🚀 Launching Arcee Agent: The Cutting-Edge Specialized 7B Language Model
Trained with Spectrum, powered by
@CrusoeAI
Outperforms larger models like GPT-3.5-Turbo
Excels in interpreting, executing, and chaining function calls.
Find it here:
We have huge news today for Model Mergers...
We're now excited to announce the launch of the state-of-the-art functionality of Evolutionary Model Merging in MergeKit!
#nlp
#ai
#LLM
#LLMs
Arcee is dropping two new datasets today:
1. The Tome: A 1.75M sample dataset filtered for training strong generalist models, used for Spark & Nova.
2. Agent Data: Key for Arcee-Agent, includes Salesforce-xlam, agent-flan, Glaive-FC2 (20k samples), and Magpie-Pro.
Links below.
In collab /w
@huggingface
, Arcee is thrilled to release our MergeKit Hugging Face Space.
🙌 You now can perform model merges w/ MergeKit in an easy-to use-UI, & save models right into your Hugging Face hub.
Try it out here:
(1/2)
#nlp
#llm
#llms
Arcee AI is excited to launch 💡Llama-3-SEC💡
Built on Meta-Llama-3-70B-Instruct w/ goal of providing unparalleled insights & analysis capabilities for finance pros, investors, researchers, & anyone working w SEC filings & related data.
#nlp
#LLMs
#ai
New Model Release 🔥
Introducing Hermes 2 Theta 70B! This model is smarter, more creative, and more capable than ever before. It outperforms Llama-3 Instruct 70B on various benchmarks. A proud collaboration with
@Teknium1
,
@theemozilla
,
@karan4d
, and
@NousResearch
🆕ARCEE AI MODEL ALERT🆕
We’ve just dropped Arcee-Nova:
🤗Evaluated on the OpenLLM Leaderboard 2.0 stack
🏆Top-performing OS model on this stack
📈Approaches GPT-4 (May 2023) performance levels, marking a significant milestone.
Details here:
#LLMs
🆕ARCEE AI MODEL ALERT🆕
While working on Spark v2, we discovered a training checkpoint of InternLM-2.5 that we adored. Though it doesn't fully meet our expectations for a Spark model, its language skills are exceptional, particularly in creative writing...
🧵 (1/4)
🌟 Nothing better than getting stellar reviews of our model releases!
🌑 𝗔𝗿𝗰𝗲𝗲-𝗡𝗼𝘃𝗮 is our highest-performing
#opensource
model...
🧠 We created it by merging Qwen2-72B-Instruct w/ a custom model tuned on a generalist dataset mixture...
(1/4)
#LLM
#GenAI
#NLP
Haven’t had a chance to read up on the Spectrum method for efficiently training
#LLMs
by targeting specific layer modules?
It optimizes resource usage, cutting training time & costs IN HALF.
Check out the paper here () or learn the basics in our video ⬇️
Thanks to
@VentureBeat
for covering this huge milestone for Arcee AI…
✨Less than a year after emerging from stealth, we’ve signed a $24 million Series A led by
@emergencecap
✨
💗Proud to celebrate w/ our partners, our team, & our customers 💗
Link to article in replies ⬇️
🕵Wondering what you can do with our new 7B
#LLM
, Arcee Agent?
It may be just a 7B but it's an invaluable asset for automating multi-step business tasks and processes, across industries.
Check out the demo in this video by our
@LucasAtkins7
⬇️
(1/2)
#nlp
#GenAI
#LLMs
Introducing Weight Averaged Rewarded Policies (WARP), Google DeepMind's latest RLHF alignment method using the magic of model merging. By scaling alignment like pre-training was scaled, WARP learns sota Gemma LLM surpassing previous releases. A 🧵below.
ICYMI, we've had 👀 a few 👀 developments this week.
TL;DR:
🌟 We announced our $24 million Series A, led by
@emergencecap
🖊️ We signed
@julsimon
as our Chief Evangelist
⛅ Our hosted SaaS, Arcee Cloud, went LIVE.
Recap here ⬇
Wishing a huge welcome to
@julsimon
who has joined us as Chief Evangelist!
From AWS to Hugging Face and now to
@arcee_ai
Julien has led the way on
#AI
development... & now shares our conviction that
#SLM
-led
#GenAI
is the future👏👏
Honored to have you on this journey!!! 🚀🚀
Ready to learn more about Model Merging? For Day 15 we bring you the paper we just published: "Arcee's MergeKit: A Toolkit for Merging Large Language Models"
Hit us up here with your questions on all things MergeKit!
#nlp
#ai
#llm
#llms
Coming up: our 🎉Bonus Edition 🎉 of the Small Language Model (SLM) show, w/ the winner of our MODEL MERGING Hackathon
@maximelabonne
talking about his winning merge.
Hosts
@Malikeh5
& @
@mmcquade_ai_u
will also chat RE our EPIC model release Llama-3-SEC
Haven't yet had a chance to check out our epic model release, Llama-3-SEC? Our
@LucasAtkins7
walks you through his queries about this week's hottest stock (NVIDIA of course) - check it out here:
#nlp
#GenAI
#LLMs
#LLM
We're thrilled to welcome
@maximelabonne
to the first episode of The Small Language Model (SLM) Show... He'll talk about his extensive experimenting w/ MODEL MERGING, w/ MergeKit Founder
@chargoddard
#nlp
#llm
#ai
Here at Arcee AI, every day we’re making the world more aware of the power of Small Language Models (SLMs), & today we’re dropping the 🦾highest-scoring model ever 🦾in the 7B-15B range: 🎇Arcee Spark🎇, which outperforms Mixtral-8x7B & Llama-3-8B-Instruct!
🧵 (1/3)
Another day, another amazing stat RE our Arcee Spark model.
W/ an EQ-Bench score of 71.4, Arcee Spark outranks:
• Phi-3-medium from Microsoft (double its size),
• Claude Sonnet 3,
& is almost even w/ miquella120b 👏
Find it here:
#LLMs
#GenAI
#NLP
🤔 What should you be doing with
#llama31
, the latest and greatest open source
#LLM
?
🧠 You should be MERGING IT.
👩🏫 You should be TRAINING IT.
💁♂️ You should be adapting it to your domain & use case.
Our Jacob Solawetz wrote you a guide:
#NLP
#GenAI
📣ICYMI – Check out this terrific chat about the power of Model Merging, featuring Arcee's
@chargoddard
(the founder of MergeKit) & hosted by the talented folks over at
@AIMakerspace
:
#nlp
#genai
#LLMs
👏 Congrats to Arcee-Nova for its high ranking on the Open
#Arabic
#LLM
Leaderboard on
@huggingface
.
🏆 It's 𝗙𝗢𝗨𝗥𝗧𝗛 𝗼𝘃𝗲𝗿𝗮𝗹𝗹...
🥇 And 𝗙𝗜𝗥𝗦𝗧 𝗶𝗻 𝘁𝗵𝗿𝗲𝗲 𝗯𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸𝘀: ACVA, MMLU, and EXAMS.
#GenAI
#NLP
Congrats 2 Arcee AI's
@LucasAtkins7
&
@FernandoNetoAI
as co-authors on this paper introducing a method to accelerate
#LLM
training by selectively targeting layer modules based on their signal-to-noise ratio, freezing the rest. Drop questions for them here!
Model Merging onstage at MIT today for the Imagination in Action: Forging the Future of Business with AI Summit. Arcee's
@chargoddard
talking all things MergeKit! Video recording of the talk coming soon here🚀
#ai
#nlp
#LLM
#LLMs
Small Language Models (SML) are the future of AI.
"Small" (SML) instead of "Large" (LLM). These small models are highly specialized models with superhuman abilities on specific tasks.
Here are two techniques to build these models:
• Spectrum
• Model Merging
I give you a
... in MergeKit, though the current version of transformers messes with the InternLM tokenizer when using save.pretrained. It took quite a while to get a working model. Proceed with caution!
(4/4)
#NLP
#GenAI
Starting to really understand the shortcomings of LLMs?It's time to check out SLMs–our Small Language Model system that's the answer for enterprises looking for bespoke
#ai
trained on THEIR data. Big results at a great price 🦾
#nlp
#ai
#LLM
#LLMs
It's been just a couple of weeks since we launched the MergeKit GUI (), & we're already closing in on 7k merges!
BTW don't forget to submit your merges for our Model Merging Hackathon-you have until May 6! Details here:
#nlp
#LLM
Ready to learn more about the technique called Spectrum for optimizing
#LLM
training?
Spectrum:
⏲ Reduces training time
📝 Improves memory efficiency
🧠 Minimizes catastrophic forgetting.
More in this article ⬇️ by our
@LucasAtkins7
#nlp
#GenAI
Greet week of collab between
@arcee_ai
&
@huggingface
: along w/ the release of 2 Arcee / MergeKit Hugging Face Spaces (one of which is a Featured Space: ), our founders & other team members got to hang with Hugging Face CEO
@ClementDelangue
🤗 in Miami 🌴
Here at Arcee AI, that's exactly what our end-to-end platform for training and deploying CUSTOM LANGUAGE MODELS gives our customers.
We're dropping the ☁️ version, Arcee Cloud, TOMORROW!
Check out the DEMO-linked in the comments!
(2/2)
#NLP
#AI
#LLMs
BIG news - Arcee AI’s Small Language Model show is now available on your favorite podcast platform.
Subscribe on Apple Podcasts here:
And on Spotify here.
Check out
@LucasAtkins7
's look at the model we dropped today: the 7B Arcee Spark:
🌟 it's the highest-scoring model by far in the 7B-15B range
🌟 it gets jokes, and can explain jokes better than some of us 😆
#nlp
#LLM
#GenAi
Weekend reminder to MODEL MERGERS - get 💸 for your work by entering the
@NeurIPSConf
model merging competition - details below.
@arcee_ai
is proud to be a co-sponsor as we help bring Model Merging to the world 🦾🌎
🚨 Model Merging competition
@NeurIPSConf
!🚀
Can you revolutionize model selection and merging?Let's create the best LLMs!🧠✨
💻Come for science
💰Stay for $8K
💬Discord:
🔗Sign up:
Sponsors:
@huggingface
@SakanaAILabs
@arcee_ai
With the launch of our Small Language Model (SLM) training & merging platform's in-browser version, Arcee Cloud, we're seeing a spike in people getting started w/ Model Merging. Here are 3 great resources to guide you:
1) Our docs, of course: ...
(1/2)
Introducing Hermes 2 Theta 70B!
Hermes 2 Theta is smarter, more creative, and capable of more then ever before.
It takes a strong lead over Llama-3 Instruct 70B across a wide variety of benchmarks, and is a continuation of our collaboration with
@chargoddard
and
@arcee_ai
.
🚨 Model Merging competition
@NeurIPSConf
!🚀
Can you revolutionize model selection and merging?Let's create the best LLMs!🧠✨
💻Come for science
💰Stay for $8K
💬Discord:
🔗Sign up:
Sponsors:
@huggingface
@SakanaAILabs
@arcee_ai
Another great Model Merging tip, from MergeKit founder
@chargoddard
.
QUESTION: When to use merging over a Mixture of Experts (MoE) model?
ANSWER: I encourage regular merging over frankenMoEs in pretty much all circumstances. They're a really cool hack and there are ...
(1/3)
👏Love seeing these download numbers w/ our Arcee Spark 7B model, released just a week ago.👏
It's the highest-scoring model by far in the 7B-15B range, outperforming Mixtral-8x7B and Llama-3-8B-Instruct.
Find it here:
We're excited to announce that our solution for training, merging, and deploying domain-specific language models is coming soon in a Hosted SaaS called Arcee Cloud.
Get on the waitlist here , & read more here:
#ai
#NLP
#LLMs
Today's episode is seriously mind-expanding. In it, Mark (
@mmcquade_ai_u
) and Charles (
@chargoddard
) detail how they're pushing the A.I. frontier through LLM merging, extremely efficient (even CPU-only!) LLM training, and *Small* Language Models.
Watch here:
Our team is growing fast here at Arcee AI! We're excited to welcome
@shahrzad_sayeh
as our new ML Product Engineer🚀🚀🚀
Shahrzad is a major contributor to building out the Arcee platform and is already a prolific blogger for us - here's her latest piece:
⏲ The countdown is on for our MODEL MERGING HACKATHON, Co-Sponsored by Arcee and AWS.
💰 You have till 5/13 to submit entries in the following categories, for which we're awarding a total of $9k in cash:
🤔 Best New Merge
🤝 Best Integration with Other Ecosystems... (1/2)
Which gets you better results for domain-specific LLMs: QLoRA or standard Continual Pre-Training (CPT)?
Arcee's Head of Applied NLP Research, Shamane Siri, PhD, explains why (& when) standard CPT reigns supreme:
#nlp
#ai
#LLMs
📣 The latest update to Arcee's MergeKit: you can now use MergeKit to extract LoRA adapters from any fine-tuned model.
⬇ Details here from our Research Engineer Thomas Gautier!
#nlp
#ai
#LLM
#LLMs
What companies want RE
#GenAI
...
📈 Find the best model to start from
🖱️ Easily adapt it to their domain in a few clicks
💸 Keep costs under control
✅ Get ROI out-of-the-box
🔁 Then, repeat the process for a different project…
(1/2)
Special shout-out to Julien Chaumond
@julien_c
& Lucain Pouget
@Wauplin
at Hugging Face for their help building this epic merging playground 🚀🚀🚀
And congrats to the Arcee team as well on the release of this fantastic tool 👏👏👏
(2/2)
#nlp
#ai
#llm
#llms
🚀Just 3 days until the launch of our hosted SaaS! As
#GenAI
evolves towards smaller, specialized models, Arcee is leading the charge.🌟 Get an exclusive sneak peek inside Arcee Cloud & witness the power of deploying your model on our cutting-edge platform! Tour link in comments
Today we are releasing an experimental new model in collaboration with
@chargoddard
and
@arcee_ai
, Hermes 2 Θ, our first model merge, combining Hermes 2 Pro, and Llama-3 Instruct, and then further RLHF'ed from there.
Available on HuggingFace:
This model
Why merge 2 models vs. fine-tuning a single model? Our experts answer that for Day 17 of March Merge Madness... starting witht the reduction of the risk of catastrophic forgetting 🧠
#nlp
#ai
#llm
#llms
Are 7B language models the future? Check out the popularity of our recently-released stats for Arcee Agent and Arcee Spark, in the video below⬇️ then you tell us!
#nlp
#GenAI
#LLMs
#SLMs
Want to learn how to take Llama-3.1-8B (
#llama31
), train it on your
#slack
data, and get an
#LLM
with immediate, practical business value? Link to our guide in the comments⬇️
#NLP
#GenAI
As the leader in Small Language Models for enterprises, we couldn't have said it better ourselves: check out this great
@venturebeat
article by
@jathomason
on why the GenAI world is shifting its focus from
#LLMs
to SLMs.
#ai
#nlp
#LLM
Ready to learn more about the model training techniques that helped us land our Series A?
No one better to learn from than
@JonKrohnLearns
!
Fantastic interview on Model Merging & Spectrum✨
#nlp
#GenAI
#LLMs
#SLMs
It's Day 5 of 🏀 March Merge Madness🏀 and we're bringing you one of our biggest takeaways so far...
In our discussions with companies, we’ve realized this:
💡 Even though MODEL MERGING is set to TRANSFORM the world of LLMs as we know it… (1/3)
🤔Wondering what you can do with 𝗔𝗿𝗰𝗲𝗲-𝗡𝗼𝘃𝗮?
Here are just some of the biz use cases:
✅ Customer Service✅ Content Creation✅ Software Dev✅Data Analysis✅R&D✅Legal & Compliance
✅Education & Training...
(3/4)
Great news for Model Mergers: we've extended the deadline for our hackathon, co-sponsored by Arcee/MergeKit &
@awscloud
, to MAY 13!
Full details here:
💵 💵 💵$9k in cash prizes across 3 categories! 💵 💵 💵
Heading to the AI summit @ MIT this week? Be sure to catch the talk on Model Merging & MergeKit by Arcee's
@chargoddard
- this Thursday on the MIT campus
#nlp
#ai
#LLM
#LLMs
Kinda wild that you can merge models with SoTA techniques at the click of a button! 🤯
Presenting MergeKit UI - Drop in your config, access token and voila, you get a merged model back!
Supported merging methods:
1. Model Soups
2. SLERP
3. Task Arithmetic
4. TIES
5. DARE TIES
Big news today for MODEL MERGERS: the chance to win 💰💰💰!
Arcee is hosting a MODEL MERGING HACKATHON with a total of $9k in prizes across 3 categories.
Details in our blog:
#nlp
#ai
#LLM
#LLMs
Announcing the general availability of the
@MongoDB
AI Applications Program (MAAP)!
Here at
@arcee_ai
we're excited to be a part of this incredible ecosystem and are ready to help customers take advantage of
#AI
.
Learn more here:
#MAAPLaunch
How do we deliver high-quality domain-specific language models at a fraction of the cost of other companies?
Check out our white paper:
#nlp
#llm
#llms
#ai
In the countdown to the May 13 deadline for the Model Merging Hackathon co-sponsored by Arcee &
@awscloud
, here's a 🧵 w/ tips for Mergers–from MergeKit founder
@chargoddard
:
💡If you're trying to merge 2 models (and exactly 2), SLERP is always a good first choice...
(1/3)
Terrific seeing the enthusiasm around Arcee's work to democratize LLMs during our CTO/Co-Founder Jacob Solawetz's talk @
#MadHats
AI, co-presented with Ganapathi Krishnamoorthi of AWS Trainium.
Jacob pictured w/ Rohit Talluri & Nick Hartman of AWS 🚀 🚀 🚀
#nlp
#ai
#LLMs
By now most of us know what Model Merging is.
But have you ventured into Evolutionary Model Merging?
That's the topic of our show this week, featuring
@chargoddard
, Arcee CTO
@JacobSolawetz
, AND a new Co-Host,
@Malikeh5
! Live here on X @ 11am PT / 2pm ET this Wednesday...
So much happening here at Arcee that it's hard to keep up.
January:
@TechCrunch
announces our seed round 🚀
February: we merged w/ mergekit 🚀
Can't wait to see what happens in March 🚀