llama_cpp 0.6.0 → 0.7.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/CHANGELOG.md +8 -0
- data/ext/llama_cpp/llama_cpp.cpp +49 -3
- data/ext/llama_cpp/src/ggml-cuda.cu +122 -72
- data/ext/llama_cpp/src/ggml-metal.m +4 -5
- data/ext/llama_cpp/src/ggml-metal.metal +9 -2
- data/ext/llama_cpp/src/ggml-opencl.cpp +119 -53
- data/ext/llama_cpp/src/ggml.c +755 -320
- data/ext/llama_cpp/src/ggml.h +13 -0
- data/ext/llama_cpp/src/k_quants.c +744 -2
- data/ext/llama_cpp/src/llama.cpp +779 -113
- data/ext/llama_cpp/src/llama.h +22 -6
- data/ext/llama_cpp/src/unicode.h +462 -0
- data/lib/llama_cpp/version.rb +2 -2
- data/lib/llama_cpp.rb +1 -1
- data/sig/llama_cpp.rbs +5 -0
- metadata +3 -2
data/ext/llama_cpp/src/ggml.h
CHANGED
@@ -401,10 +401,14 @@ extern "C" {
|
|
401
401
|
GGML_OP_CLAMP,
|
402
402
|
GGML_OP_CONV_1D,
|
403
403
|
GGML_OP_CONV_2D,
|
404
|
+
GGML_OP_CONV_TRANSPOSE_1D,
|
404
405
|
GGML_OP_CONV_TRANSPOSE_2D,
|
405
406
|
GGML_OP_POOL_1D,
|
406
407
|
GGML_OP_POOL_2D,
|
407
408
|
|
409
|
+
GGML_OP_CONV_1D_STAGE_0, // internal
|
410
|
+
GGML_OP_CONV_1D_STAGE_1, // internal
|
411
|
+
|
408
412
|
GGML_OP_UPSCALE, // nearest interpolate
|
409
413
|
|
410
414
|
GGML_OP_FLASH_ATTN,
|
@@ -1386,6 +1390,14 @@ extern "C" {
|
|
1386
1390
|
int s,
|
1387
1391
|
int d);
|
1388
1392
|
|
1393
|
+
GGML_API struct ggml_tensor * ggml_conv_transpose_1d(
|
1394
|
+
struct ggml_context * ctx,
|
1395
|
+
struct ggml_tensor * a,
|
1396
|
+
struct ggml_tensor * b,
|
1397
|
+
int s0,
|
1398
|
+
int p0,
|
1399
|
+
int d0);
|
1400
|
+
|
1389
1401
|
GGML_API struct ggml_tensor * ggml_conv_2d(
|
1390
1402
|
struct ggml_context * ctx,
|
1391
1403
|
struct ggml_tensor * a,
|
@@ -1759,6 +1771,7 @@ extern "C" {
|
|
1759
1771
|
GGML_OPT_NO_CONTEXT,
|
1760
1772
|
GGML_OPT_INVALID_WOLFE,
|
1761
1773
|
GGML_OPT_FAIL,
|
1774
|
+
GGML_OPT_CANCEL,
|
1762
1775
|
|
1763
1776
|
GGML_LINESEARCH_FAIL = -128,
|
1764
1777
|
GGML_LINESEARCH_MINIMUM_STEP,
|