finetune : add -ngl parameter (#3762)

* Add '-ngl' support to finetune.cpp

* Add fprintf in ggml_cuda_op_add

When I tried CUDA offloading during finetuning following the readme, I got an assert here.
This probably isn't an important case because inference later gives a warning saying you should use f16 or f32 instead when using lora

* Add 'finetune.sh', which currently fails when using GPU

"error: operator (): Finetuning on tensors with type 'f16' is not yet supported"

* tweak finetune.sh

* Suppress some warnings in ggml.c

* Add f16 implementation to ggml_compute_forward_add_f16_f32

* Add an f16 case to ggml_add_cast_impl and llama_build_lora_finetune_graphs

* finetune.sh: Edit comments

* Add "add_f16_f32_f32_cuda"

* Tweak an error message

* finetune.sh: Add an optional LLAMA_MODEL_DIR variable

* finetune.sh: Add an optional LLAMA_TRAINING_DIR variable

* train : minor

* tabs to spaces

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>

1 files changed, 38 insertions, 15 deletions

diff --git a/ggml.c b/ggml.c
index 84407b12..80d68225 100644
--- a/ggml.c
+++ b/ggml.c
@@ -3153,7 +3153,7 @@ static struct ggml_tensor * ggml_add_cast_impl(
     // TODO: support less-strict constraint
     //       GGML_ASSERT(ggml_can_repeat(b, a));
     GGML_ASSERT(ggml_can_repeat_rows(b, a));
-    GGML_ASSERT(ggml_is_quantized(a->type)); // currently only supported for quantized input
+    GGML_ASSERT(ggml_is_quantized(a->type) || a->type == GGML_TYPE_F16); // currently only supported for quantized input and f16
 
     bool is_node = false;
 
@@ -6927,9 +6927,15 @@ static void ggml_compute_forward_add_f16_f32(
 
     GGML_ASSERT(src0->type == GGML_TYPE_F16);
     GGML_ASSERT(src1->type == GGML_TYPE_F32);
-    GGML_ASSERT(dst->type  == GGML_TYPE_F16);
 
-    GGML_ASSERT( nb0 == sizeof(ggml_fp16_t));
+    if (dst->type == GGML_TYPE_F32) {
+        GGML_ASSERT( nb0 == sizeof(float));
+    }
+    else {
+        GGML_ASSERT(dst->type  == GGML_TYPE_F16);
+        GGML_ASSERT( nb0 == sizeof(ggml_fp16_t));
+    }
+
     GGML_ASSERT(nb00 == sizeof(ggml_fp16_t));
 
     // rows per thread
@@ -6940,18 +6946,35 @@ static void ggml_compute_forward_add_f16_f32(
     const int ir1 = MIN(ir0 + dr, nr);
 
     if (nb10 == sizeof(float)) {
-        for (int ir = ir0; ir < ir1; ++ir) {
-            // src0, src1 and dst are same shape => same indices
-            const int i3 = ir/(ne2*ne1);
-            const int i2 = (ir - i3*ne2*ne1)/ne1;
-            const int i1 = (ir - i3*ne2*ne1 - i2*ne1);
-
-            ggml_fp16_t * dst_ptr  = (ggml_fp16_t *) ((char *) dst->data  + i3*nb3  + i2*nb2  + i1*nb1);
-            ggml_fp16_t * src0_ptr = (ggml_fp16_t *) ((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01);
-            float *       src1_ptr = (float *)       ((char *) src1->data + i3*nb13 + i2*nb12 + i1*nb11);
-
-            for (int i = 0; i < ne0; i++) {
-                dst_ptr[i] = GGML_FP32_TO_FP16(GGML_FP16_TO_FP32(src0_ptr[i]) + src1_ptr[i]);
+        if (dst->type == GGML_TYPE_F16) {
+            for (int ir = ir0; ir < ir1; ++ir) {
+                // src0, src1 and dst are same shape => same indices
+                const int i3 = ir/(ne2*ne1);
+                const int i2 = (ir - i3*ne2*ne1)/ne1;
+                const int i1 = (ir - i3*ne2*ne1 - i2*ne1);
+
+                ggml_fp16_t * dst_ptr  = (ggml_fp16_t *) ((char *) dst->data  + i3*nb3  + i2*nb2  + i1*nb1);
+                ggml_fp16_t * src0_ptr = (ggml_fp16_t *) ((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01);
+                float *       src1_ptr = (float *)       ((char *) src1->data + i3*nb13 + i2*nb12 + i1*nb11);
+
+                for (int i = 0; i < ne0; i++) {
+                    dst_ptr[i] = GGML_FP32_TO_FP16(GGML_FP16_TO_FP32(src0_ptr[i]) + src1_ptr[i]);
+                }
+            }
+        } else {
+            for (int ir = ir0; ir < ir1; ++ir) {
+                // src0, src1 and dst are same shape => same indices
+                const int i3 = ir/(ne2*ne1);
+                const int i2 = (ir - i3*ne2*ne1)/ne1;
+                const int i1 = (ir - i3*ne2*ne1 - i2*ne1);
+
+                float *       dst_ptr  = (float *)       ((char *) dst->data  + i3*nb3  + i2*nb2  + i1*nb1);
+                ggml_fp16_t * src0_ptr = (ggml_fp16_t *) ((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01);
+                float *       src1_ptr = (float *)       ((char *) src1->data + i3*nb13 + i2*nb12 + i1*nb11);
+
+                for (int i = 0; i < ne0; i++) {
+                    dst_ptr[i] = GGML_FP16_TO_FP32(src0_ptr[i]) + src1_ptr[i];
+                }
             }
         }
     }