biondizzle/deepseek-v4-quant
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix: use full argparse flag names (--calib_size, --kv_cache_qformat)

Browse Source
This commit is contained in:
biondizzle
2026-05-09 14:54:51 +00:00
parent b2849a8944
commit 1a36a655ea
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
scripts/quantize_nvfp4.py
View File

@@ -214,9 +214,9 @@ def run_calibration(model_path, export_dir, calib_save_path, amax_snapshot_path,
"hf_ptq.py",
"--model", model_path,
"--quant", QUANT,
"--calib", str(calib_size),
"--calib_size", str(calib_size),
"--calib_seq", str(calib_seq),
"--kv_cache_quant", KV_CACHE_QUANT,
"--kv_cache_qformat", KV_CACHE_QUANT,
"--tp", str(TP),
"--export_path", export_dir,
"--trust_remote_code",