Feat: Apertus model implementation #15852
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -52,6 +52,15 @@ static inline float op_sqrt(float x) { | |
| return sqrtf(x); | ||
| } | ||
|
|
||
| static inline float op_xielu(float x, float alpha_n, float alpha_p, float beta, float eps) { | ||
| if (x > 0.0f) { | ||
| return alpha_p * x * x + beta * x; | ||
| } else { | ||
| const float min_x_eps = fminf(x, eps); | ||
| return (expm1f(min_x_eps) - x) * alpha_n + beta * x; | ||
| } | ||
| } | ||
|
|
||
| static inline float op_sin(float x) { | ||
| return sinf(x); | ||
| } | ||
|
|
@@ -64,8 +73,8 @@ static inline float op_log(float x) { | |
| return logf(x); | ||
| } | ||
|
|
||
| template <typename Op, typename src0_t, typename dst_t> | ||
| static inline void vec_unary_op(const Op& op, int64_t n, dst_t * y, const src0_t * x) { | ||
pwilkin marked this conversation as resolved.
Outdated
Show resolved
Hide resolved
|
||
| constexpr auto src0_to_f32 = type_conversion_table<src0_t>::to_f32; | ||
| constexpr auto f32_to_dst = type_conversion_table<dst_t >::from_f32; | ||
|
|
||
|
|
@@ -74,8 +83,8 @@ static inline void vec_unary_op(int64_t n, dst_t * y, const src0_t * x) { | |
| } | ||
| } | ||
|
|
||
| template <typename Op, typename src0_t, typename dst_t> | ||
| static void apply_unary_op(const Op& op, const ggml_compute_params * params, ggml_tensor * dst) { | ||
| const ggml_tensor * src0 = dst->src[0]; | ||
|
|
||
| GGML_ASSERT(ggml_is_contiguous_1(src0) && ggml_is_contiguous_1(dst) && ggml_are_same_shape(src0, dst)); | ||
|
|
@@ -95,25 +104,25 @@ static void apply_unary_op(const ggml_compute_params * params, ggml_tensor * dst | |
| dst_t * dst_ptr = (dst_t *) ((char *) dst->data + i03*nb3 + i02*nb2 + i01*nb1 ); | ||
| const src0_t * src0_ptr = (const src0_t *) ((const char *) src0->data + i03*nb03 + i02*nb02 + i01*nb01); | ||
|
|
||
| vec_unary_op<decltype(op), src0_t, dst_t>(op, ne0, dst_ptr, src0_ptr); | ||
| } | ||
| } | ||
|
|
||
| // TODO: Use the 'traits' lookup table (for type conversion fns), instead of a mass of 'if' conditions with long templates | ||
| template <typename Op> | ||
| static void unary_op(const Op& op, const ggml_compute_params * params, ggml_tensor * dst) { | ||
|
||
| const ggml_tensor * src0 = dst->src[0]; | ||
|
|
||
| /* */ if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) { // all f32 | ||
| apply_unary_op<decltype(op), float, float>(op, params, dst); | ||
| } else if (src0->type == GGML_TYPE_F16 && dst->type == GGML_TYPE_F16) { // all f16 | ||
| apply_unary_op<decltype(op), ggml_fp16_t, ggml_fp16_t>(op, params, dst); | ||
| } else if (src0->type == GGML_TYPE_BF16 && dst->type == GGML_TYPE_BF16) { // all bf16 | ||
| apply_unary_op<decltype(op), ggml_bf16_t, ggml_bf16_t>(op, params, dst); | ||
| } else if (src0->type == GGML_TYPE_BF16 && dst->type == GGML_TYPE_F32) { | ||
| apply_unary_op<decltype(op), ggml_bf16_t, float>(op, params, dst); | ||
| } else if (src0->type == GGML_TYPE_F16 && dst->type == GGML_TYPE_F32) { | ||
| apply_unary_op<decltype(op), ggml_fp16_t, float>(op, params, dst); | ||
| } else { | ||
| fprintf(stderr, "%s: unsupported types: dst: %s, src0: %s\n", __func__, | ||
| ggml_type_name(dst->type), ggml_type_name(src0->type)); | ||
|
|
@@ -122,65 +131,89 @@ static void unary_op(const ggml_compute_params * params, ggml_tensor * dst) { | |
| } | ||
|
|
||
| void ggml_compute_forward_abs(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_abs, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_sgn(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_sgn, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_neg(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_neg, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_step(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_step, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_tanh(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_tanh, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_elu(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_elu, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_relu(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_relu, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_sigmoid(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_sigmoid, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_hardsigmoid(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_hardsigmoid, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_exp(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_exp, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_hardswish(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_hardswish, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_sqr(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_sqr, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_sqrt(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_sqrt, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_sin(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_sin, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_cos(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_cos, params, dst); | ||
| } | ||
|
|
||
| void ggml_compute_forward_log(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| unary_op(op_log, params, dst); | ||
| } | ||
|
|
||
| static float softplus(float input, float beta=1.0f, float threshold=20.0f) { | ||
| if (input * beta > threshold) return input; | ||
| return (1/beta) * logf(1 + expf(beta * input)); | ||
| } | ||
|
|
||
| void ggml_compute_forward_xielu(const ggml_compute_params * params, ggml_tensor * dst) { | ||
| // Get the XIELU parameters from the operation | ||
| const float * op_params = (const float*)dst->op_params; | ||
| float alpha_n = op_params[0]; | ||
| float alpha_p = op_params[1]; | ||
| const float beta = op_params[2]; | ||
| const float eps = op_params[3]; | ||
|
|
||
| // alpha_p = softplus(alpha_p); | ||
| // alpha_n = beta + softplus(alpha_n); | ||
|
|
||
| const auto xielu_op_params = [alpha_n, alpha_p, beta, eps](float f) { | ||
| return op_xielu(f, alpha_n, alpha_p, beta, eps); | ||
| }; | ||
|
|
||
| unary_op(xielu_op_params, params, dst); | ||
| } | ||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Any reason not to put this along the rest of the unary ops like RELU, GELU, etc.?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Not really :>
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@ggerganov Done.