mlx-community/Llama-3.2-1B-Instruct-4bit
Text Generation
•
Updated
•
3.4M
•
12
Jorge Vicente Mendoza
0xjorgev
·
AI & ML interests
Stable Diffusion , llama, transformers, GPTs
Recent Activity
liked
a model
4 days ago
mlx-community/Llama-3.2-1B-Instruct-4bit
liked
a model
4 days ago
mlx-community/Phi-3-mini-4k-instruct-4bit
liked
a model
6 days ago
anthonymikinka/OpenELM-270M-Instruct-128-FP16ComputePrecisionv2_chunked_pipeline
Organizations
0xjorgev's activity
liked
2
models
4 days ago
liked
3
models
6 days ago
Can't compile on Xcode
3
#2 opened 2 months ago
by
0xjorgev
liked
a
model
2 months ago
liked
4
models
3 months ago
reacted to
pcuenq's
post with 🚀
4 months ago
Post
5054
OpenELM in Core ML
Apple recently released a set of efficient LLMs in sizes varying between 270M and 3B parameters. Their quality, according to benchmarks, is similar to OLMo models of comparable size, but they required half the pre-training tokens because they use layer-wise scaling, where the number of attention heads increases in deeper layers.
I converted these models to Core ML, for use on Apple Silicon, using this script: https://gist.github.com/pcuenca/23cd08443460bc90854e2a6f0f575084. The converted models were uploaded to this community in the Hub for anyone that wants to integrate inside their apps: corenet-community/openelm-core-ml-6630c6b19268a5d878cfd194
The conversion was done with the following parameters:
- Precision: float32.
- Sequence length: fixed to 128.
With swift-transformers (https://github.com/huggingface/swift-transformers), I'm getting about 56 tok/s with the 270M on my M1 Max, and 6.5 with the largest 3B model. These speeds could be improved by converting to
I'm also looking at optimizing inference using an experimental kv cache in swift-transformers. It's a bit tricky because the layers have varying number of attention heads, but I'm curious to see how much this feature can accelerate performance in this model family :)
Regarding the instruct fine-tuned models, I don't know the chat template that was used. The models use the Llama 2 tokenizer, but the Llama 2 chat template, or the default Alignment Handbook one that was used to train, are not recognized. Any ideas on this welcome!
Apple recently released a set of efficient LLMs in sizes varying between 270M and 3B parameters. Their quality, according to benchmarks, is similar to OLMo models of comparable size, but they required half the pre-training tokens because they use layer-wise scaling, where the number of attention heads increases in deeper layers.
I converted these models to Core ML, for use on Apple Silicon, using this script: https://gist.github.com/pcuenca/23cd08443460bc90854e2a6f0f575084. The converted models were uploaded to this community in the Hub for anyone that wants to integrate inside their apps: corenet-community/openelm-core-ml-6630c6b19268a5d878cfd194
The conversion was done with the following parameters:
- Precision: float32.
- Sequence length: fixed to 128.
With swift-transformers (https://github.com/huggingface/swift-transformers), I'm getting about 56 tok/s with the 270M on my M1 Max, and 6.5 with the largest 3B model. These speeds could be improved by converting to
float16. However, there's some precision loss somewhere and generation doesn't work in float16 mode yet. I'm looking into this and will keep you posted! Or take a look at this issue if you'd like to help: https://github.com/huggingface/swift-transformers/issues/95I'm also looking at optimizing inference using an experimental kv cache in swift-transformers. It's a bit tricky because the layers have varying number of attention heads, but I'm curious to see how much this feature can accelerate performance in this model family :)
Regarding the instruct fine-tuned models, I don't know the chat template that was used. The models use the Llama 2 tokenizer, but the Llama 2 chat template, or the default Alignment Handbook one that was used to train, are not recognized. Any ideas on this welcome!
- 4 replies
upvoted
a
collection
4 months ago
Is it possible to port from .mlpackage to .mlmodelc
3
#4 opened 4 months ago
by
0xjorgev
reacted to
MonsterMMORPG's
post with 🔥
6 months ago
Post
5343
LivePortrait AI: Transform Static Photos into Talking Videos. Now supporting Video-to-Video conversion and Superior Expression Transfer at Remarkable Speed
A new tutorial is anticipated to showcase the latest changes and features in V3, including Video-to-Video capabilities and additional enhancements.
This post provides information for both Windows (local) and Cloud installations (Massed Compute, RunPod, and free Kaggle Account).
🔗 Windows Local Installation Tutorial ️⤵️
▶️ https://youtu.be/FPtpNrmuwXk
🔗 Cloud (no-GPU) Installations Tutorial for Massed Compute, RunPod and free Kaggle Account ️⤵️
▶️ https://youtu.be/wG7oPp01COg
The V3 update introduces video-to-video functionality. If you're seeking a one-click installation method for LivePortrait, an open-source zero-shot image-to-animation application on Windows, for local use, this tutorial is essential. It introduces the cutting-edge image-to-animation open-source generator Live Portrait. Simply provide a static image and a driving video to create an impressive animation in seconds. LivePortrait is incredibly fast and adept at preserving facial expressions from the input video. The results are truly astonishing.
With the V3 update adding video-to-video functionality, those interested in using LivePortrait but lacking a powerful GPU, using a Mac, or preferring cloud-based solutions will find this tutorial invaluable. It guides you through the one-click installation and usage of LivePortrait on #MassedCompute, #RunPod, and even a free #Kaggle account. After following this tutorial, you'll find running LivePortrait on cloud services as straightforward as running it locally. LivePortrait is the latest state-of-the-art static image to talking animation generator, surpassing even paid services in both speed and quality.
A new tutorial is anticipated to showcase the latest changes and features in V3, including Video-to-Video capabilities and additional enhancements.
This post provides information for both Windows (local) and Cloud installations (Massed Compute, RunPod, and free Kaggle Account).
🔗 Windows Local Installation Tutorial ️⤵️
▶️ https://youtu.be/FPtpNrmuwXk
🔗 Cloud (no-GPU) Installations Tutorial for Massed Compute, RunPod and free Kaggle Account ️⤵️
▶️ https://youtu.be/wG7oPp01COg
The V3 update introduces video-to-video functionality. If you're seeking a one-click installation method for LivePortrait, an open-source zero-shot image-to-animation application on Windows, for local use, this tutorial is essential. It introduces the cutting-edge image-to-animation open-source generator Live Portrait. Simply provide a static image and a driving video to create an impressive animation in seconds. LivePortrait is incredibly fast and adept at preserving facial expressions from the input video. The results are truly astonishing.
With the V3 update adding video-to-video functionality, those interested in using LivePortrait but lacking a powerful GPU, using a Mac, or preferring cloud-based solutions will find this tutorial invaluable. It guides you through the one-click installation and usage of LivePortrait on #MassedCompute, #RunPod, and even a free #Kaggle account. After following this tutorial, you'll find running LivePortrait on cloud services as straightforward as running it locally. LivePortrait is the latest state-of-the-art static image to talking animation generator, surpassing even paid services in both speed and quality.
- 2 replies
liked
3
models
about 1 year ago
Unable to load model in SwitChat example
6
#7 opened over 1 year ago
by
ltouati
liked
a
Space
about 1 year ago