Article
How to Run a Hugging Face Model in JAX (Part 3)
By
•
•
6Han
qihqi
AI & ML interests
None yet
Recent Activity
upvoted
an
article
14 days ago
How to Run a Hugging Face Model in JAX (Part 3)
commented on
their
article
14 days ago
How to Run a Hugging Face Model in JAX (Part 1)
commented on
their
article
14 days ago
How to Run a Hugging Face Model in JAX (Part 2)
Organizations
upvoted
an
article
14 days ago
commented on
How to Run a Hugging Face Model in JAX (Part 1)
14 days ago
https://huggingface.co/blog/qihqi/huggingface-jax-03 using StaticCache for sequence generation. Thanks for the suggestion!
commented on
How to Run a Hugging Face Model in JAX (Part 2)
14 days ago
published
an
article
14 days ago
Article
How to Run a Hugging Face Model in JAX (Part 3)
By
•
•
6
reacted to
pcuenq's
post with 🔥
24 days ago
Post
8657
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!
- 5 replies
commented on
How to Run a Hugging Face Model in JAX (Part 1)
26 days ago
Amazing work! Loved the trick to register the custom types as Pytrees.
If I understand it correctly, the KV cache will not be in use in the jitted code. Did you explore the use of a fixed-size
StaticCacheinstead, or will that be part of your sequence generation followup?
Hi Pedro,
Yes that is a good idea. I didn't know about StaticCache let me explore that. Thanks!
commented on
How to Run a Hugging Face Model in JAX (Part 1)
28 days ago
upvoted
an
article
28 days ago
Article
How to Run a Hugging Face Model in JAX (Part 2)
By
•
•
5
published
an
article
28 days ago
Article
How to Run a Hugging Face Model in JAX (Part 2)
By
•
•
5
upvoted
an
article
about 1 month ago
Article
How to Run a Hugging Face Model in JAX (Part 1)
By
•
•
22
published
an
article
about 1 month ago
Article
How to Run a Hugging Face Model in JAX (Part 1)
By
•
•
22