aws-neuron/optimum-neuron-cache
Updated
ā¢
22
AI & ML interests
None defined yet.
Recent Activity
View all activity
dacorvoĀ
updated
a
model
2 days ago
Post
2996
Is there a "one-size-fits-all" recipe for quantizing Large Language Models? š¤
As part of my ongoing work in mixed-precision quantization, I've been exploring this question by measuring layer-by-layer sensitivity. The goal is to see if we can find universal rules for which layers can be quantized aggressively without impacting performance.The results are fascinating and reveal two key insights:
1ļøā£ Sensitivity profiles are like architectural "fingerprints." Models from the same family share strikingly similar sensitivity patterns. As you can see in the charts below for the Gemma and SmolLM families, the ranking and relative sensitivity of the layers remain remarkably consistent. This suggests that the underlying architecture is a primary driver of a model's quantization behavior.
2ļøā£ A "universal" mixed-precision quantization strategy is challenging. While models within a family are similar, these "fingerprints" change dramatically when comparing different architectures like LLaMA, Qwen, and StableLM. This highlights the difficulty in creating a generalized mixed-precision configuration that works optimally across all model families.
However, there is one near-universal truth we uncovered: the mlp.down_proj layer consistently emerges as one of the most sensitive components across all models studied.
This finding strongly resonates with the work in "The Super Weight in Large Language Models" (by Mengxia Yu et al.). The paper identifies that functionally critical parameters, or "super weights," are concentrated in these down_proj layers. Our empirical results provide clear validation for this theory, showing these layers are highly intolerant to precision loss.
In short, while every architecture has a unique sensitivity profile, a fingerprint shaped not only by its core design but also by its specific training dataset and optimization approach, some components remain universally critical!
What are your thoughts?
As part of my ongoing work in mixed-precision quantization, I've been exploring this question by measuring layer-by-layer sensitivity. The goal is to see if we can find universal rules for which layers can be quantized aggressively without impacting performance.The results are fascinating and reveal two key insights:
1ļøā£ Sensitivity profiles are like architectural "fingerprints." Models from the same family share strikingly similar sensitivity patterns. As you can see in the charts below for the Gemma and SmolLM families, the ranking and relative sensitivity of the layers remain remarkably consistent. This suggests that the underlying architecture is a primary driver of a model's quantization behavior.
2ļøā£ A "universal" mixed-precision quantization strategy is challenging. While models within a family are similar, these "fingerprints" change dramatically when comparing different architectures like LLaMA, Qwen, and StableLM. This highlights the difficulty in creating a generalized mixed-precision configuration that works optimally across all model families.
However, there is one near-universal truth we uncovered: the mlp.down_proj layer consistently emerges as one of the most sensitive components across all models studied.
This finding strongly resonates with the work in "The Super Weight in Large Language Models" (by Mengxia Yu et al.). The paper identifies that functionally critical parameters, or "super weights," are concentrated in these down_proj layers. Our empirical results provide clear validation for this theory, showing these layers are highly intolerant to precision loss.
In short, while every architecture has a unique sensitivity profile, a fingerprint shaped not only by its core design but also by its specific training dataset and optimization approach, some components remain universally critical!
What are your thoughts?
- 4 replies
Post
2144
Deploy GPT OSS models with Hugging Face on Azure AI!
Weāre thrilled to enable OpenAI GPT OSS models on Azure AI Model Catalog for Azure users to try the model securely the day of its release.
In our official launch blogpost, thereās a section on how to deploy the model to your Azure AI Hub. Get started today!
https://huggingface.co/blog/welcome-openai-gpt-oss#azure
Weāre thrilled to enable OpenAI GPT OSS models on Azure AI Model Catalog for Azure users to try the model securely the day of its release.
In our official launch blogpost, thereās a section on how to deploy the model to your Azure AI Hub. Get started today!
https://huggingface.co/blog/welcome-openai-gpt-oss#azure
Post
245
We now have the newest Open AI models available on the Dell Enterprise Hub!
We built the Dell Enterprise Hub to provide access to the latest and greatest model from the Hugging Face community to our on-prem customers. Weāre happy to give secure access to this amazing contribution from Open AI on the day of its launch!
https://dell.huggingface.co/
We built the Dell Enterprise Hub to provide access to the latest and greatest model from the Hugging Face community to our on-prem customers. Weāre happy to give secure access to this amazing contribution from Open AI on the day of its launch!
https://dell.huggingface.co/
Post
316
šŖ Qwen/Qwen3ā235BāA22BāInstructā2507āFP8 is now available in Microsoft Azure for oneāclick deployment! š
Check out their blogpost: https://qwenlm.github.io/blog/qwen3/
You can now find it in the Hugging Face Collection in Azure ML or Azure AI Foundry, along with 10k other Hugging Face models š¤š¤
Qwen/Qwen3-235B-A22B-Instruct-2507-FP8
Bear with us for the nonāquantized version.
Check out their blogpost: https://qwenlm.github.io/blog/qwen3/
You can now find it in the Hugging Face Collection in Azure ML or Azure AI Foundry, along with 10k other Hugging Face models š¤š¤
Qwen/Qwen3-235B-A22B-Instruct-2507-FP8
Bear with us for the nonāquantized version.
Post
1542
In our recent push to make more models available on Azure, we recently added SmolLM v3 in the catalog! š
@juanjucm wrote a really detailed guide on how to deploy on Azure AI š¤
https://huggingface.co/docs/microsoft-azure/azure-ai/examples/deploy-smollm3
If you want to see other models, please let us know
@juanjucm wrote a really detailed guide on how to deploy on Azure AI š¤
https://huggingface.co/docs/microsoft-azure/azure-ai/examples/deploy-smollm3
If you want to see other models, please let us know
- 1 reply
Post
198
š New in Azure Model Catalog: NVIDIA Parakeet TDT 0.6B V2
We're excited to welcome Parakeet TDT 0.6B V2āa state-of-the-art English speech-to-text modelāto the Azure Foundry Model Catalog.
What is it?
A powerful ASR model built on the FastConformer-TDT architecture, offering:
š Word-level timestamps
āļø Automatic punctuation & capitalization
š Strong performance across noisy and real-world audio
It runs with NeMo, NVIDIAās optimized inference engine.
Want to give it a try? š§ You can test it with your own audio (up to 3 hours) on Hugging Face Spaces before deploying.If it fits your need, deploy easily from the Hugging Face Hub or Azure ML Studio with secure, scalable infrastructure!
š Learn more by following this guide written by @alvarobartt
https://huggingface.co/docs/microsoft-azure/azure-ai/examples/deploy-nvidia-parakeet-asr
We're excited to welcome Parakeet TDT 0.6B V2āa state-of-the-art English speech-to-text modelāto the Azure Foundry Model Catalog.
What is it?
A powerful ASR model built on the FastConformer-TDT architecture, offering:
š Word-level timestamps
āļø Automatic punctuation & capitalization
š Strong performance across noisy and real-world audio
It runs with NeMo, NVIDIAās optimized inference engine.
Want to give it a try? š§ You can test it with your own audio (up to 3 hours) on Hugging Face Spaces before deploying.If it fits your need, deploy easily from the Hugging Face Hub or Azure ML Studio with secure, scalable infrastructure!
š Learn more by following this guide written by @alvarobartt
https://huggingface.co/docs/microsoft-azure/azure-ai/examples/deploy-nvidia-parakeet-asr
Post
1257
If you want to dive into how the HF team worked with
@seungrokj
at
@AMD
to optimize kernels on MI300, you should give a read to our latest blog!
Such a great educational material for anyone curious about the world of optimizing low level ML.
https://huggingface.co/blog/mi300kernels
to optimize kernels on MI300, you should give a read to our latest blog!
Such a great educational material for anyone curious about the world of optimizing low level ML.
https://huggingface.co/blog/mi300kernels
badaouiĀ
authored
a
paper
about 1 month ago
Post
1627
In case you missed it, Hugging Face expanded its collaboration with Azure a few weeks ago with a curated catalog of 10,000 models, accessible from Azure AI Foundry and Azure ML!
@alvarobartt cooked during these last days to prepare the one and only documentation you need, if you wanted to deploy Hugging Face models on Azure. It comes with an FAQ, great guides and examples on how to deploy VLMs, LLMs, smolagents and more to come very soon.
We need your feedback: come help us and let us know what else you want to see, which model we should add to the collection, which model task we should prioritize adding, what else we should build a tutorial for. Youāre just an issue away on our GitHub repo!
https://huggingface.co/docs/microsoft-azure/index
@alvarobartt cooked during these last days to prepare the one and only documentation you need, if you wanted to deploy Hugging Face models on Azure. It comes with an FAQ, great guides and examples on how to deploy VLMs, LLMs, smolagents and more to come very soon.
We need your feedback: come help us and let us know what else you want to see, which model we should add to the collection, which model task we should prioritize adding, what else we should build a tutorial for. Youāre just an issue away on our GitHub repo!
https://huggingface.co/docs/microsoft-azure/index
jeffboudierĀ
posted
an
update
about 2 months ago
Post
490
AMD summer hackathons are here!
A chance to get hands-on with MI300X GPUs and accelerate models.
š«š· Paris - Station F - July 5-6
š®š³ Mumbai - July 12-13
š®š³ Bengaluru - July 19-20
Hugging Face and GPU Mode will be on site and on July 6 in Paris @ror will share lessons learned while building new kernels to accelerate Llama 3.1 405B on ROCm
Register to Paris event: https://lu.ma/fmvdjmur?tk=KeAbiP
All dates: https://lu.ma/calendar/cal-3sxhD5FdxWsMDIz
A chance to get hands-on with MI300X GPUs and accelerate models.
š«š· Paris - Station F - July 5-6
š®š³ Mumbai - July 12-13
š®š³ Bengaluru - July 19-20
Hugging Face and GPU Mode will be on site and on July 6 in Paris @ror will share lessons learned while building new kernels to accelerate Llama 3.1 405B on ROCm
Register to Paris event: https://lu.ma/fmvdjmur?tk=KeAbiP
All dates: https://lu.ma/calendar/cal-3sxhD5FdxWsMDIz
Post
3217
Hackathons in Paris on July 5th and 6th!
Hugging Face just wrapped 4 months of deep work with AMD to push kernel-level optimization on their MI300X GPUs. Now, it's time to share everything we learned.
Join us in Paris at STATION F for a hands-on weekend of workshops and a hackathon focused on making open-source LLMs faster and more efficient on AMD.
Prizes, amazing host speakers, ... if you want more details, navigate to https://lu.ma/fmvdjmur!
Hugging Face just wrapped 4 months of deep work with AMD to push kernel-level optimization on their MI300X GPUs. Now, it's time to share everything we learned.
Join us in Paris at STATION F for a hands-on weekend of workshops and a hackathon focused on making open-source LLMs faster and more efficient on AMD.
Prizes, amazing host speakers, ... if you want more details, navigate to https://lu.ma/fmvdjmur!
- 2 replies
Post
2403
Webinar Alert
Build your first chatbot with a Hugging Face Spaces frontend and Gaudi-powered backend with @bconsolvo ! He will teach you how to build an LLM-powered chatbot using Streamlit and Hugging Face Spacesāintegrating a model endpoint hosted on an IntelĀ® GaudiĀ® accelerator.
Beginners are welcome
https://web.cvent.com/event/70e11f23-7c52-4994-a918-96fa9d5e935f/summary
Build your first chatbot with a Hugging Face Spaces frontend and Gaudi-powered backend with @bconsolvo ! He will teach you how to build an LLM-powered chatbot using Streamlit and Hugging Face Spacesāintegrating a model endpoint hosted on an IntelĀ® GaudiĀ® accelerator.
Beginners are welcome
https://web.cvent.com/event/70e11f23-7c52-4994-a918-96fa9d5e935f/summary
- 1 reply
jeffboudierĀ
posted
an
update
2 months ago
Post
1687
Today we launched Training Cluster as a Service, to make the new DGX Cloud Lepton supercloud easily accessible to AI researchers.
Hugging Face will collaborate with NVIDIA to provision and set up GPU training clusters to make them available for the duration of training runs.
Hugging Face organizations can sign up here: https://huggingface.co/training-cluster
Hugging Face will collaborate with NVIDIA to provision and set up GPU training clusters to make them available for the duration of training runs.
Hugging Face organizations can sign up here: https://huggingface.co/training-cluster
jeffboudierĀ
posted
an
update
3 months ago
Post
2462
š Congrats
@jinanz
adding TimesFM times series forecasting to Transformers!
Learn how to use TimesFM in this blog post by the Nutanix team: https://huggingface.co/blog/Nutanix/introducing-timesfm-for-time-series-forcasting
Learn how to use TimesFM in this blog post by the Nutanix team: https://huggingface.co/blog/Nutanix/introducing-timesfm-for-time-series-forcasting
jeffboudierĀ
posted
an
update
3 months ago
Post
496
Wrapping up a week of shipping and announcements with Dell Enterprise Hub now featuring AI Applications, on-device models for AI PCs, a new CLI and Python SDK... all you need for building AI on premises!
Blog post has all the details: https://huggingface.co/blog/dell-ai-applications
Blog post has all the details: https://huggingface.co/blog/dell-ai-applications
jeffboudierĀ
posted
an
update
3 months ago
Post
2596
Transcribing 1 hour of audio for less than $0.01 š¤Æ
@mfuntowicz cooked with 8x faster Whisper speech recognition - whisper-large-v3-turbo transcribes at 100x real time on a $0.80/hr L4 GPU!
How they did it: https://huggingface.co/blog/fast-whisper-endpoints
1-click deploy with HF Inference Endpoints: https://endpoints.huggingface.co/new?repository=openai%2Fwhisper-large-v3-turbo&vendor=aws®ion=us-east&accelerator=gpu&instance_id=aws-us-east-1-nvidia-l4-x1&task=automatic-speech-recognition&no_suggested_compute=true
@mfuntowicz cooked with 8x faster Whisper speech recognition - whisper-large-v3-turbo transcribes at 100x real time on a $0.80/hr L4 GPU!
How they did it: https://huggingface.co/blog/fast-whisper-endpoints
1-click deploy with HF Inference Endpoints: https://endpoints.huggingface.co/new?repository=openai%2Fwhisper-large-v3-turbo&vendor=aws®ion=us-east&accelerator=gpu&instance_id=aws-us-east-1-nvidia-l4-x1&task=automatic-speech-recognition&no_suggested_compute=true
jeffboudierĀ
posted
an
update
3 months ago
Post
3023
So many orgs on HF would really benefit from security and governance built into Enterprise Hub - I wrote a guide on why and how upgrade: https://huggingface.co/spaces/jeffboudier/how-to-upgrade-to-enterprise
For instance, did you know about Resource Groups?
For instance, did you know about Resource Groups?
Post
1999
If you haven't had the chance to test the latest open model from Meta, Llama 4 Maverick, go try it on AMD MI 300 on Hugging Face!
amd/llama4-maverick-17b-128e-mi-amd
amd/llama4-maverick-17b-128e-mi-amd
jeffboudierĀ
posted
an
update
4 months ago
Post
2210
Llama4 is out and Scout is already on the Dell Enterprise Hub to deploy on Dell systems š dell.huggingface.co