blob: 7bbb9e4827e0abd5d8cae867ba1f1925a0afe8d5 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
|
(define-module (epistemia packages machine-learning)
#:use-module (guix packages)
#:use-module (guix build-system go)
#:use-module (guix download)
#:use-module (guix utils)
#:use-module (guix git-download)
#:use-module (guix gexp)
#:use-module ((guix licenses) #:prefix license:)
#:use-module (gnu packages base)
#:use-module (gnu packages golang-check)
#:use-module (gnu packages golang-web)
#:use-module (gnu packages golang-xyz)
#:use-module (gnu packages llvm)
#:use-module (gnu packages machine-learning)
#:use-module (gnu packages rocm)
#:use-module (gnu packages rocm-libs))
(define-public llama-cpp-rocm
(let ((tag "b9491"))
(package
(inherit llama-cpp)
(name "llama-cpp-rocm")
(version (string-append "0.0.0-" tag))
(source
(origin
(method git-fetch)
(uri (git-reference
(url "https://github.com/ggml-org/llama.cpp")
(commit tag)))
(file-name (git-file-name name tag))
(sha256
(base32 "096rpq10ranq9yn8sk4s10myb0b7qzmq3crvv1w7mqwk1nmg0vxq"))))
(arguments
(substitute-keyword-arguments (package-arguments llama-cpp)
((#:configure-flags flags #~'())
#~(append #$flags
'(;; CMAKE_HIP_FLAGS is to workaround regression in ROCm 7.2
"-DCMAKE_HIP_FLAGS=\"-mllvm --amdgpu-unroll-threshold-local=600\""
#$(string-append "-DGPU_TARGETS=" (current-amd-gpu-targets-string))
;; temporarily disable webui building until I figure out how to add second source for hf bucket
"-DLLAMA_USE_PREBUILT_UI=OFF" "-DLLAMA_BUILD_UI=OFF")))
((#:phases phases)
#~(modify-phases #$phases
;; mtp added test that also needs internet for downloading its model
(add-after 'fix-tests 'fix-tests-new
(lambda _
(substitute* "tests/CMakeLists.txt"
(("llama_build_and_test\\(test-state-restore-fragmented\\.cpp .*")
"")
(("set_tests_properties\\(test-state-restore-fragmented PROPERTIES.*")
""))
(substitute* "tests/CMakeLists.txt"
(("llama_build_and_test\\(test-recurrent-state-rollback\\.cpp .*")
"")
(("set_tests_properties\\(test-recurrent-state-rollback PROPERTIES.*")
""))
(substitute* "tests/CMakeLists.txt"
(("llama_build_and_test\\(test-save-load-state\\.cpp .*")
"")
(("set_tests_properties\\(test-save-load-state PROPERTIES.*")
""))))
(delete 'fix-python-shebang)))))
(inputs
(modify-inputs (package-inputs llama-cpp)
(replace "ggml" ggml-rocm))))))
(define-public ggml-rocm
(package
(inherit ggml)
(name "ggml-rocm")
;; (version "0.13.1")
(version "9ca88e98e6d9e84cf77959931130d38c5b56c19b")
(source
(origin
(method git-fetch)
(uri (git-reference
;; (url "https://github.com/ggml-org/ggml")
(url "https://git.vineetk.net/ggml")
;; (commit (string-append "v" version))
(commit version)))
(file-name (git-file-name name version))
(sha256
(base32 "0llmlnkcg8zrdj6v2wpl018iw6j0sp1fsg2if3w3zd91hv66jq5r"))))
(arguments
(substitute-keyword-arguments (package-arguments (@@ (gnu packages machine-learning) ggml))
((#:configure-flags flags #~'())
#~(append #$flags (list (string-append "-DCMAKE_HIP_COMPILER_ROCM_ROOT=" #$rocm-hip-runtime)
(string-append "-DCMAKE_HIP_FLAGS=--rocm-path=" #$rocm-hip-runtime
" --rocm-device-lib-path=" #$rocm-device-libs "/amdgcn/bitcode")
#$(string-append "-DGPU_TARGETS=" (current-amd-gpu-targets-string))
"-DGGML_CPU=ON"
"-DGGML_HIP=ON"
"-DGGML_HIP_ROCWMMA_FATTN=OFF"
"-DGGML_CUDA_FA_ALL_QUANTS=ON"
;; for stable-diffusion.cpp
"-DGGML_CPU_ALL_VARIANTS=ON"
"-DGGML_MAX_NAME=128")))))
(inputs
(modify-inputs (package-inputs ggml)
(append clang-rocm
lld-rocm
rocm-cmake
rocm-device-libs
rocm-hipcc
rocm-hip-runtime
rocr-runtime
hipblas
hipblas-common
rocblas)))))
|
