NyxKrage/Microsoft_Phi-4
Updated
•
2.43k
•
42
Joseph
Joseph717171
AI & ML interests
None yet
Recent Activity
liked
a model
about 21 hours ago
NyxKrage/Microsoft_Phi-4
upvoted
a
paper
about 21 hours ago
Deliberation in Latent Space via Differentiable Cache Augmentation
liked
a model
3 days ago
black-forest-labs/FLUX.1-schnell
Organizations
Joseph717171's activity
liked
a
model
about 21 hours ago
upvoted
a
paper
about 21 hours ago
liked
2
models
3 days ago
reacted to
singhsidhukuldeep's
post with 🚀🧠
3 days ago
Post
3532
Exciting breakthrough in AI:
@Meta
's new Byte Latent Transformer (BLT) revolutionizes language models by eliminating tokenization!
The BLT architecture introduces a groundbreaking approach that processes raw bytes instead of tokens, achieving state-of-the-art performance while being more efficient and robust. Here's what makes it special:
>> Key Innovations
Dynamic Patching: BLT groups bytes into variable-sized patches based on entropy, allocating more compute power where the data is more complex. This results in up to 50% fewer FLOPs during inference compared to traditional token-based models.
Three-Component Architecture:
• Lightweight Local Encoder that converts bytes to patch representations
• Powerful Global Latent Transformer that processes patches
• Local Decoder that converts patches back to bytes
>> Technical Advantages
• Matches performance of Llama 3 at 8B parameters while being more efficient
• Superior handling of non-English languages and rare character sequences
• Remarkable 99.9% accuracy on spelling tasks
• Better scaling properties than token-based models
>> Under the Hood
The system uses an entropy model to determine patch boundaries, cross-attention mechanisms for information flow, and hash n-gram embeddings for improved representation. The architecture allows simultaneous scaling of both patch and model size while maintaining fixed inference costs.
This is a game-changer for multilingual AI and could reshape how we build future language models. Excited to see how this technology evolves!
The BLT architecture introduces a groundbreaking approach that processes raw bytes instead of tokens, achieving state-of-the-art performance while being more efficient and robust. Here's what makes it special:
>> Key Innovations
Dynamic Patching: BLT groups bytes into variable-sized patches based on entropy, allocating more compute power where the data is more complex. This results in up to 50% fewer FLOPs during inference compared to traditional token-based models.
Three-Component Architecture:
• Lightweight Local Encoder that converts bytes to patch representations
• Powerful Global Latent Transformer that processes patches
• Local Decoder that converts patches back to bytes
>> Technical Advantages
• Matches performance of Llama 3 at 8B parameters while being more efficient
• Superior handling of non-English languages and rare character sequences
• Remarkable 99.9% accuracy on spelling tasks
• Better scaling properties than token-based models
>> Under the Hood
The system uses an entropy model to determine patch boundaries, cross-attention mechanisms for information flow, and hash n-gram embeddings for improved representation. The architecture allows simultaneous scaling of both patch and model size while maintaining fixed inference costs.
This is a game-changer for multilingual AI and could reshape how we build future language models. Excited to see how this technology evolves!
- 2 replies
reacted to
davanstrien's
post with 🔥
4 days ago
Post
1530
Introducing FineWeb-C 🌐🎓, a community-built dataset for improving language models in ALL languages.
Inspired by FineWeb-Edu the community is labelling the educational quality of texts for many languages.
318 annotators, 32K+ annotations, 12 languages - and growing! 🌍
data-is-better-together/fineweb-c
Inspired by FineWeb-Edu the community is labelling the educational quality of texts for many languages.
318 annotators, 32K+ annotations, 12 languages - and growing! 🌍
data-is-better-together/fineweb-c
upvoted
a
paper
5 days ago
reacted to
FranckAbgrall's
post with 🤗🔥
5 days ago
Post
1046
🆕 It should now be easier to identify discussions or pull requests where repository owners are participating on HF, let us know it that helps 💬🤗
- 1 reply
reacted to
m-ric's
post with 🧠👀👍
5 days ago
Post
2029
𝐇𝐮𝐠𝐠𝐢𝐧𝐠 𝐅𝐚𝐜𝐞 𝐫𝐞𝐥𝐞𝐚𝐬𝐞𝐬 𝐏𝐢𝐜𝐨𝐭𝐫𝐨𝐧, 𝐚 𝐦𝐢𝐜𝐫𝐨𝐬𝐜𝐨𝐩𝐢𝐜 𝐥𝐢𝐛 𝐭𝐡𝐚𝐭 𝐬𝐨𝐥𝐯𝐞𝐬 𝐋𝐋𝐌 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝟒𝐃 𝐩𝐚𝐫𝐚𝐥𝐥𝐞𝐥𝐢𝐳𝐚𝐭𝐢𝐨𝐧 🥳
🕰️ Llama-3.1-405B took 39 million GPU-hours to train, i.e. about 4.5 thousand years.
👴🏻 If they had needed all this time, we would have GPU stories from the time of Pharaoh 𓂀: "Alas, Lord of Two Lands, the shipment of counting-stones arriving from Cathay was lost to pirates, this shall delay the building of your computing temple by many moons "
🛠️ But instead, they just parallelized the training on 24k H100s, which made it take just a few months.
This required parallelizing across 4 dimensions: data, tensor, context, pipeline.
And it is infamously hard to do, making for bloated code repos that hold together only by magic.
🤏 𝗕𝘂𝘁 𝗻𝗼𝘄 𝘄𝗲 𝗱𝗼𝗻'𝘁 𝗻𝗲𝗲𝗱 𝗵𝘂𝗴𝗲 𝗿𝗲𝗽𝗼𝘀 𝗮𝗻𝘆𝗺𝗼𝗿𝗲! Instead of building mega-training codes, Hugging Face colleagues cooked in the other direction, towards tiny 4D parallelism libs. A team has built Nanotron, already widely used in industry.
And now a team releases Picotron, a radical approach to code 4D Parallelism in just a few hundred lines of code, a real engineering prowess, making it much easier to understand what's actually happening!
⚡ 𝗜𝘁'𝘀 𝘁𝗶𝗻𝘆, 𝘆𝗲𝘁 𝗽𝗼𝘄𝗲𝗿𝗳𝘂𝗹:
Counting in MFU (Model FLOPs Utilization, how much the model actually uses all the compute potential), this lib reaches ~50% on SmolLM-1.7B model with 8 H100 GPUs, which is really close to what huge libs would reach. (Caution: the team is leading further benchmarks to verify this)
Go take a look 👉 https://github.com/huggingface/picotron/tree/main/picotron
🕰️ Llama-3.1-405B took 39 million GPU-hours to train, i.e. about 4.5 thousand years.
👴🏻 If they had needed all this time, we would have GPU stories from the time of Pharaoh 𓂀: "Alas, Lord of Two Lands, the shipment of counting-stones arriving from Cathay was lost to pirates, this shall delay the building of your computing temple by many moons "
🛠️ But instead, they just parallelized the training on 24k H100s, which made it take just a few months.
This required parallelizing across 4 dimensions: data, tensor, context, pipeline.
And it is infamously hard to do, making for bloated code repos that hold together only by magic.
🤏 𝗕𝘂𝘁 𝗻𝗼𝘄 𝘄𝗲 𝗱𝗼𝗻'𝘁 𝗻𝗲𝗲𝗱 𝗵𝘂𝗴𝗲 𝗿𝗲𝗽𝗼𝘀 𝗮𝗻𝘆𝗺𝗼𝗿𝗲! Instead of building mega-training codes, Hugging Face colleagues cooked in the other direction, towards tiny 4D parallelism libs. A team has built Nanotron, already widely used in industry.
And now a team releases Picotron, a radical approach to code 4D Parallelism in just a few hundred lines of code, a real engineering prowess, making it much easier to understand what's actually happening!
⚡ 𝗜𝘁'𝘀 𝘁𝗶𝗻𝘆, 𝘆𝗲𝘁 𝗽𝗼𝘄𝗲𝗿𝗳𝘂𝗹:
Counting in MFU (Model FLOPs Utilization, how much the model actually uses all the compute potential), this lib reaches ~50% on SmolLM-1.7B model with 8 H100 GPUs, which is really close to what huge libs would reach. (Caution: the team is leading further benchmarks to verify this)
Go take a look 👉 https://github.com/huggingface/picotron/tree/main/picotron
- 1 reply
updated
2
models
7 days ago
New activity in
Joseph717171/Llama-3.1-SuperNova-Lite-8.0B-OQ8_0-F32.EF32.IQ4_K-Q8_0-GGUF
7 days ago
Good for summarization.
6
#1 opened 25 days ago
by
ThiloteE
liked
a
model
7 days ago
updated
a
model
8 days ago
liked
2
models
8 days ago
