PollardWeights/Qwen2.5-7B-Instruct-Pollard
card: tok/s per rung, published byte sizes, corrected runtime column, stock-rebuild comparison
add the Calib-3.0 importance matrix the allocation was measured with
Rebuild IQ3_S on stock atoms so it loads in stock llama.cpp
Rebuild IQ4_XS on stock atoms so it loads in stock llama.cpp
State which rungs need ik_llama.cpp, read from the files' tensor types
Regenerate on the Pollard master card template
Upload README.md with huggingface_hub
Upload Qwen2.5-7B-Instruct-Pollard-IQ1_KT.gguf with huggingface_hub
Upload Qwen2.5-7B-Instruct-Pollard-Q6_K.gguf with huggingface_hub
Upload Qwen2.5-7B-Instruct-Pollard-IQ4_XS.gguf with huggingface_hub
Upload Qwen2.5-7B-Instruct-Pollard-IQ3_S.gguf with huggingface_hub
Upload README.md with huggingface_hub
initial commit