Modelpublic

ChesterProgrammer/Qwen3.5-9B-ConsistentChat-100steps

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face

0likes4downloads

Model Card

Uploaded finetuned model

—Developed by: ChesterProgrammer
—License: apache-2.0
—Finetuned from model : unsloth/Qwen3.5-9B

This qwen3_5 model was trained 2x faster with Unsloth and Huggingface's TRL library.

Hyperparams

Method Lora(16 bit) Epochs 0 Batch size 8 Grad Accum 2 Learning rate 0.0002 Optimizer AdamW 8-bit Max steps 100 Context length 2048 Warmup steps 4 Packing False weight decay 0.001 seed 3407

LoRA Rank 16 Alpha 32 Dropout 0 Variant lora

Dataset jiawei-ucas/ConsistentChat

Elapsed: 22m 51s 0.07 steps/s Tokens: 1572400