Refuel (@RefuelAI) profile picture

Solve enterprise data tasks at superhuman accuracy. Acquired by @togethercompute

San Francisco, CA
Joined May 2021
Photo by Refuel (@RefuelAI)
🚀 Big news: Together AI has acquired @RefuelAI! Refuel specializes in models and tools that turn messy, unstructured data into clean, structured input—exactly what teams need to build high-quality, production-grade AI applications. Details below 👇
2
4
1
28
2,780
Data intelligence too cheap to meter RefuelLLM-2-mini (75.02%), our latest 1.5B param SLM, outperforms all comparable models including Phi-3.5 (65.3%), Qwen2.5 (67.62%), Gemma2 (64.52%), Llama3-3B (55.8%) and Llama3-1B (39.92%) across our benchmark of data processing tasks such as labeling, enrichment and structure extraction RefuelLLM-2-mini is a Qwen2-1.5B base model, trained on a corpus of 2750+ datasets spanning tasks such as classification, reading comprehension, structured attribute extraction and entity resolution, using the same recipe as other models in the Refuel-LLM family. It's fast! We’re open sourcing the model weights, available on @huggingface - huggingface.co/refuelai/Qwen… If you'd like to access models, along with fine tuning support, DM me or reach out to us: refuel.ai/get-started Grateful to our early customers for their partnership, and the entire @RefuelAI team for their hard work 🚀
Thrilled to introduce RefuelLLM-2, our latest family of LLMs built for data labeling and enrichment tasks. RefuelLLM-2 (83.82%) outperforms GPT-4-Turbo (80.88%), Claude-3-Opus (79.19%), Llama3-70B (78.2%) and Gemini-1.5-Pro (74.59%) on a benchmark of ~30 data labeling tasks: RefuelLLM-2-small (79.67%), aka Llama-3-Refueled, outperforms all comparable LLMs including Claude3-Sonnet (70.99%), Haiku (69.23%) and GPT-3.5-Turbo (68.13%). We’re open sourcing the model: huggingface.co/refuelai/Llam… You can try out the models here and give us some feedback! labs.refuel.ai/playground. The code and data used for benchmarking the LLMs is available in our Autolabel library: github.com/refuel-ai/autolab… One more thing: RefuelLLM-2 family of models output much better calibrated confidence scores - a useful lever to reject, retry or ensemble low confidence outputs.
3
3
14
2,299
Thank you @databricks @DbrxMosaicAI for the keynote shoutout! Always great connecting with new and old friends at the @Data_AI_Summit
2
19
1,349
We're kicking off the Data+AI Summit with the #MosaicX #Meetup: San Francisco Edition on Monday, June 10th. We're at the #Moscone Center South, 2nd floor, with over 1500 registrants and 39 speakers across four tracks. It's a "slightly" packed agenda with: ✅ Discussion panels on #Hardware, Build & Risks, Data Panel, and #VC Panel with a special session on #OLMo ✅ #Research track on importance of high quality #data, common challenges in #RAG development, #diffusion models, and more ✅ A use cases track on composable #CDP, building models, #multimodal, #agents, and more ✅ In the building LLMs track, we discuss the challenges, tools/techniques to build them, fast #LLM inference, and building #GenAI apps. While we are fully packed, if you are already registered for #DataAISummit, we will have a waitlist at the door. We have speakers from @databricks @DbrxMosaicAI @LaminiAI @Oracle @VoltronData @Replit @AiSquared_ @robusthq @gretel_ai @superannotate @EssenceVenture @AmplifyPartners @llama_index @QuotientAI @ActionIQinc @RefuelAI @OrbyAI @yousearchengine @NumbersStnAI @lancedb @huggingface mosaicx.events/events/june-1…
6
14
6
52
37,302
We're trending on @huggingface! Check out: huggingface.co/refuelai/Llam… Try out the model here: labs.refuel.ai/playground
Thrilled to introduce RefuelLLM-2, our latest family of LLMs built for data labeling and enrichment tasks. RefuelLLM-2 (83.82%) outperforms GPT-4-Turbo (80.88%), Claude-3-Opus (79.19%), Llama3-70B (78.2%) and Gemini-1.5-Pro (74.59%) on a benchmark of ~30 data labeling tasks: RefuelLLM-2-small (79.67%), aka Llama-3-Refueled, outperforms all comparable LLMs including Claude3-Sonnet (70.99%), Haiku (69.23%) and GPT-3.5-Turbo (68.13%). We’re open sourcing the model: huggingface.co/refuelai/Llam… You can try out the models here and give us some feedback! labs.refuel.ai/playground. The code and data used for benchmarking the LLMs is available in our Autolabel library: github.com/refuel-ai/autolab… One more thing: RefuelLLM-2 family of models output much better calibrated confidence scores - a useful lever to reject, retry or ensemble low confidence outputs.
1
5
2
19
14,301
Thrilled to introduce RefuelLLM-2, our latest family of LLMs built for data labeling and enrichment tasks. RefuelLLM-2 (83.82%) outperforms GPT-4-Turbo (80.88%), Claude-3-Opus (79.19%), Llama3-70B (78.2%) and Gemini-1.5-Pro (74.59%) on a benchmark of ~30 data labeling tasks: RefuelLLM-2-small (79.67%), aka Llama-3-Refueled, outperforms all comparable LLMs including Claude3-Sonnet (70.99%), Haiku (69.23%) and GPT-3.5-Turbo (68.13%). We’re open sourcing the model: huggingface.co/refuelai/Llam… You can try out the models here and give us some feedback! labs.refuel.ai/playground. The code and data used for benchmarking the LLMs is available in our Autolabel library: github.com/refuel-ai/autolab… One more thing: RefuelLLM-2 family of models output much better calibrated confidence scores - a useful lever to reject, retry or ensemble low confidence outputs.
15
63
20
307
100,990
Refuel retweeted
Labeling LLM data with confidence is hard. This post by @nihit_desai and @BansalDhruva from @RefuelAI was surprisingly valuable for how we're thinking about these things in @cardinalappio 🏆 refuel.ai/blog-posts/labelin…
2
3
10
736
Refuel is super cool. They’re able to address the issue of data labeling at scale, it’s truly impressive. They turn the process into ‘use Refuel, focus on the data science.’ @RefuelAI #MachineLearning #LLMs
Hi @mmarshall - at @RefuelAI, we have a fine-tuned version of Llama-2 deployed in production -- it's used by our customers for data cleaning and labeling use cases. We've served millions of data points in just the last month - happy to share more if interested. Many thanks to the great work by @ylecun and team at Meta, and our infra partners at @MosaicML refuel.ai/blog-posts/announc…
1
2
593
searchenginejournal.com/goog… At @RefuelAI we've made the bet that LLMs can annotate and enrich data better than human annotators (refuel.ai/blog-posts/llm-lab…), a big unlock for creating datasets to finetune and evaluate AI applications. My gut sense is, Google realizes the same.
2
8
1,103
Replying to @antonosika
this is buildable right now, no? @hwchase17 has a notebook of a research agent, @sama is exposing logprobs for 3.5 & 4 via API, and @TruLensML or @RefuelAI can eval relative truthiness if given some (probably pullable on the fly?) ground truth baseline
1
1
2
285
2/ @RefuelAI for annotating LLM data at scale refuel.ai/ LLM purpose-built for data labeling and enrichment to build AI applications that reduce hallucinations and hit 99% accuracy.
1
1
1
327
(Part 3: AI): AI Infrastructure ⚒️ @UnstructuredIO for pre-processing your data @RefuelAI for annotating LLM data at scale @dify_ai for streamlining GenAI development @vellum_ai for producing AI-powered features @baseten for deploying and scaling your models
1
1
6
412
.@Refuelai debuts a large language model to label the data needed to train other LLMs buff.ly/3tGuy1o @thecubetv @Mike_Wheatley “But this field is getting crowded, with many established players and new entrants...” - @AndyThurai @constellationr
1
2
4
606
Elated to see the response!🚀 Appreciate all your feedback thus far. We're working through requests for API access as quickly as we can. * The LLM can be used (for free) via Autolabel. Request access: refuel-ai.typeform.com/llm-a… * To sign up for Refuel Cloud (which supports Refuel LLM already): refuel.ai/get-started
We’re thrilled to announce Refuel LLM, a large language model built for data labeling and enrichment tasks. Outperforms trained human annotators, GPT-3.5-turbo, PaLM-2, Claude and a host of open-source LLMs on a benchmark of data labeling tasks: refuel.ai/blog-posts/announc… Further, Refuel LLM is easy to finetune. For each of the datasets used for finetuning experiments, we were able to get to a model that outperformed GPT-4 with less than 15 minutes of training on a cluster of 8x H100s. On average, finetuning improved label quality by 12.2% across datasets. Under the hood, it is a Llama-v2-13b base model, instruction tuned on over 5.24B tokens, spanning tasks such as classification, entity resolution, matching, reading comprehension and extraction. You can play around with the model in the playground here: app.refuel.ai/playground, or access it via Autolabel. We plan to share a more detailed technical summary in the coming weeks!
3
11
1,399
Excited to discuss Autolabel with Rishabh Bhargava Autolabel: Open-source platform for text data labeling, cleaning, and enhancing with LLMs🚀 👉 youtube.com/watch?v=TjzeaHjm…
4
1
17
4,012
It has begun! 🚀👀
1
3
10
1,349
1/ Confidence estimation is an effective tool to mitigate hallucinations when leveraging LLMs for data labeling and enrichment. Excited to share our learnings on this topic here: refuel.ai/blog-posts/labelin…
1
4
1
14
1,605
Here's my latest great find: a #Python library called Autolabel by @RefuelAI. It allows for automated labeling, cleaning, and enriching of textual data using language models. Autolabel supports a wide range of NLP tasks, saving you time and costs compared to manual labeling. By leveraging open-source models from #OpenAI, Anthropic, HuggingFace, and Google, this library enables high-precision data labeling. What are some gems you’ve found on #GitHub?
3
11
765