E RKNN: [08:48:14.826] invalid rknn llm model magic E RKNN: [08:48:14.826] load model file error!

[huochen1]

rkllm init start
I rkllm: rkllm-runtime version: 1.1.2, rknpu driver version: 0.9.8, platform: RK3588

E RKNN: [08:48:14.826] invalid rknn llm model magic
E RKNN: [08:48:14.826] load model file error!
rknn_init fail! ret=-1
rkllm init success

user: o
robot: o:0: GGML_ASSERT(view_src == NULL || data_size == 0 || data_size + view_offs <= ggml_nbytes(view_src)) failed
/home/linaro/rkllm/rkllm-runtime/examples/rkllm_api_demo/../../runtime/Linux/librkllm_api/aarch64/librkllmrt.so(+0x1c7e34)[0x7fa3bf9e34]
/home/linaro/rkllm/rkllm-runtime/examples/rkllm_api_demo/../../runtime/Linux/librkllm_api/aarch64/librkllmrt.so(+0x1c9480)[0x7fa3bfb480]
/home/linaro/rkllm/rkllm-runtime/examples/rkllm_api_demo/../../runtime/Linux/librkllm_api/aarch64/librkllmrt.so(+0x1e82a8)[0x7fa3c1a2a8]
/home/linaro/rkllm/rkllm-runtime/examples/rkllm_api_demo/../../runtime/Linux/librkllm_api/aarch64/librkllmrt.so(+0x116454)[0x7fa3b48454]
/home/linaro/rkllm/rkllm-runtime/examples/rkllm_api_demo/../../runtime/Linux/librkllm_api/aarch64/librkllmrt.so(+0x1190d4)[0x7fa3b4b0d4]
/home/linaro/rkllm/rkllm-runtime/examples/rkllm_api_demo/../../runtime/Linux/librkllm_api/aarch64/librkllmrt.so(+0x17afb0)[0x7fa3bacfb0]
/home/linaro/rkllm/rkllm-runtime/examples/rkllm_api_demo/../../runtime/Linux/librkllm_api/aarch64/librkllmrt.so(+0x11eb14)[0x7fa3b50b14]
/home/linaro/rkllm/rkllm-runtime/examples/rkllm_api_demo/../../runtime/Linux/librkllm_api/aarch64/librkllmrt.so(+0x12e774)[0x7fa3b60774]
/home/linaro/rkllm/rkllm-runtime/examples/rkllm_api_demo/../../runtime/Linux/librkllm_api/aarch64/librkllmrt.so(+0xaaab4)[0x7fa3adcab4]
./llm_demo(+0x2314)[0x5561792314]
/lib/aarch64-linux-gnu/libc.so.6(__libc_start_main+0xe8)[0x7fa36c7e18]
./llm_demo(+0x1ac4)[0x5561791ac4]
您好，这个报错是不支持这个rkllm模型是吗，为什么他还会初始化成功的

[huochen1]

有可以提供测试的rkllm模型吗。

[waydong]

hi，模型是用1.1.2版本的rkllm-toolkit转换出来的吗？

[huochen1]

收到，谢谢

[huochen1]

您好，确实需要版本对应的上才行，就是需要你板端的1.1.2，rkllm-toolkit也需要1.1.2模型转化大模型

但是这个回复好像不太行的样子，我看到test.py里面的# Build model
dataset = "./data_quant.json"

Json file format, please note to add prompt in the input，like this:

[{"input":"Human: 你好！\nAssistant: ", "target": "你好！我是人工智能助手KK！"},...] 这个/data_quant.json是需要自己写的吗，我是从魔塔社区里面下的那个千问1.8模型，里面好像没有这个/data_quant.json文件，最近刚接触到部署的大模型可能这个有点不太懂了，期待您的回复

[yaoyaoyao123]

我遇到了相同的问题，请问你有1.1.4版本的rkllm模型吗

[huochen1]

没有1.1.4的我就按照文档转了1.1.2的试了试不过那个乱回答的问题不知道怎么弄

[waydong]

没有1.1.4的我就按照文档转了1.1.2的试了试不过那个乱回答的问题不知道怎么弄

llm_demo.cpp中要加上提示词，不然模型生成效果会有较大影响：
text = PROMPT_TEXT_PREFIX + input_str + PROMPT_TEXT_POSTFIX;

[Rocketman17]

I encountered the same error: "E RKNN: invalid rknn llm model magic E RKNN: load model file error!"
Converted two models into rkllm format (TinyLlama and Qwen-chat) but in both cases it throws the same error.

[waydong]

I encountered the same error: "E RKNN: invalid rknn llm model magic E RKNN: load model file error!" Converted two models into rkllm format (TinyLlama and Qwen-chat) but in both cases it throws the same error.

Please attempt to reconvert the model utilizing toolkit version 1.1.4.

[Rocketman17]

Thanks for the reply,
Apparently, conversion process was stalling in middle causing creation of corrupted model file.
I got working model now.