This model is trained to improve DB Bench (database operation) performance while preserving the base model's ALFWorld capability without ALFWorld-specific training data.

Loss is applied to all assistant turns in the multi-turn trajectory, enabling the model to learn SQL generation, action selection, and error recovery.

Training Data

DB Bench v4 (u-10bei/dbbench_sft_dataset_react_v4): 1,200 samples
Distilled data (weak category reinforcement via GPT-OSS-120B): 440 samples
Total: 1,640 samples
ALFWorld data intentionally excluded to preserve base model performance

Training Configuration

Base model: Qwen/Qwen2.5-7B-Instruct
Method: LoRA → merged to 16bit
Max sequence length: 2048
Epochs: 3
Learning rate: 2e-6
LoRA: r=64, alpha=128
Optimizer: AdamW (cosine scheduler)
Framework: Unsloth 2025.12.7

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "koguma-ai/dbbench-distill-qwen25-7b-0302"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

Limitations

Optimized for DB Bench tasks; ALFWorld performance relies on base model capability
Distilled data targets weak categories (MAX, comparison, INSERT, complex queries)

koguma-ai/dbbench-distill-qwen25-7b-0302-07

Qwen2.5-7B DB Bench Focused SFT + Distillation

Training Objective

Training Data

Training Configuration

Usage

Limitations

koguma-ai/
dbbench-distill-qwen25-7b-0302-07