mirror of
https://github.com/LostRuins/koboldcpp.git
synced 2026-09-19 17:25:07 +02:00
Merge commit '7d701b59296bc6f5ba504d5a4cddaf416c449a15' into concedo_experimental
# Conflicts: # .github/workflows/python-type-check.yml # examples/pydantic_models_to_grammar.py # ggml/src/ggml-hexagon/ggml-hexagon.cpp # ggml/src/ggml-hexagon/htp/htp-ops.h # ggml/src/ggml-hexagon/htp/hvx-arith.h # ggml/src/ggml-hexagon/htp/main.c # ggml/src/ggml-hexagon/htp/unary-ops.c # ggml/src/ggml-hexagon/htp/unary-ops.h # ggml/src/ggml-opencl/ggml-opencl.cpp # ggml/src/ggml-opencl/kernels/conv2d.cl # ggml/src/ggml-opencl/kernels/conv2d_f16_f32.cl # scripts/jinja/jinja-tester.py # scripts/snapdragon/ggml-hexagon-profile.py # scripts/tool_bench.py # tests/CMakeLists.txt # tests/test-backend-ops.cpp # tests/test-recurrent-state-rollback.cpp
This commit is contained in:
+2
-2
@@ -839,7 +839,7 @@ ggml_tensor * clip_graph::build_attn(
|
||||
}
|
||||
|
||||
cur = ggml_flash_attn_ext(ctx0, q, k, v, kq_mask, kq_scale, 0.0f, 0.0f);
|
||||
ggml_flash_attn_ext_set_prec(cur, GGML_PREC_F32);
|
||||
ggml_prec_set_acc(cur, GGML_PREC_F32);
|
||||
if (sinks != nullptr) {
|
||||
ggml_flash_attn_ext_add_sinks(cur, sinks);
|
||||
}
|
||||
@@ -852,7 +852,7 @@ ggml_tensor * clip_graph::build_attn(
|
||||
|
||||
ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
|
||||
// F32 may not needed for vision encoders?
|
||||
// ggml_mul_mat_set_prec(kq, GGML_PREC_F32);
|
||||
// ggml_prec_set_acc(kq, GGML_PREC_F32);
|
||||
|
||||
kq = ggml_soft_max_ext(ctx0, kq, kq_mask, kq_scale, 0.0f);
|
||||
if (sinks != nullptr) {
|
||||
|
||||
Reference in New Issue
Block a user