mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-11-15 15:29:53 +00:00
06943a69f6
* ggml : move rope type enum to ggml.h
This commit moves the `llama_rope_type` enum from `llama.h` to
`ggml.h` and changes its name to `ggml_rope_type`.
The motivation for this change is to address the TODO in `llama.h` and
use the enum in ggml.
Note: This commit does not change the `mode` parameter to be of type
`enum ggml_rope_type`. The name `mode` and its usage suggest that it
might be more generic and possibly used as a bit field for multiple
flags. Further investigation/discussion may be needed to determine
if `mode` should be restricted to RoPE types.
* squash! ggml : move rope type enum to ggml.h
This commit removes GGML_ROPE_TYPE_NONE and GGML_ROPE_TYPE_GLM from
ggml.h, and back the llama_rope_type enum.
I've kept the assert for GGML_ROPE_TYPE_GLM as I'm not sure if it is
safe to remove it yet.
* squash! ggml : move rope type enum to ggml.h
This commit removes the enum ggml_rope_type from ggml.h and replaces it
with a define (GGML_ROPE_TYPE_NEOX). This define is used in the code to
check if the mode is set to GPT-NeoX. Also the enum llama_rope_type has
been updated to reflect this change.
* squash! ggml : move rope type enum to ggml.h
This commit contains a suggestion enable the GGML_ROPE_TYPE_NEOX
macro/define to be passed to the shader compiler.
* squash! ggml : move rope type enum to ggml.h
This commit fixes the editorconfig-checker warnings.
* squash! ggml : move rope type enum to ggml.h
Update comment for ggml_rope function.
* Revert "squash! ggml : move rope type enum to ggml.h"
This reverts commit 6261222bd0.
* squash! ggml : move rope type enum to ggml.h
Add GGML_ROPE_TYPE_NEOX to rope_common.comp.
* remove extra line
---------
Co-authored-by: slaren <slarengh@gmail.com>
276 lines
10 KiB
C++
276 lines
10 KiB
C++
#include "rope.hpp"
|
|
|
|
struct rope_corr_dims {
|
|
float v[2];
|
|
};
|
|
|
|
static float rope_yarn_ramp(const float low, const float high, const int i0) {
|
|
const float y = (i0 / 2 - low) / sycl::max(0.001f, high - low);
|
|
return 1.0f - sycl::min(1.0f, sycl::max(0.0f, y));
|
|
}
|
|
|
|
// YaRN algorithm based on LlamaYaRNScaledRotaryEmbedding.py from https://github.com/jquesnelle/yarn
|
|
// MIT licensed. Copyright (c) 2023 Jeffrey Quesnelle and Bowen Peng.
|
|
static void rope_yarn(
|
|
float theta_extrap, float freq_scale, rope_corr_dims corr_dims, int64_t i0, float ext_factor, float mscale,
|
|
float * cos_theta, float * sin_theta) {
|
|
// Get n-d rotational scaling corrected for extrapolation
|
|
float theta_interp = freq_scale * theta_extrap;
|
|
float theta = theta_interp;
|
|
if (ext_factor != 0.0f) {
|
|
float ramp_mix = rope_yarn_ramp(corr_dims.v[0], corr_dims.v[1], i0) * ext_factor;
|
|
theta = theta_interp * (1 - ramp_mix) + theta_extrap * ramp_mix;
|
|
|
|
// Get n-d magnitude scaling corrected for interpolation
|
|
mscale *= 1.0f + 0.1f * sycl::log(1.0f / freq_scale);
|
|
}
|
|
*cos_theta = sycl::cos(theta) * mscale;
|
|
*sin_theta = sycl::sin(theta) * mscale;
|
|
}
|
|
|
|
template<typename T, bool has_ff>
|
|
static void rope_norm(
|
|
const T * x, T * dst, int ne0, int n_dims, const int32_t * pos, float freq_scale, int p_delta_rows,
|
|
float ext_factor, float attn_factor, rope_corr_dims corr_dims, float theta_scale, const float * freq_factors,
|
|
const sycl::nd_item<3> &item_ct1) {
|
|
const int i0 = 2 * (item_ct1.get_local_range(1) * item_ct1.get_group(1) +
|
|
item_ct1.get_local_id(1));
|
|
|
|
if (i0 >= ne0) {
|
|
return;
|
|
}
|
|
|
|
const int row = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
|
|
item_ct1.get_local_id(2);
|
|
|
|
if (i0 >= n_dims) {
|
|
const int i = row*ne0 + i0;
|
|
|
|
dst[i + 0] = x[i + 0];
|
|
dst[i + 1] = x[i + 1];
|
|
|
|
return;
|
|
}
|
|
|
|
const int i = row*ne0 + i0;
|
|
const int i2 = row/p_delta_rows;
|
|
|
|
const float theta_base = pos[i2] * sycl::pow(theta_scale, i0 / 2.0f);
|
|
|
|
const float freq_factor = has_ff ? freq_factors[i0/2] : 1.0f;
|
|
|
|
float cos_theta;
|
|
float sin_theta;
|
|
|
|
rope_yarn(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, &cos_theta, &sin_theta);
|
|
|
|
const float x0 = x[i + 0];
|
|
const float x1 = x[i + 1];
|
|
|
|
dst[i + 0] = x0*cos_theta - x1*sin_theta;
|
|
dst[i + 1] = x0*sin_theta + x1*cos_theta;
|
|
}
|
|
|
|
template<typename T, bool has_ff>
|
|
static void rope_neox(
|
|
const T * x, T * dst, int ne0, int n_dims, const int32_t * pos, float freq_scale, int p_delta_rows,
|
|
float ext_factor, float attn_factor, rope_corr_dims corr_dims, float theta_scale, const float * freq_factors,
|
|
const sycl::nd_item<3> &item_ct1) {
|
|
const int i0 = 2 * (item_ct1.get_local_range(1) * item_ct1.get_group(1) +
|
|
item_ct1.get_local_id(1));
|
|
|
|
if (i0 >= ne0) {
|
|
return;
|
|
}
|
|
|
|
const int row = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
|
|
item_ct1.get_local_id(2);
|
|
|
|
if (i0 >= n_dims) {
|
|
const int i = row*ne0 + i0;
|
|
|
|
dst[i + 0] = x[i + 0];
|
|
dst[i + 1] = x[i + 1];
|
|
|
|
return;
|
|
}
|
|
|
|
const int i = row*ne0 + i0/2;
|
|
const int i2 = row/p_delta_rows;
|
|
|
|
const float theta_base = pos[i2] * sycl::pow(theta_scale, i0 / 2.0f);
|
|
|
|
const float freq_factor = has_ff ? freq_factors[i0/2] : 1.0f;
|
|
|
|
float cos_theta;
|
|
float sin_theta;
|
|
|
|
rope_yarn(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, &cos_theta, &sin_theta);
|
|
|
|
const float x0 = x[i + 0];
|
|
const float x1 = x[i + n_dims/2];
|
|
|
|
dst[i + 0] = x0*cos_theta - x1*sin_theta;
|
|
dst[i + n_dims/2] = x0*sin_theta + x1*cos_theta;
|
|
}
|
|
|
|
template <typename T>
|
|
static void rope_norm_sycl(
|
|
const T *x, T *dst, int ne0, int n_dims, int nr, const int32_t *pos, float freq_scale, int p_delta_rows,
|
|
float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, queue_ptr stream) {
|
|
GGML_ASSERT(ne0 % 2 == 0);
|
|
const sycl::range<3> block_dims(1, SYCL_ROPE_BLOCK_SIZE, 1);
|
|
const int num_blocks_x = (ne0 + 2*SYCL_ROPE_BLOCK_SIZE - 1) / (2*SYCL_ROPE_BLOCK_SIZE);
|
|
const sycl::range<3> block_nums(1, num_blocks_x, nr);
|
|
|
|
const float theta_scale = powf(freq_base, -2.0f/n_dims);
|
|
|
|
dpct::has_capability_or_fail(stream->get_device(),
|
|
{sycl::aspect::fp16});
|
|
|
|
if (freq_factors == nullptr) {
|
|
/*
|
|
DPCT1049:40: The work-group size passed to the SYCL kernel may exceed
|
|
the limit. To get the device limit, query
|
|
info::device::max_work_group_size. Adjust the work-group size if needed.
|
|
*/
|
|
stream->parallel_for(
|
|
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
|
[=](sycl::nd_item<3> item_ct1) {
|
|
rope_norm<T, false>(x, dst, ne0, n_dims, pos, freq_scale, p_delta_rows,
|
|
ext_factor, attn_factor, corr_dims, theta_scale, freq_factors,
|
|
item_ct1);
|
|
});
|
|
} else {
|
|
/*
|
|
DPCT1049:41: The work-group size passed to the SYCL kernel may exceed
|
|
the limit. To get the device limit, query
|
|
info::device::max_work_group_size. Adjust the work-group size if needed.
|
|
*/
|
|
stream->parallel_for(
|
|
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
|
[=](sycl::nd_item<3> item_ct1) {
|
|
rope_norm<T, true>(x, dst, ne0, n_dims, pos, freq_scale, p_delta_rows,
|
|
ext_factor, attn_factor, corr_dims, theta_scale, freq_factors,
|
|
item_ct1);
|
|
});
|
|
}
|
|
}
|
|
|
|
template <typename T>
|
|
static void rope_neox_sycl(
|
|
const T *x, T *dst, int ne0, int n_dims, int nr, const int32_t *pos, float freq_scale, int p_delta_rows,
|
|
float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, queue_ptr stream) {
|
|
GGML_ASSERT(ne0 % 2 == 0);
|
|
const sycl::range<3> block_dims(1, SYCL_ROPE_BLOCK_SIZE, 1);
|
|
const int num_blocks_x = (ne0 + 2*SYCL_ROPE_BLOCK_SIZE - 1) / (2*SYCL_ROPE_BLOCK_SIZE);
|
|
const sycl::range<3> block_nums(1, num_blocks_x, nr);
|
|
|
|
const float theta_scale = powf(freq_base, -2.0f/n_dims);
|
|
|
|
dpct::has_capability_or_fail(stream->get_device(),
|
|
{sycl::aspect::fp16});
|
|
|
|
if (freq_factors == nullptr) {
|
|
stream->parallel_for(
|
|
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
|
[=](sycl::nd_item<3> item_ct1) {
|
|
rope_neox<T, false>(x, dst, ne0, n_dims, pos, freq_scale,
|
|
p_delta_rows, ext_factor, attn_factor,
|
|
corr_dims, theta_scale, freq_factors,
|
|
item_ct1);
|
|
});
|
|
} else {
|
|
stream->parallel_for(
|
|
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
|
[=](sycl::nd_item<3> item_ct1) {
|
|
rope_neox<T, true>(x, dst, ne0, n_dims, pos, freq_scale,
|
|
p_delta_rows, ext_factor, attn_factor,
|
|
corr_dims, theta_scale, freq_factors,
|
|
item_ct1);
|
|
});
|
|
}
|
|
}
|
|
|
|
void ggml_sycl_op_rope(
|
|
ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1, ggml_tensor *dst,
|
|
const float *src0_dd, const float *src1_dd, float *dst_dd, const queue_ptr &main_stream) {
|
|
const ggml_tensor * src2 = dst->src[2];
|
|
|
|
GGML_ASSERT(src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_F16);
|
|
GGML_ASSERT( dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
|
|
GGML_ASSERT(src0->type == dst->type);
|
|
|
|
const int64_t ne00 = src0->ne[0];
|
|
const int64_t ne01 = src0->ne[1];
|
|
const int64_t nr = ggml_nrows(src0);
|
|
|
|
//const int n_past = ((int32_t *) dst->op_params)[0];
|
|
const int n_dims = ((int32_t *) dst->op_params)[1];
|
|
const int mode = ((int32_t *) dst->op_params)[2];
|
|
//const int n_ctx = ((int32_t *) dst->op_params)[3];
|
|
const int n_ctx_orig = ((int32_t *) dst->op_params)[4];
|
|
|
|
// RoPE alteration for extended context
|
|
float freq_base;
|
|
float freq_scale;
|
|
float ext_factor;
|
|
float attn_factor;
|
|
float beta_fast;
|
|
float beta_slow;
|
|
|
|
memcpy(&freq_base, (int32_t *) dst->op_params + 5, sizeof(float));
|
|
memcpy(&freq_scale, (int32_t *) dst->op_params + 6, sizeof(float));
|
|
memcpy(&ext_factor, (int32_t *) dst->op_params + 7, sizeof(float));
|
|
memcpy(&attn_factor, (int32_t *) dst->op_params + 8, sizeof(float));
|
|
memcpy(&beta_fast, (int32_t *) dst->op_params + 9, sizeof(float));
|
|
memcpy(&beta_slow, (int32_t *) dst->op_params + 10, sizeof(float));
|
|
|
|
const bool is_neox = mode & GGML_ROPE_TYPE_NEOX;
|
|
|
|
const int32_t * pos = (const int32_t *) src1_dd;
|
|
|
|
const float * freq_factors = nullptr;
|
|
if (src2 != nullptr) {
|
|
freq_factors = (const float *) src2->data;
|
|
}
|
|
|
|
rope_corr_dims corr_dims;
|
|
ggml_rope_yarn_corr_dims(n_dims, n_ctx_orig, freq_base, beta_fast, beta_slow, corr_dims.v);
|
|
|
|
// compute
|
|
if (is_neox) {
|
|
if (src0->type == GGML_TYPE_F32) {
|
|
rope_neox_sycl(
|
|
(const float *)src0_dd, (float *)dst_dd, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
|
|
attn_factor, corr_dims, freq_factors, main_stream
|
|
);
|
|
} else if (src0->type == GGML_TYPE_F16) {
|
|
rope_neox_sycl(
|
|
(const sycl::half *)src0_dd, (sycl::half *)dst_dd, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
|
|
attn_factor, corr_dims, freq_factors, main_stream
|
|
);
|
|
} else {
|
|
GGML_ABORT("fatal error");
|
|
}
|
|
} else {
|
|
if (src0->type == GGML_TYPE_F32) {
|
|
rope_norm_sycl(
|
|
(const float *)src0_dd, (float *)dst_dd, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
|
|
attn_factor, corr_dims, freq_factors, main_stream
|
|
);
|
|
} else if (src0->type == GGML_TYPE_F16) {
|
|
rope_norm_sycl(
|
|
(const sycl::half *)src0_dd, (sycl::half *)dst_dd, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
|
|
attn_factor, corr_dims, freq_factors, main_stream
|
|
);
|
|
} else {
|
|
GGML_ABORT("fatal error");
|
|
}
|
|
}
|
|
|
|
(void) src1;
|
|
(void) dst;
|
|
(void) src1_dd;
|
|
}
|