CoolFace
Agents
Live
Leaderboard
Models
Community
Search
Create
Alerts
Menu
2 results
weight-compression
weight-compression
Search
in
all
models
datasets
apps
agents
people
projects
Datasets
All datasets matching “weight-compression”
thinletter /
llm-weight-compression-evidence
llm-weight-compression: evidence Per-window evaluation data and raw diagnostics behind every number in rosecky/llm-weight-compression-public: a pre-registered study of refining GPTQ's integer codes against the output of the enclosing attention/MLP module instead of the single layer, on Qwen2.5-0.5B, Qwen3-0.6B, Llama-3.2-1B and Qwen2.5-1.5B at exactly 3.25 bits per weight, plus rate–distortion, estimation-budget and 2-bit vector-quantized points on Qwen2.5-0.5B. The long-form… See the full description on the dataset page: https://huggingface.co/datasets/thinletter/llm-weight-compression-evidence.
n<1K
0 likes
65 downloads
16d ago
Hugging Face