cmedina/agenx-lora-training
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: rename 0-indexed saved files to 1-indexed

Browse Source
This commit is contained in:
Christian Medina
2026-07-03 00:16:15 -04:00
parent b4538c013b
commit 23dcdd2c6f
1 changed files with 13 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
quantize_streaming.py
View File

@@ -30,12 +30,23 @@ def streaming_quantize(model_path: str, output_path: str):
output_path = Path(output_path) output_path = Path(output_path)
output_path.mkdir(parents=True, exist_ok=True) output_path.mkdir(parents=True, exist_ok=True)
config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
import glob import glob
shards = sorted(glob.glob(f"{model_path}/*.safetensors")) shards = sorted(glob.glob(f"{model_path}/*.safetensors"))
# Rename existing 0-indexed files to 1-indexed
for existing in output_path.glob("*.safetensors"):
parts = existing.name.split("-")
if len(parts) >= 3 and existing.name.startswith("model-00000-"):
num = int(parts[1])
new_name = f"model-{num+1:05d}-of-{len(shards):05d}.safetensors"
new_path = output_path / new_name
existing.rename(new_path)
print(f"Renamed: {existing.name} -> {new_name}")
print(f"Found {len(shards)} shards\n") print(f"Found {len(shards)} shards\n")
config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
# Process 4 shards per GPU (8 total) for max parallelism # Process 4 shards per GPU (8 total) for max parallelism
shards_per_gpu = 4 shards_per_gpu = 4
num_gpus = 2 num_gpus = 2