|
--- |
|
frameworks: |
|
- Pytorch |
|
license: apache-2.0 |
|
tasks: |
|
- text-generation |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
--- |
|
Fine-tuning the llama3-8b-instruct model using the [msagent-pro](https://modelscope.cn/datasets/iic/MSAgent-Pro/summary) dataset and the loss_scale technique with [swift](https://github.com/modelscope/swift), the script is as follows: |
|
```bash |
|
NPROC_PER_NODE=8 \ |
|
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \ |
|
MASTER_PORT=29500 \ |
|
swift sft \ |
|
--model_type llama3-8b-instruct \ |
|
--learning_rate 2e-5 \ |
|
--sft_type lora \ |
|
--dataset msagent-pro \ |
|
--gradient_checkpointing true \ |
|
--gradient_accumulation_steps 8 \ |
|
--deepspeed default-zero3 \ |
|
--lora_target_modules ALL \ |
|
--use_loss_scale true \ |
|
--save_strategy epoch \ |
|
--batch_size 1 \ |
|
--num_train_epochs 2 \ |
|
--max_length 4096 \ |
|
--preprocess_num_proc 4 \ |
|
--use_loss_scale true \ |
|
--loss_scale_config_path agent-flan \ |
|
--ddp_backend nccl \ |
|
``` |
|
|
|
Comparison with the Original Model on the ToolBench Evaluation Set |
|
|
|
| Model | ToolBench (in-domain) | | | | | ToolBench (out-of-domain) | | | | |
|
|-------------------------|----------------------------------------------|-------|-------|-------|-------|--------------------------------------------|-------|-------|-------| |
|
| | Plan.EM | Act.EM| HalluRate (lower is better) | Avg.F1 | R-L | Plan.EM | Act.EM| HalluRate (lower is better) | Avg.F1 | R-L | |
|
| llama3-8b-instruct | 74.22 | 36.17 | 15.68 | 20.0 | 12.14 | 69.47 | 34.21 | 14.72 | 20.25 | 14.07 | |
|
| llama3-8b-agent-instruct-v2 | **85.15** | **58.1** | **1.57** | **52.10** | **26.02** | **85.79** | **59.43** | **2.56** | **52.19** | **31.43** | |
|
|
|
For detailed explanations of the evaluation metrics, please refer to [document](https://github.com/modelscope/eval-scope/tree/main/llmuses/third_party/toolbench_static) |
|
|
|
Deploy this model: |
|
```shell |
|
USE_HF=True swift deploy \ |
|
--model_id_or_path modelscope/llama3-8b-agent-instruct-v2 \ |
|
--model_type llama3-8b-instruct \ |
|
--infer_backend vllm \ |
|
--tools_prompt toolbench |
|
``` |