faxenoff/code-daemon-summary-v1
replace with the locally built v1: 3 added/updated, 2 superseded removed
refresh the card and upload the metadata files that no publish run sent
Card: document the imatrix-rebuilt Q3_K_M and why Q4/Q5 are unchanged
Q3_K_M: quantized direct from F16 with a production-calibrated imatrix (was a requant of Q5_K_M). Held-out domain perplexity 2.3256 vs 2.3401, paired t=+3.8 over 31 chunks. Q4/Q5 unchanged - measured no-op at those bit widths.
docs: measured throughput + memory footprint
Upload README.md with huggingface_hub
Upload code-daemon-summary-v1-Q3_K_M.gguf with huggingface_hub
Upload code-daemon-summary-v1-Q5_K_M.gguf with huggingface_hub
Upload code-daemon-summary-v1-Q4_K_M.gguf with huggingface_hub
Upload README.md with huggingface_hub
initial commit
