graph : utilize ggml_build_forward_select() to avoid reallocations (#18898)

* graph : avoid branches between embedding and token inputs

* models : make deepstack graphs (e.g. Qwen3 VL) have constant topology

* ci : enable -DGGML_SCHED_NO_REALLOC=ON for server CI

* cont : pad token embeddings to n_embd_inp
This commit is contained in:
Georgi Gerganov authored and GitHub committed 2026-01-23 18:22:34 +02:00
1 parent cb6caca191
commit 557515be1e
7 files changed
+69 -53

No files matched your search

+2 -2
View File
@@ -72,7 +72,7 @@ jobs:
- name: Build
id: cmake_build
run: |
cmake -B build -DLLAMA_BUILD_BORINGSSL=ON
cmake -B build -DLLAMA_BUILD_BORINGSSL=ON -DGGML_SCHED_NO_REALLOC=ON
cmake --build build --config ${{ matrix.build_type }} -j ${env:NUMBER_OF_PROCESSORS} --target llama-server
- name: Python setup
@@ -108,7 +108,7 @@ jobs:
- name: Build
id: cmake_build
run: |
cmake -B build -DLLAMA_BUILD_BORINGSSL=ON
cmake -B build -DLLAMA_BUILD_BORINGSSL=ON -DGGML_SCHED_NO_REALLOC=ON
cmake --build build --config Release -j ${env:NUMBER_OF_PROCESSORS} --target llama-server
- name: Python setup