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

fix: use NF4 naming convention for quantized model

Browse Source
This commit is contained in:
Christian Medina
2026-07-02 22:00:15 -04:00
parent 2f997377f5
commit 52b93837f9
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
quantize_model.py
View File

@@ -41,7 +41,7 @@ if __name__ == "__main__":
import argparse import argparse
parser = argparse.ArgumentParser() parser = argparse.ArgumentParser()
parser.add_argument("--model-path", type=str, default="/data/models/Ornith-1.0-35B") parser.add_argument("--model-path", type=str, default="/data/models/Ornith-1.0-35B")
parser.add_argument("--output-path", type=str, default="/data/models/Ornith-1.0-35B-4bit-nf4") parser.add_argument("--output-path", type=str, default="/data/models/Ornith-1.0-35B-nf4")
args = parser.parse_args() args = parser.parse_args()
quantize_model(args.model_path, args.output_path) quantize_model(args.model_path, args.output_path)

2
training/configs/ornith-35b-lora.yaml
View File

@@ -1,7 +1,7 @@
# LoRA Training Configuration for Ornith-1.0-35B # LoRA Training Configuration for Ornith-1.0-35B
# Dataset: cyron_summary_lora_dataset (20k examples) # Dataset: cyron_summary_lora_dataset (20k examples)
base_model: /data/models/Ornith-1.0-35B base_model: /data/models/Ornith-1.0-35B-nf4
model_type: Qwen3_5MoeForCausalLM model_type: Qwen3_5MoeForCausalLM
tokenizer_type: AutoTokenizer tokenizer_type: AutoTokenizer