From: Georgi Gerganov Date: Tue, 22 Aug 2023 09:02:00 +0000 (+0300) Subject: ggml : rename xPos variables to unify coding style X-Git-Tag: upstream/0.0.1642~1272 X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=f9fbe2e7548d7d85eb7aae49c1525eff8be9ad3f;p=pkg%2Fggml%2Fsources%2Fggml ggml : rename xPos variables to unify coding style ggml-ci --- diff --git a/include/ggml/ggml.h b/include/ggml/ggml.h index 5c838689..1a34bd9b 100644 --- a/include/ggml/ggml.h +++ b/include/ggml/ggml.h @@ -1223,8 +1223,8 @@ extern "C" { struct ggml_tensor * a, int n_past, int n_dims, - float scale_base, - bool downscale); + float base, + bool down); // rotary position embedding backward, i.e compute dx from dy // a - dy @@ -1238,7 +1238,7 @@ extern "C" { float freq_base, float freq_scale, float xpos_base, - bool xpos_downscale); + bool xpos_down); // alibi position embedding // in-place, returns view(a) diff --git a/src/ggml.c b/src/ggml.c index 6a4b6ed4..b7831919 100644 --- a/src/ggml.c +++ b/src/ggml.c @@ -6716,7 +6716,7 @@ static struct ggml_tensor * ggml_rope_impl( float freq_base, float freq_scale, float xpos_base, - bool xpos_downscale, + bool xpos_down, bool inplace) { GGML_ASSERT(n_past >= 0); bool is_node = false; @@ -6730,8 +6730,8 @@ static struct ggml_tensor * ggml_rope_impl( int32_t params[8] = { n_past, n_dims, mode, n_ctx }; memcpy(params + 4, &freq_base, sizeof(float)); memcpy(params + 5, &freq_scale, sizeof(float)); - memcpy(params + 6, &xpos_base, sizeof(float)); - memcpy(params + 7, &xpos_downscale, sizeof(bool)); + memcpy(params + 6, &xpos_base, sizeof(float)); + memcpy(params + 7, &xpos_down, sizeof(bool)); ggml_set_op_params(result, params, sizeof(params)); result->op = GGML_OP_ROPE; @@ -6790,9 +6790,9 @@ struct ggml_tensor * ggml_rope_xpos_inplace( struct ggml_tensor * a, int n_past, int n_dims, - float scale_base, - bool downscale) { - return ggml_rope_impl(ctx, a, n_past, n_dims, 0, 0, 10000.0f, 1.0f, scale_base, downscale, true); + float base, + bool down) { + return ggml_rope_impl(ctx, a, n_past, n_dims, 0, 0, 10000.0f, 1.0f, base, down, true); } // ggml_rope_back @@ -6807,7 +6807,7 @@ struct ggml_tensor * ggml_rope_back( float freq_base, float freq_scale, float xpos_base, - bool xpos_downscale) { + bool xpos_down) { GGML_ASSERT(n_past >= 0); GGML_ASSERT((mode & 4) == 0 && "ggml_rope_back() for ChatGLM not implemented yet"); @@ -6820,10 +6820,10 @@ struct ggml_tensor * ggml_rope_back( struct ggml_tensor * result = ggml_dup_tensor(ctx, a); int32_t params[8] = { n_past, n_dims, mode, n_ctx }; - memcpy(params + 4, &freq_base, sizeof(float)); + memcpy(params + 4, &freq_base, sizeof(float)); memcpy(params + 5, &freq_scale, sizeof(float)); - memcpy(params + 6, &xpos_base, sizeof(float)); - memcpy(params + 7, &xpos_downscale, sizeof(bool)); + memcpy(params + 6, &xpos_base, sizeof(float)); + memcpy(params + 7, &xpos_down, sizeof(bool)); ggml_set_op_params(result, params, sizeof(params)); result->op = GGML_OP_ROPE_BACK; @@ -12177,7 +12177,7 @@ static void ggml_compute_forward_rope_f32( // these two only relevant for xPos RoPE: float xpos_base; - bool xpos_downscale; + bool xpos_down; const int n_past = ((int32_t *) dst->op_params)[0]; const int n_dims = ((int32_t *) dst->op_params)[1]; @@ -12186,7 +12186,7 @@ static void ggml_compute_forward_rope_f32( memcpy(&freq_base, (int32_t *) dst->op_params + 4, sizeof(float)); memcpy(&freq_scale, (int32_t *) dst->op_params + 5, sizeof(float)); memcpy(&xpos_base, (int32_t *) dst->op_params + 6, sizeof(float)); - memcpy(&xpos_downscale, (int32_t *) dst->op_params + 7, sizeof(bool)); + memcpy(&xpos_down, (int32_t *) dst->op_params + 7, sizeof(bool)); assert(n_past >= 0); @@ -12260,7 +12260,7 @@ static void ggml_compute_forward_rope_f32( const float sin_theta = sinf(theta); // zeta scaling for xPos only: float zeta = xpos_base != 0.0f ? powf((i0 + 0.4f * ne0) / (1.4f * ne0), (n_past + i2) / xpos_base) : 1.0f; - if (xpos_downscale) zeta = 1.0f / zeta; + if (xpos_down) zeta = 1.0f / zeta; theta *= theta_scale; @@ -12470,16 +12470,16 @@ static void ggml_compute_forward_rope_back_f32( // these two only relevant for xPos RoPE: float xpos_base; - bool xpos_downscale; + bool xpos_down; 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 = ((int32_t *) dst->op_params)[3]; UNUSED(n_ctx); memcpy(&freq_base, (int32_t *) dst->op_params + 4, sizeof(float)); memcpy(&freq_scale, (int32_t *) dst->op_params + 5, sizeof(float)); memcpy(&xpos_base, (int32_t *) dst->op_params + 6, sizeof(float)); - memcpy(&xpos_downscale, (int32_t *) dst->op_params + 7, sizeof(bool)); + memcpy(&xpos_down, (int32_t *) dst->op_params + 7, sizeof(bool)); assert(n_past >= 0); @@ -12524,7 +12524,7 @@ static void ggml_compute_forward_rope_back_f32( const float sin_theta = sinf(theta); // zeta scaling for xPos only: float zeta = xpos_base != 0.0f ? powf((i0 + 0.4f * ne0) / (1.4f * ne0), (n_past + i2) / xpos_base) : 1.0f; - if (xpos_downscale) zeta = 1.0f / zeta; + if (xpos_down) zeta = 1.0f / zeta; theta *= theta_scale; @@ -16012,12 +16012,14 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor const int n_dims = ((int32_t *) tensor->op_params)[1]; const int mode = ((int32_t *) tensor->op_params)[2]; const int n_ctx = ((int32_t *) tensor->op_params)[3]; - float freq_base, freq_scale, xpos_base; - bool xpos_downscale; + float freq_base; + float freq_scale; + float xpos_base; + bool xpos_down; memcpy(&freq_base, (int32_t *) tensor->op_params + 4, sizeof(float)); memcpy(&freq_scale, (int32_t *) tensor->op_params + 5, sizeof(float)); memcpy(&xpos_base, (int32_t *) tensor->op_params + 6, sizeof(float)); - memcpy(&xpos_downscale, (int32_t *) tensor->op_params + 7, sizeof(bool)); + memcpy(&xpos_down, (int32_t *) tensor->op_params + 7, sizeof(bool)); src0->grad = ggml_add_impl(ctx, src0->grad, @@ -16030,7 +16032,7 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor freq_base, freq_scale, xpos_base, - xpos_downscale), + xpos_down), inplace); } } break; @@ -16041,12 +16043,14 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor const int n_dims = ((int32_t *) tensor->op_params)[1]; const int mode = ((int32_t *) tensor->op_params)[2]; const int n_ctx = ((int32_t *) tensor->op_params)[3]; - float freq_base, freq_scale, xpos_base; - bool xpos_downscale; + float freq_base; + float freq_scale; + float xpos_base; + bool xpos_down; memcpy(&freq_base, (int32_t *) tensor->op_params + 4, sizeof(float)); memcpy(&freq_scale, (int32_t *) tensor->op_params + 5, sizeof(float)); memcpy(&xpos_base, (int32_t *) tensor->op_params + 6, sizeof(float)); - memcpy(&xpos_downscale, (int32_t *) tensor->op_params + 7, sizeof(bool)); + memcpy(&xpos_down, (int32_t *) tensor->op_params + 7, sizeof(bool)); src0->grad = ggml_add_impl(ctx, src0->grad, @@ -16059,7 +16063,7 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor freq_base, freq_scale, xpos_base, - xpos_downscale, + xpos_down, false), inplace); }