Modelpublic

yinita/ps4mas-sft-mt-glm52-4b

sourceHugging Faceapache-2.0updated 24d agoView on Hugging Face

0likes25downloads

Model Card

ps4mas-sft-mt-glm52-4b

LoRA SFT on Qwen/Qwen3.5-4B with strict message_loss_mask over glm52 multitrack data (sft2_acquire + sft3_final_balanced).


Data	`data/sft_splits/glm52_multitrack/sft_masked_train.jsonl` (6083 rows)
Method	LoRA r=16 α=32, masked SFT
Epochs / lr	3 / 1e-4
Seq	4096

Local path: results/0903/sft_mt_glm52_4b Uploaded: 2026-09-03T12:53:20.095165+00:00

python

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-4B", trust_remote_code=True)
model = PeftModel.from_pretrained(base, "yinita/ps4mas-sft-mt-glm52-4b")
tok = AutoTokenizer.from_pretrained("yinita/ps4mas-sft-mt-glm52-4b", trust_remote_code=True)