[使用问题] Ai00推理报错:ERROR [ai00_core::run] tensor shape not match: (96, 2048, 1, 1) vs. (2048, 2048, 1, 1)

萌新第一次使用Ai00,成功加载模型后使用Chat聊天时报错

错误信息:

2025-02-03T11:36:24.264Z INFO  [ai00_core::run] queued task at slot 15
2025-02-03T11:36:24.275Z INFO  [ai00_core::run] slot 15 schedules future back of length 37
2025-02-03T11:36:24.283Z ERROR [ai00_core::run] tensor shape not match: (96, 2048, 1, 1) vs. (2048, 2048, 1, 1)
2025-02-03T11:36:57.766Z INFO  [ai00_core] loading model "assets/models\\RWKV-x070-World-1.5B-v3-20250127-ctx4096.st"
2025-02-03T11:36:57.766Z INFO  [ai00_core] ModelInfo {
    version: V7,
    num_layer: 24,
    num_emb: 2048,
    num_hidden: 8192,
    num_vocab: 65536,
    num_head: 32,
    custom: V7(
        CustomInfo {
            w: 2048,
            a: 2048,
            g: 2048,
            v: 2048,
        },
    ),
}
2025-02-03T11:36:57.766Z INFO  [ai00_core] type: SafeTensors
2025-02-03T11:36:58.167Z INFO  [ai00_core] AdapterInfo {
    name: "NVIDIA GeForce GTX 1650",
    vendor: 4318,
    device: 8089,
    device_type: DiscreteGpu,
    driver: "NVIDIA",
    driver_info: "566.36",
    backend: Vulkan,
}
2025-02-03T11:36:58.315Z INFO  [web_rwkv::context] context 1930 destroyed
2025-02-03T11:37:05.943Z INFO  [ai00_core] model loaded
2025-02-03T11:37:17.498Z INFO  [ai00_core::run] start at empty slot 15
2025-02-03T11:37:17.514Z INFO  [ai00_core::run] slot 15 checks out cache of length 0
2025-02-03T11:37:17.514Z INFO  [ai00_core::run] queued task at slot 15
2025-02-03T11:37:17.525Z INFO  [ai00_core::run] slot 15 schedules future back of length 37
2025-02-03T11:37:17.644Z ERROR [ai00_core::run] tensor shape not match: (96, 2048, 1, 1) vs. (2048, 2048, 1, 1)
2025-02-03T11:37:26.969Z INFO  [ai00_core::run] start at empty slot 14
2025-02-03T11:37:26.984Z INFO  [ai00_core::run] slot 14 checks out cache of length 0
2025-02-03T11:37:26.984Z INFO  [ai00_core::run] queued task at slot 14
2025-02-03T11:37:26.986Z ERROR [ai00_core::run] tensor shape not match: (96, 2048, 1, 1) vs. (2048, 2048, 1, 1)

环境:Windows,前后端
权重:用RWKV-Runner的WebGPU (Python)转换得到的RWKV-x070-World-1.5B-v3-20250127-ctx4096.st


已解决。官方群里大佬解答,RWKV-Runner可能暂时没支持将RWKV-7模型转换成safetensors格式,应该先用ai00_server的转换脚本
(显然是我没好好看README的使用说明 :joy:#convert-the-model