This repository provides a LoRA adapter fine-tuned from kuririrn/qwen3-4b-agent-trajectory_alf_admissible-lora-constraint_gen-dist_allign using LoRA + Unsloth.

This repository contains LoRA adapter weights only. The base model must be loaded separately.

Datasets

This adapter is trained on a single trajectory-style SFT dataset:

DBBench weak-type trajectories: kuririrn/dbbench_sft_dataset_react_v3v4_weaktypes
- Subset of DBBench SFT data focusing on weak-type categories such as counting, comparison, aggregation-SUM, aggregation-MAX, INSERT.
- These correspond to low-accuracy regimes in the baseline agent and are targeted to improve DBBench performance.

Training Objective

This adapter is trained to improve multi-turn agent task performance on DBBench (database operations), especially weak-type categories.

Loss is applied to all assistant turns so the model learns tool calls, SQL reasoning, and error recovery for DBBench environments.

Training Configuration

Base model: kuririrn/qwen3-4b-agent-trajectory_alf_admissible-lora-constraint_gen-dist_allign
Method: LoRA (full precision base)
Max sequence length: 2048
Epochs: 1
Learning rate: 5e-07
LoRA r: 64
LoRA alpha: 128

Intended Use

This adapter is intended to be used as a tool-using agent for ALFWorld and/or DBBench-style tasks, depending on the training configuration:

When trained only on ALFWorld, it focuses on household tasks.
When trained only on DBBench, it focuses on database operations.
When trained on both, it aims to balance both domains.

How to Load

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained(
    "kuririrn/qwen3-4b-agent-trajectory_alf_admissible-lora-constraint_gen-dist_allign",
    torch_dtype="auto",
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("kuririrn/qwen3-4b-agent-trajectory_alf_admissible-lora-constraint_gen-dist_allign")

model = PeftModel.from_pretrained(
    base_model,
    "/content/lora_agentbench_qwen3_4b",
)

model.eval()

Limitations and Bias

This adapter is trained on synthetic and task-specific trajectories for ALFWorld and/or DBBench. It may:

Overfit to the style of prompts and environments in these datasets.
Hallucinate actions or SQL queries outside the training distribution.
Fail on tasks that are very different from ALFWorld/DBBench.

Always validate the model's outputs, especially when using it for real-world decision making or database operations.

kuririrn/qwen3-4b-agent-trajectory_2stageSFT_alfadm_dbweek

AgentBench LoRA Adapter (kuririrn/qwen3-4b-agent-trajectory_alf_admissible-lora-constraint_gen-dist_allign)