Large Language Models
-
List of Awesome AI Agents like AutoGPT and BabyAGI / Many open-source Agents with code included!
In my opinion the most interesting Agents: Auto-GPT Github: https://github.com/Significant-Gravitas/Auto-GPT BabyAGI Github: https://github.com/yoheinakajima/babyagi Voyager Github: https://github.com/MineDojo/Voyager / Paper: https://arxiv.org/abs/2305.16291 I would also add: ChemCrow: Augmenting large-language models with chemistry tools Github: https://github.com/ur-whitelab/chemcrow-public/ Paper: https://arxiv.org/abs/2304.05376
-
[D] - Are there any AI benchmarks that involve successful longterm problem solving when running as autonomous agents (like in autogpt)? How do we compare the effectiveness of models as agents?
Does this beat the voyager? I read about it and wondered what if we add a skill library to langchain/llamaindex agents. It could be the same vector store for storing static data but after each task is performed, the agent will evaluate and archive the recipe of steps to perform a new task. Next time when the agent is asked to perform a task, it can just look at the library to retrieve a recipe. Unlike traditional fine tuning, you dont update the model parameters, these recipes are much more interpretable and can be manually edited/inserted by humans. There may also be an automatic way to convert wikihow articles or youtube tutorials into recipes.
-
GPT-4 was set free in Minecraft, here's what happened next...
Source. P.S. If you love geeking over AI updates, I have this free newsletter you might want to check out. Thank you!
Source.
What are some alternatives?
SqueezeLLM - [ICML 2024] SqueezeLLM: Dense-and-Sparse Quantization
GITM - Ghost in the Minecraft: Generally Capable Agents for Open-World Environments via Large Language Models with Text-based Knowledge and Memory
GPTQ-for-LLaMa - 4 bits quantization of LLaMA using GPTQ
tree-of-thought-llm - [NeurIPS 2023] Tree of Thoughts: Deliberate Problem Solving with Large Language Models
langchain4j-examples
mineflayer - Create Minecraft bots with a powerful, stable, and high level JavaScript API.
CML_AMP_AI_Text_Summarization_with_Amazon_Bedrock - CML_AMP_AI_Text_Summarization_with_Amazon_Bedrock
gorilla - Gorilla: An API store for LLMs
kafka-streams-dashboards - showcases Grafana dashboards for Kafka Stream applications leveraging client JMX metrics.
data-in-motion - This is repository for tutorials of Data In Motion starting with Data Distribution
qlora - QLoRA: Efficient Finetuning of Quantized LLMs