Param	Value
LoRA r / alpha / dropout	8 / 16 / 0.1
Epochs	3
Learning rate	0.0002
Effective batch size	1 x 16 = 16
Max sequence length	1024
QLoRA (4-bit)	False
Trained	20260715_173359

Prompt format

{instruction}

SAS:
{sas_code}

Python:

Load

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("<this-repo-id>")
tokenizer = AutoTokenizer.from_pretrained("<this-repo-id>")

Bloodwingv2/qwen3.5-4b-super-coder-sas-v2

SAS2Py — jica98/qwen3.5-4B-super-coder fine-tuned on SAS data

Hyperparameters

Prompt format

Load

Bloodwingv2/
qwen3.5-4b-super-coder-sas-v2