RWKV is an RNN with transformer-level LLM performance. It can be directly trained like a GPT (parallelizable). So it's combining the best of RNN and transformer - great performance, fast inference, saves VRAM, fast training, "infinite" ctx_len, and free sentence embedding.
Apache License 2.0
12.05k
stars
827
forks
source link
报错求解RuntimeError: CUDA error: CUBLAS_STATUS_INVALID_VALUE when calling `cublasGemmEx( handle, opa, opb, m, n, k, &falpha, a,CUDA_R_16F, lda, b, CUDA_R_16F, ldb, &fbeta, c, CUDA_R_16F, ldc, CUDA_R_32F, CUBLAS_GEMM_DFALT_TENSOR_OP)` #144
多次遇到类似的问题,是模型结构定义问题吗?![image](https://github.com/BlinkDL/RWKV-LM/assets/87795698/e6fb9e68-37e3-45d0-a9da-40cfe463e2b4)