RTAQ checkpoints: routed experts at about 2 bits, stored as BF16 tensors.
MinJae Choi
minjaechoi
webbrain-one-257412's profile picture
webbrain-one-9586305a's profile picture
1000luca's profile picture·
AI & ML interests
Efficient ML, Quantization, On-Device AI, Hardware-aware Algorithm Design
Recent Activity
updated a collection 2 days ago
RTAQ: Reasoning Termination-Aware Quantization for MoE updated a model 2 days ago
minjaechoi/qwen3p6-35b-a3b-2p02bit-r33-lowbit updated a collection 2 days ago
RTAQ: Reasoning Termination-Aware Quantization for MoEOrganizations
None yet
models 90
minjaechoi/qwen3p6-35b-a3b-2p02bit-r33-lowbit
Text Generation • Updated • 15
minjaechoi/qwen3next-80b-a3b-2p02bit-r22-lowbit
Text Generation • Updated • 14
minjaechoi/nemotron3-nano-30b-a3b-2p03bit-r22-lowbit
Text Generation • Updated • 13
minjaechoi/qwen3p6-35b-a3b-levelbank-asym-norot-k15-r83
Updated
minjaechoi/qwen3p6-35b-a3b-2p03bit-r82
Text Generation • 35B • Updated • 56
minjaechoi/qwen3p6-35b-a3b-levelbank-sym-r81
Updated
minjaechoi/qwen3p6-35b-a3b-levelbank-bipea-asym-dp-r80
Updated
minjaechoi/qwen3p6-35b-a3b-fisher-anchor-rebuilt-partial-r79
Updated
minjaechoi/qwen3p6-35b-a3b-levelbank-asym-r78
Updated
minjaechoi/qwen3p6-35b-a3b-2p03bit-r76
Text Generation • 35B • Updated • 28