biMEMO/Qwen3.8-27B-int4-AutoRound

Update README.md

Remove internal measurement/network detail from benchmark methodology note

Update README.md

Widen benchmark table with descriptive headers and full unit names (real markdown padding does not affect HF rendering)

Widen and center-align benchmark table columns for readability

Move context-limited note to footnote for table readability

Correct benchmark numbers: measure on-host to eliminate client network overhead, fix 1-GPU gpu-memory-utilization for vLLM 0.27.1, reorder Quick Start 1/2/4 GPU

Update README.md

Update README.md

Add biMEMO Discord link

Update README.md

Add per-GPU-count Quick Start commands (1/2/4 GPUs), all benchmarked configs

Add MTP quantization approach section with A/B comparison vs full-BF16 MTP block

Upload folder using huggingface_hub

initial commit