qwen25_7b_lora_agentbench_v20
This repository provides a merged model fine-tuned from
ShogoMu/qwen25_7b_lora_agentbench_v11. The fine-tuning was performed using LoRA + Unsloth and the resulting adapter has been merged back into the base model weights.
This repository contains full model weights, making it ready for inference
without the need to load a separate adapter.
Training Objective
This model is optimized for multi-turn agent tasks, specifically for
ALFWorld (household navigation/interaction) and DBBench (database operations).
The training process applied loss to all assistant turns in the multi-turn
trajectories, allowing the model to learn not just final answers, but also
intermediate reasoning (Thought), environment observation processing,
action selection, and error recovery.
Training Configuration
- Base model: ShogoMu/qwen25_7b_lora_agentbench_v11
- Method: LoRA (merged post-training)
- Max sequence length: 2048
- Epochs: 2
- Learning rate: 2e-06
- LoRA Parameters: r=64, alpha=128
Dataset Sampling
A subset was used for each dataset (format: id:N = up to N samples, id:N% = up to N% of dataset):
- u-10bei/sft_alfworld_trajectory_dataset_v5: 500
Usage
This model can be loaded using the standard transformers library or
deployed with vLLM (recommended for evaluation).
Transformers
1from transformers import AutoModelForCausalLM, AutoTokenizer
2import torch
3
4model_id = "your_hf_id/your_repo_name"
5
6tokenizer = AutoTokenizer.from_pretrained(model_id)
7model = AutoModelForCausalLM.from_pretrained(
8 model_id,
9 torch_dtype=torch.bfloat16,
10 device_map="auto",
11)
