Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference
Paper
•
2412.13663
•
Published
•
103
Distillation Hugs
community
AI & ML interests
None defined yet.
Recent Activity
Distillation-Hugs's activity
tomaarsen
authored
a
paper
6 days ago
Post
6417
We outperform Llama 70B with Llama 3B on hard math by scaling test-time compute 🔥
How? By combining step-wise reward models with tree search algorithms :)
We show that smol models can match or exceed the performance of their much larger siblings when given enough "time to think"
We're open sourcing the full recipe and sharing a detailed blog post.
In our blog post we cover:
📈 Compute-optimal scaling: How we implemented DeepMind's recipe to boost the mathematical capabilities of open models at test-time.
🎄 Diverse Verifier Tree Search (DVTS): An unpublished extension we developed to the verifier-guided tree search technique. This simple yet effective method improves diversity and delivers better performance, particularly at large test-time compute budgets.
🧭 Search and Learn: A lightweight toolkit for implementing search strategies with LLMs and built for speed with vLLM
Here's the links:
- Blog post: HuggingFaceH4/blogpost-scaling-test-time-compute
- Code: https://github.com/huggingface/search-and-learn
Enjoy!
How? By combining step-wise reward models with tree search algorithms :)
We show that smol models can match or exceed the performance of their much larger siblings when given enough "time to think"
We're open sourcing the full recipe and sharing a detailed blog post.
In our blog post we cover:
📈 Compute-optimal scaling: How we implemented DeepMind's recipe to boost the mathematical capabilities of open models at test-time.
🎄 Diverse Verifier Tree Search (DVTS): An unpublished extension we developed to the verifier-guided tree search technique. This simple yet effective method improves diversity and delivers better performance, particularly at large test-time compute budgets.
🧭 Search and Learn: A lightweight toolkit for implementing search strategies with LLMs and built for speed with vLLM
Here's the links:
- Blog post: HuggingFaceH4/blogpost-scaling-test-time-compute
- Code: https://github.com/huggingface/search-and-learn
Enjoy!
- 2 replies
Post
1782
SmolVLM speeding locally on a laptop thanks to mlx-vlm and
@Gradio ! Try it with two lines:
pip install git+https://github.com/andimarafioti/mlx-vlm.git@stream-generate-fix
python -m mlx_vlm.chat_ui --model mlx-community/SmolVLM-Instruct-8bit
Gotta love the MLX community! Big thanks to @pcuenq and @prince_canuma !
@Gradio ! Try it with two lines:
pip install git+https://github.com/andimarafioti/mlx-vlm.git@stream-generate-fix
python -m mlx_vlm.chat_ui --model mlx-community/SmolVLM-Instruct-8bit
Gotta love the MLX community! Big thanks to @pcuenq and @prince_canuma !
Post
3242
Let's go! We are releasing SmolVLM, a smol 2B VLM built for on-device inference that outperforms all models at similar GPU RAM usage and tokens throughputs.
- SmolVLM generates tokens 7.5 to 16 times faster than Qwen2-VL! 🤯
- Other models at this size crash a laptop, but SmolVLM comfortably generates 17 tokens/sec on a macbook! 🚀
- SmolVLM can be fine-tuned on a Google collab! Or process millions of documents with a consumer GPU!
- SmolVLM even outperforms larger models in video benchmarks, despite not even being trained on videos!
Check out more!
Demo: HuggingFaceTB/SmolVLM
Blog: https://huggingface.co/blog/smolvlm
Model: HuggingFaceTB/SmolVLM-Instruct
Fine-tuning script: https://github.com/huggingface/smollm/blob/main/finetuning/Smol_VLM_FT.ipynb
- SmolVLM generates tokens 7.5 to 16 times faster than Qwen2-VL! 🤯
- Other models at this size crash a laptop, but SmolVLM comfortably generates 17 tokens/sec on a macbook! 🚀
- SmolVLM can be fine-tuned on a Google collab! Or process millions of documents with a consumer GPU!
- SmolVLM even outperforms larger models in video benchmarks, despite not even being trained on videos!
Check out more!
Demo: HuggingFaceTB/SmolVLM
Blog: https://huggingface.co/blog/smolvlm
Model: HuggingFaceTB/SmolVLM-Instruct
Fine-tuning script: https://github.com/huggingface/smollm/blob/main/finetuning/Smol_VLM_FT.ipynb
Post
5289
I just released Sentence Transformers v3.3.0 & it's huge! 4.5x speedup for CPU with OpenVINO int8 static quantization, training with prompts for a free perf. boost, PEFT integration, evaluation on NanoBEIR, and more! Details:
1. We integrate Post-Training Static Quantization using OpenVINO, a very efficient solution for CPUs that processes 4.78x as many texts per second on average, while only hurting performance by 0.36% on average. There's a new
2. We add the option to train with prompts, e.g. strings like "query: ", "search_document: " or "Represent this sentence for searching relevant passages: ". It's as simple as using the
3. Sentence Transformers now supports training PEFT adapters via 7 new methods for adding new adapters or loading pre-trained ones. You can also directly load a trained adapter with SentenceTransformer as if it's a normal model. Very useful for e.g. 1) training multiple adapters on 1 base model, 2) training bigger models than otherwise possible, or 3) cheaply hosting multiple models by switching multiple adapters on 1 base model.
4. We added easy evaluation on NanoBEIR, a subset of BEIR a.k.a. the MTEB Retrieval benchmark. It contains 13 datasets with 50 queries and up to 10k documents each. Evaluation is fast, and can easily be done during training to track your model's performance on general-purpose information retrieval tasks.
Additionally, we also deprecate Python 3.8, add better compatibility with Transformers v4.46.0, and more. Read the full release notes here: https://github.com/UKPLab/sentence-transformers/releases/tag/v3.3.0
1. We integrate Post-Training Static Quantization using OpenVINO, a very efficient solution for CPUs that processes 4.78x as many texts per second on average, while only hurting performance by 0.36% on average. There's a new
export_static_quantized_openvino_model method to quantize a model.2. We add the option to train with prompts, e.g. strings like "query: ", "search_document: " or "Represent this sentence for searching relevant passages: ". It's as simple as using the
prompts argument in SentenceTransformerTrainingArguments. Our experiments show that you can easily reach 0.66% to 0.90% relative performance improvement on NDCG@10 at no extra cost by adding "query: " before each training query and "document: " before each training answer.3. Sentence Transformers now supports training PEFT adapters via 7 new methods for adding new adapters or loading pre-trained ones. You can also directly load a trained adapter with SentenceTransformer as if it's a normal model. Very useful for e.g. 1) training multiple adapters on 1 base model, 2) training bigger models than otherwise possible, or 3) cheaply hosting multiple models by switching multiple adapters on 1 base model.
4. We added easy evaluation on NanoBEIR, a subset of BEIR a.k.a. the MTEB Retrieval benchmark. It contains 13 datasets with 50 queries and up to 10k documents each. Evaluation is fast, and can easily be done during training to track your model's performance on general-purpose information retrieval tasks.
Additionally, we also deprecate Python 3.8, add better compatibility with Transformers v4.46.0, and more. Read the full release notes here: https://github.com/UKPLab/sentence-transformers/releases/tag/v3.3.0
Post
6830
📣 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
Post
1995
I've just shipped the Sentence Transformers v3.1.1 patch release, fixing the hard negatives mining utility for some models. This utility is extremely useful to get more performance out of your embedding training data.
⛏ Hard negatives are texts that are rather similar to some anchor text (e.g. a query), but are not the correct match. They're difficult for a model to distinguish from the correct answer, often resulting in a stronger model after training.
🔓 Beyond that, this release removes the numpy<2 restriction from v3.1.0. This was previously required for Windows as not all third-party libraries were updated to support numpy v2. With Sentence Transformers, you can now choose v1 or v2 of numpy.
Check out the full release notes here: https://github.com/UKPLab/sentence-transformers/releases/tag/v3.1.1
I'm looking forward to releasing v3.2, I have some exciting things planned 🚀
⛏ Hard negatives are texts that are rather similar to some anchor text (e.g. a query), but are not the correct match. They're difficult for a model to distinguish from the correct answer, often resulting in a stronger model after training.
mine_hard_negatives docs: https://sbert.net/docs/package_reference/util.html#sentence_transformers.util.mine_hard_negatives🔓 Beyond that, this release removes the numpy<2 restriction from v3.1.0. This was previously required for Windows as not all third-party libraries were updated to support numpy v2. With Sentence Transformers, you can now choose v1 or v2 of numpy.
Check out the full release notes here: https://github.com/UKPLab/sentence-transformers/releases/tag/v3.1.1
I'm looking forward to releasing v3.2, I have some exciting things planned 🚀
Post
2037
🎉SetFit v1.1.0 is out! Training efficient classifiers on CPU or GPU now uses the Sentence Transformers Trainer, and we resolved a lot of issues caused by updates of third-party libraries (like Transformers). Details:
Training a SetFit classifier model consists of 2 phases:
1. Finetuning a Sentence Transformer embedding model
2. Training a Classifier to map embeddings -> classes
🔌The first phase now uses the SentenceTransformerTrainer that was introduced in the Sentence Transformers v3 update. This brings some immediate upsides like MultiGPU support, without any (intended) breaking changes.
➡️ Beyond that, we softly deprecated the "evaluation_strategy" argument in favor of "eval_strategy" (following a Transformers deprecation), and deprecated Python 3.7. In return, we add official support for Python 3.11 and 3.12.
✨ There's some more minor changes too, like max_steps and eval_max_steps now being a hard limit instead of an approximate one, training/validation losses now logging nicely in Notebooks, and the "device" parameter no longer being ignored in some situations.
Check out the full release notes here: https://github.com/huggingface/setfit/releases/tag/v1.1.0
Or read the documentation: https://huggingface.co/docs/setfit
Or check out the public SetFit models for inspiration: https://huggingface.co/models?library=setfit&sort=created
P.s. the model in the code snippet trained in 1 minute and it can classify ~6000 sentences per second on my GPU.
Training a SetFit classifier model consists of 2 phases:
1. Finetuning a Sentence Transformer embedding model
2. Training a Classifier to map embeddings -> classes
🔌The first phase now uses the SentenceTransformerTrainer that was introduced in the Sentence Transformers v3 update. This brings some immediate upsides like MultiGPU support, without any (intended) breaking changes.
➡️ Beyond that, we softly deprecated the "evaluation_strategy" argument in favor of "eval_strategy" (following a Transformers deprecation), and deprecated Python 3.7. In return, we add official support for Python 3.11 and 3.12.
✨ There's some more minor changes too, like max_steps and eval_max_steps now being a hard limit instead of an approximate one, training/validation losses now logging nicely in Notebooks, and the "device" parameter no longer being ignored in some situations.
Check out the full release notes here: https://github.com/huggingface/setfit/releases/tag/v1.1.0
Or read the documentation: https://huggingface.co/docs/setfit
Or check out the public SetFit models for inspiration: https://huggingface.co/models?library=setfit&sort=created
P.s. the model in the code snippet trained in 1 minute and it can classify ~6000 sentences per second on my GPU.
Post
1082
Hugging face presents FineVideo 🎥! Unlocking the next generation of Video understanding 🚀
🤯3400 hours of annotated Creative Common videos with rich character descriptions, scene splits, mood, and content descriptions per scene as well as QA pairs.
🔥
@mfarre processed over 2M videos of Youtube-CC to make this incredibly powerful selection.
Very psyched to fine-tune idefics on this dataset. ⚡️
Explore the videos: HuggingFaceFV/FineVideo-Explorer
🤯3400 hours of annotated Creative Common videos with rich character descriptions, scene splits, mood, and content descriptions per scene as well as QA pairs.
🔥
@mfarre processed over 2M videos of Youtube-CC to make this incredibly powerful selection.
Very psyched to fine-tune idefics on this dataset. ⚡️
Explore the videos: HuggingFaceFV/FineVideo-Explorer
Post
3743
🚀 Sentence Transformers v3.1 is out! Featuring a hard negatives mining utility to get better models out of your data, a new strong loss function, training with streaming datasets, custom modules, bug fixes, small additions and docs changes. Here's the details:
⛏ Hard Negatives Mining Utility: Hard negatives are texts that are rather similar to some anchor text (e.g. a question), but are not the correct match. They're difficult for a model to distinguish from the correct answer, often resulting in a stronger model after training.
📉 New loss function: This loss function works very well for symmetric tasks (e.g. clustering, classification, finding similar texts/paraphrases) and a bit less so for asymmetric tasks (e.g. question-answer retrieval).
💾 Streaming datasets: You can now train with the datasets.IterableDataset, which doesn't require downloading the full dataset to disk before training. As simple as "streaming=True" in your "datasets.load_dataset".
🧩 Custom Modules: Model authors can now customize a lot more of the components that make up Sentence Transformer models, allowing for a lot more flexibility (e.g. multi-modal, model-specific quirks, etc.)
✨ New arguments to several methods: encode_multi_process gets a progress bar, push_to_hub can now be done to different branches, and CrossEncoders can be downloaded to specific cache directories.
🐛 Bug fixes: Too many to name here, check out the release notes!
📝 Documentation: A particular focus on clarifying the batch samplers in the Package Reference this release.
Check out the full release notes here ⭐: https://github.com/UKPLab/sentence-transformers/releases/tag/v3.1.0
I'm very excited to hear your feedback, and I'm looking forward to the future changes that I have planned, such as ONNX inference! I'm also open to suggestions for new features: feel free to send me your ideas.
⛏ Hard Negatives Mining Utility: Hard negatives are texts that are rather similar to some anchor text (e.g. a question), but are not the correct match. They're difficult for a model to distinguish from the correct answer, often resulting in a stronger model after training.
📉 New loss function: This loss function works very well for symmetric tasks (e.g. clustering, classification, finding similar texts/paraphrases) and a bit less so for asymmetric tasks (e.g. question-answer retrieval).
💾 Streaming datasets: You can now train with the datasets.IterableDataset, which doesn't require downloading the full dataset to disk before training. As simple as "streaming=True" in your "datasets.load_dataset".
🧩 Custom Modules: Model authors can now customize a lot more of the components that make up Sentence Transformer models, allowing for a lot more flexibility (e.g. multi-modal, model-specific quirks, etc.)
✨ New arguments to several methods: encode_multi_process gets a progress bar, push_to_hub can now be done to different branches, and CrossEncoders can be downloaded to specific cache directories.
🐛 Bug fixes: Too many to name here, check out the release notes!
📝 Documentation: A particular focus on clarifying the batch samplers in the Package Reference this release.
Check out the full release notes here ⭐: https://github.com/UKPLab/sentence-transformers/releases/tag/v3.1.0
I'm very excited to hear your feedback, and I'm looking forward to the future changes that I have planned, such as ONNX inference! I'm also open to suggestions for new features: feel free to send me your ideas.
-
- 3 replies
andito
authored
2
papers
4 months ago
Post
1614
🚀 Introducing Hugging Face's Multilingual Speech-to-Speech! 🎤
💬Our modular, cross-platform pipeline to run GPT4o-like experiences on device can now seamlessly switch languages mid-conversation with an imperceptible 100ms delay.
🌟 Building on an amazing early reception with 2600 stars on GitHub 🌟
🚀 We are expanding the library to support multiple languages
🔥 Try it out with a flag: --language fr
🤯 Or don't set the flag and let the system detect the language
💡 What feature should we add next?
💬Our modular, cross-platform pipeline to run GPT4o-like experiences on device can now seamlessly switch languages mid-conversation with an imperceptible 100ms delay.
🌟 Building on an amazing early reception with 2600 stars on GitHub 🌟
🚀 We are expanding the library to support multiple languages
🔥 Try it out with a flag: --language fr
🤯 Or don't set the flag and let the system detect the language
💡 What feature should we add next?
-
- 1 reply
andito
authored
a
paper
4 months ago
Post
2057
Releasing HQQ Llama-3.1-70b 4-bit quantized version! Check it out at
mobiuslabsgmbh/Llama-3.1-70b-instruct_4bitgs64_hqq.
Achieves 99% of the base model performance across various benchmarks! Details in the model card.
Achieves 99% of the base model performance across various benchmarks! Details in the model card.
Post
1788
Excited to announce the release of our high-quality Llama-3.1 8B 4-bit HQQ/calibrated quantized model! Achieving an impressive 99.3% relative performance to FP16, it also delivers the fastest inference speed for transformers.
mobiuslabsgmbh/Llama-3.1-8b-instruct_4bitgs64_hqq_calib
mobiuslabsgmbh/Llama-3.1-8b-instruct_4bitgs64_hqq_calib
- 1 reply
conceptofmind
authored
a
paper
5 months ago
keval-sha
updated
a
model
6 months ago
Post
3922
@Omartificial-Intelligence-Space
has trained and released 6 Arabic embedding models for semantic similarity. 4 of them outperform all previous models on the STS17 Arabic-Arabic task!
📚 Trained on a large dataset of 558k Arabic triplets translated from the AllNLI triplet dataset: Omartificial-Intelligence-Space/Arabic-NLi-Triplet
6️⃣ 6 different base models: AraBERT, MarBERT, LaBSE, MiniLM, paraphrase-multilingual-mpnet-base, mpnet-base, ranging from 109M to 471M parameters.
🪆 Trained with a Matryoshka loss, allowing you to truncate embeddings with minimal performance loss: smaller embeddings are faster to compare.
📈 Outperforms all commonly used multilingual models like intfloat/multilingual-e5-large, sentence-transformers/paraphrase-multilingual-mpnet-base-v2, and sentence-transformers/LaBSE.
Check them out here:
- Omartificial-Intelligence-Space/Arabic-mpnet-base-all-nli-triplet
- Omartificial-Intelligence-Space/Arabic-all-nli-triplet-Matryoshka
- Omartificial-Intelligence-Space/Arabert-all-nli-triplet-Matryoshka
- Omartificial-Intelligence-Space/Arabic-labse-Matryoshka
- Omartificial-Intelligence-Space/Marbert-all-nli-triplet-Matryoshka
- Omartificial-Intelligence-Space/Arabic-MiniLM-L12-v2-all-nli-triplet
Or the collection with all: Omartificial-Intelligence-Space/arabic-matryoshka-embedding-models-666f764d3b570f44d7f77d4e
My personal favourite is likely Omartificial-Intelligence-Space/Arabert-all-nli-triplet-Matryoshka: a very efficient 135M parameters & scores #1 on mteb/leaderboard.
📚 Trained on a large dataset of 558k Arabic triplets translated from the AllNLI triplet dataset: Omartificial-Intelligence-Space/Arabic-NLi-Triplet
6️⃣ 6 different base models: AraBERT, MarBERT, LaBSE, MiniLM, paraphrase-multilingual-mpnet-base, mpnet-base, ranging from 109M to 471M parameters.
🪆 Trained with a Matryoshka loss, allowing you to truncate embeddings with minimal performance loss: smaller embeddings are faster to compare.
📈 Outperforms all commonly used multilingual models like intfloat/multilingual-e5-large, sentence-transformers/paraphrase-multilingual-mpnet-base-v2, and sentence-transformers/LaBSE.
Check them out here:
- Omartificial-Intelligence-Space/Arabic-mpnet-base-all-nli-triplet
- Omartificial-Intelligence-Space/Arabic-all-nli-triplet-Matryoshka
- Omartificial-Intelligence-Space/Arabert-all-nli-triplet-Matryoshka
- Omartificial-Intelligence-Space/Arabic-labse-Matryoshka
- Omartificial-Intelligence-Space/Marbert-all-nli-triplet-Matryoshka
- Omartificial-Intelligence-Space/Arabic-MiniLM-L12-v2-all-nli-triplet
Or the collection with all: Omartificial-Intelligence-Space/arabic-matryoshka-embedding-models-666f764d3b570f44d7f77d4e
My personal favourite is likely Omartificial-Intelligence-Space/Arabert-all-nli-triplet-Matryoshka: a very efficient 135M parameters & scores #1 on mteb/leaderboard.
- 1 reply
adarshxs
authored
a
paper
6 months ago