Post
4974
🐋DeepSeek 🐋 is the real OpenAI 😯
- 6 replies
Sinisa Stanivuk
Stopwolf
AI & ML interests
Multilingual LLMs, STT and TTS models
Recent Activity
updated a model about 18 hours ago
Stopwolf/embedic-iter4-sentiment published a model about 18 hours ago
Stopwolf/embedic-iter4-sentiment updated a model 4 months ago
Stopwolf/embedic-m2v-largeOrganizations
reactedto onekq's post with 🔥 about 1 year ago
reactedto nataliaElv's post with 👀 over 1 year ago
Post
1661
Would you like to get a high-quality dataset to pre-train LLMs in your language? 🌏
At Hugging Face we're preparing a collaborative annotation effort to build an open-source multilingual dataset as part of the Data is Better Together initiative.
Follow the link below, check if your language is listed and sign up to be a Language Lead!
https://forms.gle/s9nGajBh6Pb9G72J6
At Hugging Face we're preparing a collaborative annotation effort to build an open-source multilingual dataset as part of the Data is Better Together initiative.
Follow the link below, check if your language is listed and sign up to be a Language Lead!
https://forms.gle/s9nGajBh6Pb9G72J6
reactedto prithivMLmods's post with 🔥🚀 over 1 year ago
Post
3985
I’m recently experimenting with the Flux-Ultra Realism and Real Anime LoRA models, using the Flux.1-dev model as the base. The model and its demo example are provided in the Flux LoRA DLC collections.📃
🥳Demo : 🔗 prithivMLmods/FLUX-LoRA-DLC
🥳Model:
- prithivMLmods/Canopus-LoRA-Flux-UltraRealism-2.0
- prithivMLmods/Flux-Dev-Real-Anime-LoRA
🥳For more details, please visit the README.md of the Flux LoRA DLC Space & prithivMLmods/lora-space-collections-6714b72e0d49e1c97fbd6a32
🥳Demo : 🔗 prithivMLmods/FLUX-LoRA-DLC
🥳Model:
- prithivMLmods/Canopus-LoRA-Flux-UltraRealism-2.0
- prithivMLmods/Flux-Dev-Real-Anime-LoRA
🥳For more details, please visit the README.md of the Flux LoRA DLC Space & prithivMLmods/lora-space-collections-6714b72e0d49e1c97fbd6a32
- 1 reply
reactedto tomaarsen's post with 🔥 over 1 year ago
Post
7274
📣 Sentence Transformers v3.2.0 is out, marking the biggest release for inference in 2 years! 2 new backends for embedding models: ONNX (+ optimization & quantization) and OpenVINO, allowing for speedups up to 2x-3x AND Static Embeddings for 500x speedups at 10-20% accuracy cost.
1️⃣ ONNX Backend: This backend uses the ONNX Runtime to accelerate model inference on both CPU and GPU, reaching up to 1.4x-3x speedup depending on the precision. We also introduce 2 helper methods for optimizing and quantizing models for (much) faster inference.
2️⃣ OpenVINO Backend: This backend uses Intel their OpenVINO instead, outperforming ONNX in some situations on CPU.
Usage is as simple as
🔒 Another major new feature is Static Embeddings: think word embeddings like GLoVe and word2vec, but modernized. Static Embeddings are bags of token embeddings that are summed together to create text embeddings, allowing for lightning-fast embeddings that don't require any neural networks. They're initialized in one of 2 ways:
1️⃣ via Model2Vec, a new technique for distilling any Sentence Transformer models into static embeddings. Either via a pre-distilled model with
2️⃣ Random initialization. This requires finetuning, but finetuning is extremely quick (e.g. I trained with 3 million pairs in 7 minutes). My final model was 6.6% worse than bge-base-en-v1.5, but 500x faster on CPU.
Full release notes: https://github.com/UKPLab/sentence-transformers/releases/tag/v3.2.0
Documentation on Speeding up Inference: https://sbert.net/docs/sentence_transformer/usage/efficiency.html
1️⃣ ONNX Backend: This backend uses the ONNX Runtime to accelerate model inference on both CPU and GPU, reaching up to 1.4x-3x speedup depending on the precision. We also introduce 2 helper methods for optimizing and quantizing models for (much) faster inference.
2️⃣ OpenVINO Backend: This backend uses Intel their OpenVINO instead, outperforming ONNX in some situations on CPU.
Usage is as simple as
SentenceTransformer("all-MiniLM-L6-v2", backend="onnx"). Does your model not have an ONNX or OpenVINO file yet? No worries - it'll be autoexported for you. Thank me later 😉🔒 Another major new feature is Static Embeddings: think word embeddings like GLoVe and word2vec, but modernized. Static Embeddings are bags of token embeddings that are summed together to create text embeddings, allowing for lightning-fast embeddings that don't require any neural networks. They're initialized in one of 2 ways:
1️⃣ via Model2Vec, a new technique for distilling any Sentence Transformer models into static embeddings. Either via a pre-distilled model with
from_model2vec or with from_distillation where you do the distillation yourself. It'll only take 5 seconds on GPU & 2 minutes on CPU, no dataset needed.2️⃣ Random initialization. This requires finetuning, but finetuning is extremely quick (e.g. I trained with 3 million pairs in 7 minutes). My final model was 6.6% worse than bge-base-en-v1.5, but 500x faster on CPU.
Full release notes: https://github.com/UKPLab/sentence-transformers/releases/tag/v3.2.0
Documentation on Speeding up Inference: https://sbert.net/docs/sentence_transformer/usage/efficiency.html
- 1 reply
reactedto alielfilali01's post with 👍 over 1 year ago
Post
2620
Don't you think we should add a tag "Evaluation" for datasets that are meant to be benchmarks and not for training ?
At least, when someone is collecting a group of datasets from an organization or let's say the whole hub can filter based on that tag and avoid somehow contaminating their "training" data.
At least, when someone is collecting a group of datasets from an organization or let's say the whole hub can filter based on that tag and avoid somehow contaminating their "training" data.
reactedto MoritzLaurer's post with ❤️ over 1 year ago
Post
4782
#phdone - I defended my PhD yesterday! A key lesson: it is amazing how open science and open source can empower beginners with limited resources:
I first learned about instruction-based classifiers like BERT-NLI 3-4 years ago, through the @HuggingFace ZeroShotClassificationPipeline. Digging deeper into this, it was surprisingly easy to find new datasets, newer base models, and reusable fine-tuning scripts on the HF Hub to create my own zeroshot models - although I didn't know much about fine-tuning at the time.
Thanks to the community effect of the Hub, my models were downloaded hundreds of thousands of times after a few months. Seeing my research being useful for people motivated me to improve and upload newer models. Leaving my contact details in the model cards led to academic cooperation and consulting contracts (and eventually my job at HF).
That's the power of open science & open source: learning, sharing, improving, collaborating.
I mean every word in my thesis acknowledgments (screenshot). I'm very grateful to my supervisors @vanatteveldt @CasAndreu @KasperWelbers for their guidance; to @profAndreaRenda and @CEPS_thinktank for enabling me to work part-time during the first year; to @huggingface for creating awesome tools and an awesome platform; and to many others who are not active on social media.
Links to the full thesis and the collection of my most recent models are below.
PS: If someone happens to speak Latin, let me know if my diploma contains some hidden Illuminati code or something :D
I first learned about instruction-based classifiers like BERT-NLI 3-4 years ago, through the @HuggingFace ZeroShotClassificationPipeline. Digging deeper into this, it was surprisingly easy to find new datasets, newer base models, and reusable fine-tuning scripts on the HF Hub to create my own zeroshot models - although I didn't know much about fine-tuning at the time.
Thanks to the community effect of the Hub, my models were downloaded hundreds of thousands of times after a few months. Seeing my research being useful for people motivated me to improve and upload newer models. Leaving my contact details in the model cards led to academic cooperation and consulting contracts (and eventually my job at HF).
That's the power of open science & open source: learning, sharing, improving, collaborating.
I mean every word in my thesis acknowledgments (screenshot). I'm very grateful to my supervisors @vanatteveldt @CasAndreu @KasperWelbers for their guidance; to @profAndreaRenda and @CEPS_thinktank for enabling me to work part-time during the first year; to @huggingface for creating awesome tools and an awesome platform; and to many others who are not active on social media.
Links to the full thesis and the collection of my most recent models are below.
PS: If someone happens to speak Latin, let me know if my diploma contains some hidden Illuminati code or something :D
- 4 replies
posted an update over 1 year ago
Post
1139
🇷🇸 New Benchmark for Serbian Language 🇷🇸
@DjMel and I recently released a new benchmark for Serbian language that measures General Knowledge of LLMs. We had to parse over 20 years of university entrance exams for University of Belgrade, so the dataset is of high quality.
🥇 OAI models still hold the podium places with a significant gap compared to open-source models
🤔 Qwen/Qwen2-7B-Instruct and VAGOsolutions/Llama-3-SauerkrautLM-8b-Instruct models show promising results considering they weren't trained on Serbian language
📈 Best open-source model seems to be Stopwolf/Mustra-7B-Instruct-v0.2, a merge between gordicaleksa/YugoGPT and mistralai/Mistral-7B-Instruct-v0.2
📉 Some models like google/gemma-2-9b-it turned out to be a disappointment with random guessing-like accuracy
Take a look at the whole results at the dataset page:
DjMel/oz-eval
P.S. If you have any constructive criticism or ideas for improvement, feel free to use dataset's Discussions page!
@DjMel and I recently released a new benchmark for Serbian language that measures General Knowledge of LLMs. We had to parse over 20 years of university entrance exams for University of Belgrade, so the dataset is of high quality.
🥇 OAI models still hold the podium places with a significant gap compared to open-source models
🤔 Qwen/Qwen2-7B-Instruct and VAGOsolutions/Llama-3-SauerkrautLM-8b-Instruct models show promising results considering they weren't trained on Serbian language
📈 Best open-source model seems to be Stopwolf/Mustra-7B-Instruct-v0.2, a merge between gordicaleksa/YugoGPT and mistralai/Mistral-7B-Instruct-v0.2
📉 Some models like google/gemma-2-9b-it turned out to be a disappointment with random guessing-like accuracy
Take a look at the whole results at the dataset page:
DjMel/oz-eval
P.S. If you have any constructive criticism or ideas for improvement, feel free to use dataset's Discussions page!
reactedto osanseviero's post with 🔥 about 2 years ago
Post
Diaries of Open Source. Part 5!
🤯Contextual KTO Mistral PairRM: this model combines iterative KTO, SnorkelAI DPO dataset, Allenai PairRM for ranking, Mistral for the base model, and is a very strong model with Claude 3 quality on AlpacaEval 2.0
Final model: ContextualAI/Contextual_KTO_Mistral_PairRM
Dataset: snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset
Leaderboard: https://tatsu-lab.github.io/alpaca_eval/
Base model: mistralai/Mistral-7B-Instruct-v0.2
🤏 tinyBenchmarks: Quick and cheap LLM evaluation!
Code: https://github.com/felipemaiapolo/tinyBenchmarks
Paper: tinyBenchmarks: evaluating LLMs with fewer examples (2402.14992)
Data: tinyBenchmarks/tinyMMLU
🎨Transformers.js 2.16 includes StableLM, speaker verification and diarization, and better chat templating. Try some fun demos!
- Xenova/video-object-detection
- Xenova/cross-encoder-web
- Xenova/the-tokenizer-playground
🏴☠️ Abascus Liberated-Qwen1.5-72B, a Qwen 72B-based model that strongly follows system prompts
Model: abacusai/Liberated-Qwen1.5-72B
👀Design2Code: benchmark of webpage screenshots to code
Data: SALT-NLP/Design2Code
Project https://salt-nlp.github.io/Design2Code/
Paper Design2Code: How Far Are We From Automating Front-End Engineering? (2403.03163)
🌎Data and models around the world
- One of the biggest Italian datasets https://hf.co/datasets/manalog/UsenetArchiveIT
- IndicLLMSuite: argest Pre-training and Instruction Fine-tuning dataset collection across 22 Indic languages ai4bharat/indicllmsuite-65ee7d225c337fcfa0991707
- Hebrew-Gemma-11B, the best base Hebrew model yam-peleg/Hebrew-Gemma-11B
- Komodo-7B, a family of multiple Indonesian languages LLMs Yellow-AI-NLP/komodo-7b-base
You can find the previous part at https://huggingface.co/posts/osanseviero/127895284909100
🤯Contextual KTO Mistral PairRM: this model combines iterative KTO, SnorkelAI DPO dataset, Allenai PairRM for ranking, Mistral for the base model, and is a very strong model with Claude 3 quality on AlpacaEval 2.0
Final model: ContextualAI/Contextual_KTO_Mistral_PairRM
Dataset: snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset
Leaderboard: https://tatsu-lab.github.io/alpaca_eval/
Base model: mistralai/Mistral-7B-Instruct-v0.2
🤏 tinyBenchmarks: Quick and cheap LLM evaluation!
Code: https://github.com/felipemaiapolo/tinyBenchmarks
Paper: tinyBenchmarks: evaluating LLMs with fewer examples (2402.14992)
Data: tinyBenchmarks/tinyMMLU
🎨Transformers.js 2.16 includes StableLM, speaker verification and diarization, and better chat templating. Try some fun demos!
- Xenova/video-object-detection
- Xenova/cross-encoder-web
- Xenova/the-tokenizer-playground
🏴☠️ Abascus Liberated-Qwen1.5-72B, a Qwen 72B-based model that strongly follows system prompts
Model: abacusai/Liberated-Qwen1.5-72B
👀Design2Code: benchmark of webpage screenshots to code
Data: SALT-NLP/Design2Code
Project https://salt-nlp.github.io/Design2Code/
Paper Design2Code: How Far Are We From Automating Front-End Engineering? (2403.03163)
🌎Data and models around the world
- One of the biggest Italian datasets https://hf.co/datasets/manalog/UsenetArchiveIT
- IndicLLMSuite: argest Pre-training and Instruction Fine-tuning dataset collection across 22 Indic languages ai4bharat/indicllmsuite-65ee7d225c337fcfa0991707
- Hebrew-Gemma-11B, the best base Hebrew model yam-peleg/Hebrew-Gemma-11B
- Komodo-7B, a family of multiple Indonesian languages LLMs Yellow-AI-NLP/komodo-7b-base
You can find the previous part at https://huggingface.co/posts/osanseviero/127895284909100
reactedto hakunamatata1997's post with 🤗 about 2 years ago
Post
Hello fellow huggers!
- 2 replies
reactedto davidberenstein1957's post with 🤝 about 2 years ago
Post
A while ago, I presented this Phi2 DPO fine-tune notebook with LoRa. Got some input from @ybelkada about not needing a
https://colab.research.google.com/drive/1PGMj7jlkJaCiSNNihA2NtpILsRgkRXrJ#scrollTo=wXqoH2TMnjjp
ref_model because we can just swap out the LoRa adapters during training. Cool feature 🤓https://colab.research.google.com/drive/1PGMj7jlkJaCiSNNihA2NtpILsRgkRXrJ#scrollTo=wXqoH2TMnjjp
- 5 replies
reactedto JustinLin610's post with ❤️ about 2 years ago
Post
Yesterday we just released Qwen1.5. Maybe someday I can tell more about the experience. But this is is at least a good release even if it is not yet SOTA. There is not so many SOTA by the way. This time, we actually fixed a lot of problems.
1. Context lengths are finally unified for all sizes. Previously, a lot of users kept telling us that 14B only supports 2K (Yeah even dynamic NTK does not work that well and it can only be extended to around 4-5K. Let alone those know nothing about how to use dynamic NTK).
2. If you carefully use our base language models, you will find that they understand special tokens of ChatML, which means that you can directly use LoRA to train on data with ChatML format. Why you can't do this before? This is because if the base language model does not understand the special tokens, you need to make them trained, which means that you should turn on the training of embedding. This is disgusting and it often leads to problems when you use ZeRO3.
3. We did strengthen our base language models except for 72. You should find better base language models, especially for 7 and 14. Why not 72? Nah, hard to say, but will make it better.
4. About the multilingual capabilities. Yes we finally build up our multilingual evaluation system and find out that our new base language models have nice performance in multilingual evaluation for base language models. This tells us that we should pay more attention to the post-training with multilingual data. And we did that too. This is why this time we tell you something about multilingual performance. It is for sure much much better than our models before this release.
5. Chat models are the most promising stuff. Before this release, we gave you the SFT models. But this time, we had very nice SFT+DPO models. Yeah not only annotators like them but also users like them. I am sure you developers will feel that way too.
1. Context lengths are finally unified for all sizes. Previously, a lot of users kept telling us that 14B only supports 2K (Yeah even dynamic NTK does not work that well and it can only be extended to around 4-5K. Let alone those know nothing about how to use dynamic NTK).
2. If you carefully use our base language models, you will find that they understand special tokens of ChatML, which means that you can directly use LoRA to train on data with ChatML format. Why you can't do this before? This is because if the base language model does not understand the special tokens, you need to make them trained, which means that you should turn on the training of embedding. This is disgusting and it often leads to problems when you use ZeRO3.
3. We did strengthen our base language models except for 72. You should find better base language models, especially for 7 and 14. Why not 72? Nah, hard to say, but will make it better.
4. About the multilingual capabilities. Yes we finally build up our multilingual evaluation system and find out that our new base language models have nice performance in multilingual evaluation for base language models. This tells us that we should pay more attention to the post-training with multilingual data. And we did that too. This is why this time we tell you something about multilingual performance. It is for sure much much better than our models before this release.
5. Chat models are the most promising stuff. Before this release, we gave you the SFT models. But this time, we had very nice SFT+DPO models. Yeah not only annotators like them but also users like them. I am sure you developers will feel that way too.
- 5 replies
reactedto BramVanroy's post with ❤️ about 2 years ago
Post
📣 DPO Dutch model release + datasets
After teasing for a while, I am finally releasing **GEITje 7B Ultra**, building upon the great GEITje 7B by @Rijgersberg . New contributions include: large new datasets for SFT (instruction/chat), two datasets for DPO training (i.e. RLAIF), and an SFT and DPO version of GEITje. The READMEs describe everything well (I hope), and I'll also share more info on social medias tomorrow.
For me this is a huge release, the datasets more so than the models. I'm especially pleased with UltraChat, which I created with the intent of having a diverse dataset - the model must be able to communicate with different types of users. So the user questions are created as if they were written by different personas, e.g. language learners, young children, experts, critics, etc. The focus with this is "building a good communication bot that is accessible and can handle different kinds of user input".
I wish I could find the time to also write a paper to get some "academic recognition" but that'll have to wait for now. I just want to bring it to the public so that others can play with it and use it to build new, cool stuff!
I hope that you can all appreciate the work. Let's build some cool stuff with it!
Models:
- Demo: https://huggingface.co/spaces/BramVanroy/GEITje-7B-ultra
- DPO Model: BramVanroy/GEITje-7B-ultra
- SFT model (not recommended): BramVanroy/GEITje-7B-ultra-sft
Datasets with GPT-4 turbo completions:
- No robots (~10k instructions): BramVanroy/no_robots_dutch
- UltraChat (~200k instructions): BramVanroy/ultrachat_200k_dutch
- UltraFeedback (DPO with GPT4+GEITje chat, ~50k): BramVanroy/ultra_feedback_dutch
- Orca DPO Pairs (DPO with GPT4+GEITje chat, ~10k): BramVanroy/orca_dpo_pairs_dutch
After teasing for a while, I am finally releasing **GEITje 7B Ultra**, building upon the great GEITje 7B by @Rijgersberg . New contributions include: large new datasets for SFT (instruction/chat), two datasets for DPO training (i.e. RLAIF), and an SFT and DPO version of GEITje. The READMEs describe everything well (I hope), and I'll also share more info on social medias tomorrow.
For me this is a huge release, the datasets more so than the models. I'm especially pleased with UltraChat, which I created with the intent of having a diverse dataset - the model must be able to communicate with different types of users. So the user questions are created as if they were written by different personas, e.g. language learners, young children, experts, critics, etc. The focus with this is "building a good communication bot that is accessible and can handle different kinds of user input".
I wish I could find the time to also write a paper to get some "academic recognition" but that'll have to wait for now. I just want to bring it to the public so that others can play with it and use it to build new, cool stuff!
I hope that you can all appreciate the work. Let's build some cool stuff with it!
Models:
- Demo: https://huggingface.co/spaces/BramVanroy/GEITje-7B-ultra
- DPO Model: BramVanroy/GEITje-7B-ultra
- SFT model (not recommended): BramVanroy/GEITje-7B-ultra-sft
Datasets with GPT-4 turbo completions:
- No robots (~10k instructions): BramVanroy/no_robots_dutch
- UltraChat (~200k instructions): BramVanroy/ultrachat_200k_dutch
- UltraFeedback (DPO with GPT4+GEITje chat, ~50k): BramVanroy/ultra_feedback_dutch
- Orca DPO Pairs (DPO with GPT4+GEITje chat, ~10k): BramVanroy/orca_dpo_pairs_dutch
- 3 replies