mirror of
https://github.com/likelovewant/ollama-for-amd.git
synced 2025-12-21 14:26:30 +00:00
b95693056c
* feat: Bump llama.cpp to the latest master (17f7f4b) This brings in significant improvements to prefill performance for all models using the SSM_CONV and SSM_SCAN ops (granite4, jamba, falcon-h, nemotron-h, Qwen3 Next) on Apple Metal. See https://github.com/ggml-org/llama.cpp/pull/17876 Branch: LlamaCPPMetalSSMImprovements Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Update patches 1-4 Branch: LlamaCPPMetalSSMImprovements Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Update patches 5-12 Branch: LlamaCPPMetalSSMImprovements Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Update patches 13-18 Branch: LlamaCPPMetalSSMImprovements Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Update patch 20 Branch: LlamaCPPMetalSSMImprovements Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Update patches 21-31 Branch: LlamaCPPMetalSSMImprovements Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Sync vendored code The two files I'm not sure about here are the swap from gemma3-iswa.cpp to gemma3.cpp (I chose to include this because I think it's required), and the inclusion of `ggml-zendnn.h` which I chose to omit. Branch: LlamaCPPMetalSSMImprovements Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> --------- Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
107 lines
5.1 KiB
Diff
107 lines
5.1 KiB
Diff
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
|
|
From: Michael Yang <git@mxy.ng>
|
|
Date: Thu, 16 Oct 2025 20:37:19 -0700
|
|
Subject: [PATCH] interleave multi rope
|
|
|
|
since ollama doesn't use mrope for anything else, change it to mean the
|
|
interleaved version used for qwen3vl
|
|
---
|
|
ggml/src/ggml-cpu/ops.cpp | 8 ++++----
|
|
ggml/src/ggml-cuda/rope.cu | 8 ++++----
|
|
ggml/src/ggml-metal/ggml-metal.metal | 8 ++++----
|
|
ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl | 8 ++++----
|
|
4 files changed, 16 insertions(+), 16 deletions(-)
|
|
|
|
diff --git a/ggml/src/ggml-cpu/ops.cpp b/ggml/src/ggml-cpu/ops.cpp
|
|
index 7d1733adb..f4aae5332 100644
|
|
--- a/ggml/src/ggml-cpu/ops.cpp
|
|
+++ b/ggml/src/ggml-cpu/ops.cpp
|
|
@@ -5599,14 +5599,14 @@ static void ggml_mrope_cache_init(
|
|
|
|
float theta = theta_t;
|
|
if (is_imrope) { // qwen3vl apply interleaved mrope
|
|
- if (sector % 3 == 1 && sector < 3 * sections[1]) {
|
|
+ if (sector % 3 == 1 && sector < 1 + 3 * sections[1]) {
|
|
theta = theta_h;
|
|
- } else if (sector % 3 == 2 && sector < 3 * sections[2]) {
|
|
+ } else if (sector % 3 == 2 && sector < 2 + 3 * sections[2]) {
|
|
theta = theta_w;
|
|
} else if (sector % 3 == 0 && sector < 3 * sections[0]) {
|
|
theta = theta_t;
|
|
- } else {
|
|
- theta = theta_e;
|
|
+ // } else {
|
|
+ // theta = theta_e;
|
|
}
|
|
} else {
|
|
if (sector >= sections[0] && sector < sec_w) {
|
|
diff --git a/ggml/src/ggml-cuda/rope.cu b/ggml/src/ggml-cuda/rope.cu
|
|
index 88ed79111..71ca60214 100644
|
|
--- a/ggml/src/ggml-cuda/rope.cu
|
|
+++ b/ggml/src/ggml-cuda/rope.cu
|
|
@@ -200,14 +200,14 @@ static __global__ void rope_multi(
|
|
|
|
float theta_base = 0.0;
|
|
if (is_imrope) {
|
|
- if (sector % 3 == 1 && sector < 3 * sections.v[1]) { // h
|
|
+ if (sector % 3 == 1 && sector < 1 + 3 * sections.v[1]) { // h
|
|
theta_base = pos[channel_x + ne2 * 1]*powf(theta_scale, i0/2.0f);
|
|
- } else if (sector % 3 == 2 && sector < 3 * sections.v[2]) { // w
|
|
+ } else if (sector % 3 == 2 && sector < 2 + 3 * sections.v[2]) { // w
|
|
theta_base = pos[channel_x + ne2 * 2]*powf(theta_scale, i0/2.0f);
|
|
} else if (sector % 3 == 0 && sector < 3 * sections.v[0]) { // t
|
|
theta_base = pos[channel_x]*powf(theta_scale, i0/2.0f);
|
|
- } else {
|
|
- theta_base = pos[channel_x + ne2 * 3]*powf(theta_scale, i0/2.0f);
|
|
+ // } else {
|
|
+ // theta_base = pos[channel_x + ne2 * 3]*powf(theta_scale, i0/2.0f);
|
|
}
|
|
} else {
|
|
if (sector < sections.v[0]) {
|
|
diff --git a/ggml/src/ggml-metal/ggml-metal.metal b/ggml/src/ggml-metal/ggml-metal.metal
|
|
index 236838e9e..c98d269d1 100644
|
|
--- a/ggml/src/ggml-metal/ggml-metal.metal
|
|
+++ b/ggml/src/ggml-metal/ggml-metal.metal
|
|
@@ -4242,14 +4242,14 @@ kernel void kernel_rope_multi(
|
|
|
|
float theta_base;
|
|
if (FC_rope_is_imrope) {
|
|
- if (sector % 3 == 1 && sector < 3 * args.sect_1) { // h
|
|
+ if (sector % 3 == 1 && sector < 1 + 3 * args.sect_1) { // h
|
|
theta_base = (float) pos[i2 + args.ne02 * 1];
|
|
- } else if (sector % 3 == 2 && sector < 3 * args.sect_2) { // w
|
|
+ } else if (sector % 3 == 2 && sector < 2 + 3 * args.sect_2) { // w
|
|
theta_base = (float) pos[i2 + args.ne02 * 2];
|
|
} else if (sector % 3 == 0 && sector < 3 * args.sect_0) { // t
|
|
theta_base = (float) pos[i2 + args.ne02 * 0];
|
|
- } else { // e
|
|
- theta_base = (float) pos[i2 + args.ne02 * 3];
|
|
+ // } else { // e
|
|
+ // theta_base = (float) pos[i2 + args.ne02 * 3];
|
|
}
|
|
} else {
|
|
if (sector < args.sect_0) {
|
|
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl b/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl
|
|
index 9726b722d..1c8c69422 100644
|
|
--- a/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl
|
|
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl
|
|
@@ -148,14 +148,14 @@ void rope_multi(const uint i0, const uint i1, rope_params p) {
|
|
|
|
float theta_base = 0.0;
|
|
if (p.is_imrope != 0) {
|
|
- if (sector % 3 == 1 && sector < 3 * p.sections[1]) {
|
|
+ if (sector % 3 == 1 && sector < 1 + 3 * p.sections[1]) {
|
|
theta_base = rope_data_pos[i02 + ne2 * 1]*pow(p.theta_scale, i0/2.0f);
|
|
- } else if (sector % 3 == 2 && sector < 3 * p.sections[2]) {
|
|
+ } else if (sector % 3 == 2 && sector < 2 + 3 * p.sections[2]) {
|
|
theta_base = rope_data_pos[i02 + ne2 * 2]*pow(p.theta_scale, i0/2.0f);
|
|
} else if (sector % 3 == 0 && sector < 3 * p.sections[0]) {
|
|
theta_base = rope_data_pos[i02]*pow(p.theta_scale, i0/2.0f);
|
|
- } else {
|
|
- theta_base = rope_data_pos[i02 + ne2 * 3]*pow(p.theta_scale, i0/2.0f);
|
|
+ //} else {
|
|
+ // theta_base = rope_data_pos[i02 + ne2 * 3]*pow(p.theta_scale, i0/2.0f);
|
|
}
|
|
} else {
|
|
if (sector < p.sections[0]) {
|