You can deploy this model and start earning money today!

Qwen-4B-DB-AlfWorld-v17

Name: Qwen-4B-DB-AlfWorld-v17
Availability: InStock

This repository provides a merged model fine-tuned from Qwen/Qwen3-4B-Instruct-2507.

All LoRA adapter weights have been merged into the base model. The resulting checkpoint is fully standalone and does not require external adapter loading.

Dataset Processing Details

ALFWorld (v4 & v5)

The two ALFWorld trajectory datasets were merged and filtered.

Filtering Strategy

Admissible Action Requirement

Only samples containing at least one user turn with: admissible actions:

were retained.

Maximum Token Length Constraint

Token length was computed using the base model tokenizer.
Samples exceeding:

2048 - 148 = 1900 tokens

were discarded.

No Structural Modification

No task-type prefix was inserted.
Original ReAct-style trajectories were preserved.

DBBench (v2 & v3)

The two DBBench datasets were merged and rebalanced by task type.

Rebalancing Strategy

Fully retained:
aggregation*
comparison*
counting*
ranking*
Capped to fixed limits:
INSERT (max 200)
UPDATE (max 250)
INSERT_error_recovery (max 60)
UPDATE_error_recovery (max 40)
Other task types:
Randomly sampled up to a total of 120 examples.

After sampling, the dataset was globally shuffled.

No explicit task-type prefix (e.g., "Task Type: DATABASE") was inserted. Original ReAct-style conversational structure was preserved.

Training Objective

The model is optimized for multi-turn reasoning and action-based tasks.

Loss is applied to all assistant turns in multi-turn trajectories, allowing the model to learn:

Observation interpretation
Step-by-step reasoning
Action execution
Tool usage
Error recovery behavior

Training Configuration

Base model: Qwen/Qwen3-4B-Instruct-2507
Method: LoRA (merged into final weights)
Max sequence length: 2048
Epochs: 4
Learning rate: 1e-05
LoRA parameters: r=32, alpha=64

Usage Example

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "Umiharu/Qwen-4B-DB-AlfWorld-v17"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
 model_id,
 torch_dtype=torch.float16,
 device_map="auto",
)

prompt = "You are a task-solving agent. Respond 'OK' if you are ready."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(
 **inputs,
 max_new_tokens=64,
 temperature=0.2,
 do_sample=False,
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

## Sources & Terms (IMPORTANT)

Training data: u-10bei/sft_alfworld_trajectory_dataset_v4, u-10bei/sft_alfworld_trajectory_dataset_v5, u-10bei/dbbench_sft_dataset_react_v2, u-10bei/dbbench_sft_dataset_react_v3

Dataset License: MIT License. This dataset is used and distributed under the terms of the MIT License.
Compliance: Users must comply with the MIT license (including copyright notice) and the base model's original terms of use.