ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

llama: Add support for RWKV v7 architecture (#12412)

Browse source 
* ggml: Add op l2_norm

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>

* ggml: Add op rwkv_wkv7

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>

* llama: Add support for RWKV7 and ARWKV7 models

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>

* llama: fix inference with RWKV6Qwen2

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>

* llama: add more (a)rwkv7 variants in size

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>

* Apply code-format changes

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>

* fix MUSA build

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>

* llama: fix shape error with rwkv using llama-parallel

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>

---------

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>
This commit is contained in:
Molly Sophia 2025-03-18 07:27:50 +08:00 • committed by GitHub
parent 60c902926c
commit 7dfad387e3
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
35 changed files with 2948 additions and 438 deletions
Download patch file Download diff file Expand all files Collapse all files

10
ggml/src/ggml-cuda/ggml-cuda.cu
View file

@ -36,7 +36,7 @@
#include "ggml-cuda/tsembd.cuh"
#include "ggml-cuda/unary.cuh"
#include "ggml-cuda/upscale.cuh"
#include "ggml-cuda/wkv6.cuh"
#include "ggml-cuda/wkv.cuh"
#include "ggml-cuda/gla.cuh"
#include "ggml.h"
@ -2196,6 +2196,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
case GGML_OP_GROUP_NORM:
ggml_cuda_op_group_norm(ctx, dst);
break;
case GGML_OP_L2_NORM:
ggml_cuda_op_l2_norm(ctx, dst);
break;
case GGML_OP_CONCAT:
ggml_cuda_op_concat(ctx, dst);
break;
@ -2304,6 +2307,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
case GGML_OP_GATED_LINEAR_ATTN:
ggml_cuda_op_gated_linear_attn(ctx, dst);
break;
case GGML_OP_RWKV_WKV7:
ggml_cuda_op_rwkv_wkv7(ctx, dst);
break;
case GGML_OP_CROSS_ENTROPY_LOSS_BACK:
ggml_cuda_cross_entropy_loss_back(ctx, dst);
break;
@ -3161,6 +3167,7 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
break;
case GGML_OP_NORM:
case GGML_OP_RMS_NORM:
case GGML_OP_L2_NORM:
return true;
case GGML_OP_RMS_NORM_BACK:
return ggml_is_contiguous(op->src[0]) && op->ne[0] % WARP_SIZE == 0;
@ -3215,6 +3222,7 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
case GGML_OP_LEAKY_RELU:
case GGML_OP_RWKV_WKV6:
case GGML_OP_GATED_LINEAR_ATTN:
case GGML_OP_RWKV_WKV7:
return true;
case GGML_OP_FLASH_ATTN_EXT: {
#ifndef FLASH_ATTN_AVAILABLE