kel-dx/llm-jp-4-8b-instruct-NVFP4

Document DGX Spark and DFlash benchmark results

Improved calibration (strict-format slice), modelopt 0.45, vLLM 0.24 validation, plugin fixes (adjust_request scoping, extract_content_ids, indentation), full reproducibility bundle (pinned revision, env freeze, calibration file), LICENSE

note SM121 validation status

fix dangling doc reference

proxy: neutral default upstream

remove egg-info artifacts

remove build artifacts

llm-jp-4-8b-instruct NVFP4 (Japanese-calibrated, mixed-precision recipe): weights, model card, vLLM plugin, full reproduction recipe

initial commit