模型微调命令预览

开发中

POST

/front/modelFineTune/generateCommandView

请求参数

Header 参数

Body 参数application/json

示例

{
    "templateId": "b21db781-7c0d-455e-af4f-3da2d926b739",
    "estimatedTime": 946,
    "maxWaitTime": 0,
    "minWaitTime": 0,
    "gpuCount": 1,
    "paramModel": 0,
    "predTotalTimeSec": 946,
    "taskPrice": 10,
    "realUnitPrice": 30.72,
    "appMode": 1,
    "maxScheduleCount": 0,
    "estimatedPrice": 11.308,
    "optTarget": "gpu_use_rate",
    "configJson": {
        "stage": "Supervised Fine-Tuning",
        "finetuning_type": "lora",
        "val_size": 0,
        "merge_model": false,
        "task_description": "",
        "learning_rate": 0.00005,
        "num_train_epochs": 3,
        "per_device_train_batch_size": 2,
        "gradient_accumulation_steps": 8,
        "global_batch_size": 16,
        "warmup_ratio": 0.1,
        "lora_rank": 8,
        "lora_alpha": 16,
        "lr_scheduler_type": "cosine",
        "max_grad_norm": 1,
        "compute_type": "bf16",
        "cutoff_len": 1024,
        "model_name_or_path": "DeepSeek-R1-0528-Qwen3-8B",
        "publicData": [
            "alpaca_zh_demo"
        ],
        "fileData": []
    },
    "configValueAsCmdKeys": []
}

请求示例代码

Shell

JavaScript

Java

Swift

PHP

Python

HTTP

Objective-C

Ruby

OCaml

Dart

curl --location --request POST '/front/modelFineTune/generateCommandView' \
--header 'FrontToken: {{FrontToken}}' \
--header 'Content-Type: application/json' \
--data-raw '{
    "templateId": "b21db781-7c0d-455e-af4f-3da2d926b739",
    "estimatedTime": 946,
    "maxWaitTime": 0,
    "minWaitTime": 0,
    "gpuCount": 1,
    "paramModel": 0,
    "predTotalTimeSec": 946,
    "taskPrice": 10,
    "realUnitPrice": 30.72,
    "appMode": 1,
    "maxScheduleCount": 0,
    "estimatedPrice": 11.308,
    "optTarget": "gpu_use_rate",
    "configJson": {
        "stage": "Supervised Fine-Tuning",
        "finetuning_type": "lora",
        "val_size": 0,
        "merge_model": false,
        "task_description": "",
        "learning_rate": 0.00005,
        "num_train_epochs": 3,
        "per_device_train_batch_size": 2,
        "gradient_accumulation_steps": 8,
        "global_batch_size": 16,
        "warmup_ratio": 0.1,
        "lora_rank": 8,
        "lora_alpha": 16,
        "lr_scheduler_type": "cosine",
        "max_grad_norm": 1,
        "compute_type": "bf16",
        "cutoff_len": 1024,
        "model_name_or_path": "DeepSeek-R1-0528-Qwen3-8B",
        "publicData": [
            "alpaca_zh_demo"
        ],
        "fileData": []
    },
    "configValueAsCmdKeys": []
}'

返回响应

🟢200成功

application/json

Body

示例

{
    "code": 1001,
    "data": "llamafactory-cli train \\\n--stage sft \\\n--do_train True \\\n--model_name_or_path /shared-only/models/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B \\\n--preprocessing_num_workers 16 \\\n--finetuning_type lora \\\n--template deepseek \\\n--flash_attn auto \\\n--dataset_dir /workspace/llamafactory/data \\\n--dataset alpaca_zh_demo \\\n--cutoff_len 1024 \\\n--learning_rate 5e-05 \\\n--num_train_epochs 3.0 \\\n--max_samples 100000 \\\n--per_device_train_batch_size 2 \\\n--gradient_accumulation_steps 8 \\\n--lr_scheduler_type cosine \\\n--max_grad_norm 1.0 \\\n--logging_steps 5 \\\n--save_steps 100 \\\n--packing False \\\n--enable_thinking True \\\n--report_to none \\\n--use_swanlab True \\\n--output_dir /workspace/user-data/models/output/DeepSeek-R1-0528-Qwen3-8B/lora/train_2025-11-26-15-07-34 \\\n--bf16 True \\\n--plot_loss True \\\n--trust_remote_code True \\\n--ddp_timeout 180000000 \\\n--include_num_input_tokens_seen True \\\n--warmup_ratio 0.1 \\\n--lora_rank 8 \\\n--lora_alpha 16 \\\n--lora_dropout 0 \\\n--lora_target all \\\n--swanlab_project llamafactory \\\n--swanlab_api_key wAOxRm0oaONkjkCUyEzGV \\\n--swanlab_mode cloud ",
    "message": "操作成功"
}

修改于 2025-11-26 07:07:46

模型微调swanLab