Differential Transformer V2
• 51
AI & ML interests
None defined yet.
Recent Activity
Papers
Covering Human Action Space for Computer Use: Data Synthesis and Benchmark
Synthetic Computers at Scale for Long-Horizon Productivity Simulation
Articles
alvarobartt
posted an update 1 day ago
Post
121
Open agents on AWS SageMaker AI with open models from the Hugging Face Hub!
> Deploy an open model from the Hugging Face Hub on SageMaker AI
> Connect the deployed model to Strands Agents
> Add built-in and custom tools for tool calling
> Expose external capabilities through MCP integration
> Bonus: talk to your agent and visualize traces with Gradio
https://alvarobartt.com/agents-on-aws-sagemaker
> Deploy an open model from the Hugging Face Hub on SageMaker AI
> Connect the deployed model to Strands Agents
> Add built-in and custom tools for tool calling
> Expose external capabilities through MCP integration
> Bonus: talk to your agent and visualize traces with Gradio
https://alvarobartt.com/agents-on-aws-sagemaker
DongChenMSRA
updated 3
models 2 days ago
ruichunma
updated a
dataset 3 days ago
ruichunma
published a
dataset 3 days ago
ruichunma
published a
model 3 days ago
JeffreyXiang
updated a
Space 4 days ago
alvarobartt
posted an update 5 days ago
Post
3202
Latest
TL; DR MoEs can be misleading to reason about from active parameters alone, since each token only activates a subset of experts, while the serving setup still needs to account for the full resident memory footprint.
🧠
🏗️ Dense models usually load and use most weights every forward pass, while MoEs load many experts but only route each token to a few of them
⚡ Active params isn't the same as memory footprint, especially for sparse architectures
📦 Runtime memory is about what is used per request/token, while loading memory also includes the expert weights that need to be resident
📚 KV cache can still dominate depending on context length, batch size, and concurrency
🔀 Expert Parallelism (EP) helps shard experts across accelerators when expert weights dominate
🚀 Data Parallelism (DP) + EP is often a good fit for throughput-oriented MoE serving
Check the repository at https://github.com/alvarobartt/hf-mem
hf-mem release added a breakdown of Mixture-of-Experts (MoE) memory usage!TL; DR MoEs can be misleading to reason about from active parameters alone, since each token only activates a subset of experts, while the serving setup still needs to account for the full resident memory footprint.
🧠
hf-mem now splits MoE memory into base model weights, routed experts, and KV cache🏗️ Dense models usually load and use most weights every forward pass, while MoEs load many experts but only route each token to a few of them
⚡ Active params isn't the same as memory footprint, especially for sparse architectures
📦 Runtime memory is about what is used per request/token, while loading memory also includes the expert weights that need to be resident
📚 KV cache can still dominate depending on context length, batch size, and concurrency
🔀 Expert Parallelism (EP) helps shard experts across accelerators when expert weights dominate
🚀 Data Parallelism (DP) + EP is often a good fit for throughput-oriented MoE serving
Check the repository at https://github.com/alvarobartt/hf-mem
Reproducing CT-RATE retrieval numbers
2
#6 opened 6 days ago
by 
gings
jmz-msft
updated a
model 8 days ago
DongChenMSRA
published 3
models 9 days ago
qianhuiwu
submitted a
paper to Daily Papers 9 days ago
Baolin
published a
dataset 9 days ago
Yif29
authored a
paper 10 days ago
ruichunma
updated a
model 10 days ago
Miaosen
submitted a
paper to Daily Papers 10 days ago
tvspina
updated a
collection 10 days ago