Skip to content

Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

chore: update gitattributes (#8860)

Michael Yang5b446cc8150d
diff --git a/.gitattributes b/.gitattributes
index 4bcd95b0b..b1279845b 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -15,6 +15,10 @@ ml/backend/**/*.cu linguist-vendored
ml/backend/**/*.cuh linguist-vendored
ml/backend/**/*.m linguist-vendored
ml/backend/**/*.metal linguist-vendored
+ml/backend/**/CMakeLists.txt linguist-vendored
+
+llama/build-info.cpp linguist-generated
+ml/backend/ggml/ggml/src/ggml-metal/ggml-metal-embed.s linguist-generated
* text=auto
*.go text eol=lf
diff --git a/.github/workflows/test.yaml b/.github/workflows/test.yaml
index f8e1cadf4..8af8812fb 100644
--- a/.github/workflows/test.yaml
+++ b/.github/workflows/test.yaml
@@ -163,5 +163,5 @@ jobs:
- uses: actions/checkout@v4
- name: Verify patches apply cleanly and do not change files
run: |
- make -f Makefile.sync clean checkout sync
+ make -f Makefile.sync clean sync
git diff --compact-summary --exit-code
diff --git a/Makefile.sync b/Makefile.sync
index 3001487de..007282746 100644
--- a/Makefile.sync
+++ b/Makefile.sync
@@ -15,7 +15,11 @@ help:
@echo " make -f $(lastword $(MAKEFILE_LIST)) clean sync"
.PHONY: sync
-sync: llama/llama.cpp ml/backend/ggml/ggml apply-patches
+sync: llama/build-info.cpp llama/llama.cpp ml/backend/ggml/ggml apply-patches
+
+.PHONY: llama/build-info.cpp
+llama/build-info.cpp: llama/build-info.cpp.in
+ sed -e 's|@FETCH_HEAD@|$(FETCH_HEAD)|' $< > $@
.PHONY: llama/llama.cpp
llama/llama.cpp: llama/vendor/ apply-patches
diff --git a/llama/build-info.cpp b/llama/build-info.cpp
index b2c1dba73..e169b926b 100644
--- a/llama/build-info.cpp
+++ b/llama/build-info.cpp
@@ -1,4 +1,4 @@
int LLAMA_BUILD_NUMBER = 0;
-char const *LLAMA_COMMIT = "ba1cb19cdd0d92e012e0f6e009e0620f854b6afd";
+char const *LLAMA_COMMIT = "46e3556e01b824e52395fb050b29804b6cff2a7c";
char const *LLAMA_COMPILER = "";
char const *LLAMA_BUILD_TARGET = "";
diff --git a/llama/build-info.cpp.in b/llama/build-info.cpp.in
new file mode 100644
index 000000000..07cbd0e4d
--- /dev/null
+++ b/llama/build-info.cpp.in
@@ -0,0 +1,4 @@
+int LLAMA_BUILD_NUMBER = 0;
+char const *LLAMA_COMMIT = "@FETCH_HEAD@";
+char const *LLAMA_COMPILER = "";
+char const *LLAMA_BUILD_TARGET = "";
diff --git a/ml/backend/ggml/ggml/src/ggml-cuda/template-instances/generate_cu_files.py b/ml/backend/ggml/ggml/src/ggml-cuda/template-instances/generate_cu_files.py
deleted file mode 100755
index d7874e6ea..000000000
--- a/ml/backend/ggml/ggml/src/ggml-cuda/template-instances/generate_cu_files.py
+++ /dev/null
@@ -1,77 +0,0 @@
-#!/usr/bin/env python3
-
-from glob import glob
-import os
-
-TYPES_KV = ["GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0", "GGML_TYPE_F16"]
-
-SOURCE_FATTN_VEC = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../fattn-vec-f{vkq_size}.cuh"
-
-DECL_FATTN_VEC_F{vkq_size}_CASE({head_size}, {type_k}, {type_v});
-"""
-
-SOURCE_FATTN_WMMA_START = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../fattn-wmma-f16.cuh"
-
-"""
-
-SOURCE_FATTN_WMMA_CASE = "DECL_FATTN_WMMA_F16_CASE({head_size}, {cols_per_block}, {kq_acc_t});\n"
-
-TYPES_MMQ = [
- "GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0",
- "GGML_TYPE_Q2_K", "GGML_TYPE_Q3_K", "GGML_TYPE_Q4_K", "GGML_TYPE_Q5_K", "GGML_TYPE_Q6_K",
- "GGML_TYPE_IQ2_XXS", "GGML_TYPE_IQ2_XS", "GGML_TYPE_IQ2_S", "GGML_TYPE_IQ3_XXS", "GGML_TYPE_IQ3_S",
- "GGML_TYPE_IQ1_S", "GGML_TYPE_IQ4_NL", "GGML_TYPE_IQ4_XS"
-]
-
-SOURCE_MMQ = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../mmq.cuh"
-
-DECL_MMQ_CASE({type});
-"""
-
-
-def get_short_name(long_quant_name):
- return long_quant_name.replace("GGML_TYPE_", "").lower()
-
-
-def get_head_sizes(type_k, type_v):
- if type_k == "GGML_TYPE_F16" and type_v == "GGML_TYPE_F16":
- return [64, 128, 256]
- if type_k == "GGML_TYPE_F16":
- return [64, 128]
- return [128]
-
-
-for filename in glob("*.cu"):
- os.remove(filename)
-
-for vkq_size in [16, 32]:
- for type_k in TYPES_KV:
- for type_v in TYPES_KV:
- for head_size in get_head_sizes(type_k, type_v):
- with open(f"fattn-vec-f{vkq_size}-instance-hs{head_size}-{get_short_name(type_k)}-{get_short_name(type_v)}.cu", "w") as f:
- f.write(SOURCE_FATTN_VEC.format(vkq_size=vkq_size, head_size=head_size, type_k=type_k, type_v=type_v))
-
-for kq_acc_t in ["half", "float"]:
- for cols_per_block in [8, 16, 32]:
- if kq_acc_t == "float" and cols_per_block == 8:
- continue
-
- with open(f"fattn-wmma-f16-instance-kq{kq_acc_t}-cpb{cols_per_block}.cu", "w") as f:
- f.write(SOURCE_FATTN_WMMA_START)
-
- for head_size in [64, 80, 96, 112, 128, 256]:
- if cols_per_block == 8 and head_size % 32 != 0: # wmma fragment is 8x32
- continue
- if kq_acc_t == "float" and cols_per_block == 32 and head_size == 256: # register spilling, bad performance
- continue
- f.write(SOURCE_FATTN_WMMA_CASE.format(kq_acc_t=kq_acc_t, cols_per_block=cols_per_block, head_size=head_size))
-
-for type in TYPES_MMQ:
- with open(f"mmq-instance-{get_short_name(type)}.cu", "w") as f:
- f.write(SOURCE_MMQ.format(type=type))