From: Morgan Smith <Morgan.J.Smith@outlook.com>
To: guix-patches@gnu.org
Cc: Morgan Smith <Morgan.J.Smith@outlook.com>
Subject: [PATCH 0/2] gnu: llama-cpp: Update to 0.0.0-b4882.
Date: Thu, 13 Mar 2025 15:42:37 -0400
| Package | Source(s) | Maintainer(s) | |
|---|---|---|---|
| guix-patches | PTS Buildd Popcon |
Report forwarded
to guix-patches@gnu.org:
bug#76999; Package guix-patches.
(Thu, 13 Mar 2025 19:49:02 GMT) (full text, mbox, link).
Acknowledgement sent
to Morgan Smith <Morgan.J.Smith@outlook.com>:
New bug report received and forwarded. Copy sent to guix-patches@gnu.org.
(Thu, 13 Mar 2025 19:49:02 GMT) (full text, mbox, link).
Message #5 received at submit@debbugs.gnu.org (full text, mbox, reply):
I was having some troubles running llama-cpp as it didn't have the ability to download things and the python scripts didn't seem to have their dependencies. This no longer installs "convert_hf_to_gguf_update.py" but that didn't work before this patch series anyways. Morgan Smith (2): gnu: Add python-gguf-llama-cpp. gnu: llama-cpp: Update to 0.0.0-b4882. gnu/local.mk | 1 - gnu/packages/machine-learning.scm | 49 +++++++++++++------ .../patches/llama-cpp-vulkan-optional.patch | 38 -------------- 3 files changed, 35 insertions(+), 53 deletions(-) delete mode 100644 gnu/packages/patches/llama-cpp-vulkan-optional.patch -- 2.48.1
Information forwarded
to guix-patches@gnu.org:
bug#76999; Package guix-patches.
(Thu, 13 Mar 2025 21:20:02 GMT) (full text, mbox, link).
Message #8 received at 76999@debbugs.gnu.org (full text, mbox, reply):
* gnu/packages/machine-learning.scm (python-gguf-llama-cpp): New variable.
Change-Id: I1c1b5f5956e3acb380b56816d180f53243b741fa
---
gnu/packages/machine-learning.scm | 15 +++++++++++++++
1 file changed, 15 insertions(+)
diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm
index 246b004156..ee5feb58fc 100644
--- a/gnu/packages/machine-learning.scm
+++ b/gnu/packages/machine-learning.scm
@@ -6490,6 +6490,21 @@ (define-public python-gguf
(description "A Python library for reading and writing GGUF & GGML format ML models.")
(license license:expat)))
+(define-public python-gguf-llama-cpp
+ (package/inherit python-gguf
+ (version "0.16.0")
+ (source (package-source llama-cpp))
+ (propagated-inputs (list python-numpy python-pyyaml python-sentencepiece
+ python-tqdm))
+ (native-inputs (list python-poetry-core))
+ (arguments
+ (substitute-keyword-arguments (package-arguments python-gguf)
+ ((#:phases phases #~%standard-phases)
+ #~(modify-phases #$phases
+ (add-after 'unpack 'chdir
+ (lambda _
+ (chdir "gguf-py")))))))))
+
(define-public python-gymnasium
(package
(name "python-gymnasium")
--
2.48.1
Information forwarded
to guix-patches@gnu.org:
bug#76999; Package guix-patches.
(Thu, 13 Mar 2025 21:21:01 GMT) (full text, mbox, link).
Message #11 received at 76999@debbugs.gnu.org (full text, mbox, reply):
* gnu/packages/machine-learning.scm (llama-cpp): Update to 0.0.0-b4882.
[inputs]: Add curl, glslang, and python-gguf-llama-cpp.
[native-inputs]: bash -> bash-minimal.
[source, homepage]: Update URL.
[python-scripts]: Check that we can run them.
[fix-tests]: Fix an additional test.
* gnu/packages/patches/llama-cpp-vulkan-optional.patch: Delete.
* gnu/local.mk: Unregister patch.
Change-Id: Ic297534cd142cb83e3964eae21b4eb807b74e9bc
---
gnu/local.mk | 1 -
gnu/packages/machine-learning.scm | 41 +++++++++++--------
.../patches/llama-cpp-vulkan-optional.patch | 38 -----------------
3 files changed, 25 insertions(+), 55 deletions(-)
delete mode 100644 gnu/packages/patches/llama-cpp-vulkan-optional.patch
diff --git a/gnu/local.mk b/gnu/local.mk
index 5425095e1d..dcff631515 100644
--- a/gnu/local.mk
+++ b/gnu/local.mk
@@ -1841,7 +1841,6 @@ dist_patch_DATA = \
%D%/packages/patches/mcrypt-CVE-2012-4527.patch \
%D%/packages/patches/libmemcached-build-with-gcc7.patch \
%D%/packages/patches/libmhash-hmac-fix-uaf.patch \
- %D%/packages/patches/llama-cpp-vulkan-optional.patch \
%D%/packages/patches/llhttp-ponyfill-object-fromentries.patch \
%D%/packages/patches/lvm2-no-systemd.patch \
%D%/packages/patches/maturin-no-cross-compile.patch \
diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm
index ee5feb58fc..b173f54fec 100644
--- a/gnu/packages/machine-learning.scm
+++ b/gnu/packages/machine-learning.scm
@@ -77,6 +77,7 @@ (define-module (gnu packages machine-learning)
#:use-module (gnu packages cmake)
#:use-module (gnu packages cpp)
#:use-module (gnu packages cran)
+ #:use-module (gnu packages curl)
#:use-module (gnu packages databases)
#:use-module (gnu packages dejagnu)
#:use-module (gnu packages documentation)
@@ -585,7 +586,7 @@ (define-public guile-aiscm-next
(deprecated-package "guile-aiscm-next" guile-aiscm))
(define-public llama-cpp
- (let ((tag "b4549"))
+ (let ((tag "b4882"))
(package
(name "llama-cpp")
(version (string-append "0.0.0-" tag))
@@ -593,19 +594,19 @@ (define-public llama-cpp
(origin
(method git-fetch)
(uri (git-reference
- (url "https://github.com/ggerganov/llama.cpp")
+ (url "https://github.com/ggml-org/llama.cpp")
(commit tag)))
(file-name (git-file-name name tag))
(sha256
- (base32 "1xf2579q0r8nv06kj8padi6w9cv30w58vdys65nq8yzm3dy452a1"))
- (patches
- (search-patches "llama-cpp-vulkan-optional.patch"))))
+ (base32 "1mhh4293lgvyvyq58hpphqk18n5g2zadafpdf9icf7xlj0cf7bqc"))))
(build-system cmake-build-system)
(arguments
(list
#:configure-flags
- #~(list "-DBUILD_SHARED_LIBS=ON"
+ #~(list #$(string-append "-DGGML_BUILD_NUMBER=" tag)
+ "-DBUILD_SHARED_LIBS=ON"
"-DGGML_VULKAN=ON"
+ "-DLLAMA_CURL=ON"
"-DGGML_BLAS=ON"
"-DGGML_BLAS_VENDOR=OpenBLAS"
(string-append "-DBLAS_INCLUDE_DIRS="
@@ -635,13 +636,16 @@ (define-public llama-cpp
(substitute* "ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp"
(("\"/bin/sh\"")
(string-append "\"" (search-input-file inputs "/bin/sh") "\"")))))
- (add-after 'unpack 'disable-unrunable-tests
+ (add-after 'unpack 'fix-tests
(lambda _
;; test-eval-callback downloads ML model from network, cannot
;; run in Guix build environment
(substitute* '("examples/eval-callback/CMakeLists.txt")
(("COMMAND llama-eval-callback")
- "COMMAND true llama-eval-callback"))))
+ "COMMAND true llama-eval-callback"))
+ ;; Help it find the test files it needs
+ (substitute* "tests/test-chat.cpp"
+ (("\"\\.\\./\"") "\"../source/\""))))
(add-before 'install 'install-python-scripts
(lambda _
(let ((bin (string-append #$output "/bin/")))
@@ -657,23 +661,28 @@ (define-public llama-cpp
(get-string-all input))))))
(chmod (string-append bin script) #o555)))
(mkdir-p bin)
- (make-script "convert_hf_to_gguf")
- (make-script "convert_llama_ggml_to_gguf")
- (make-script "convert_hf_to_gguf_update.py"))))
- (add-after 'install-python-scripts 'wrap-python-scripts
- (assoc-ref python:%standard-phases 'wrap))
+ (for-each
+ (lambda (file)
+ (make-script file)
+ ;; Run script as a sanity check
+ (invoke (string-append bin file) "-h"))
+ '(;; involves adding python-transformers package which looks involved.
+ ;; "convert_hf_to_gguf_update.py"
+ "convert_hf_to_gguf"
+ "convert_llama_ggml_to_gguf")))))
(add-after 'install 'remove-tests
(lambda* (#:key outputs #:allow-other-keys)
(for-each delete-file (find-files
(string-append (assoc-ref outputs "out")
"/bin")
"^test-")))))))
- (inputs (list python vulkan-headers vulkan-loader))
- (native-inputs (list pkg-config shaderc bash))
+ (inputs (list curl glslang python python-gguf-llama-cpp
+ vulkan-headers vulkan-loader))
+ (native-inputs (list pkg-config shaderc bash-minimal))
(propagated-inputs
(list python-numpy python-pytorch python-sentencepiece openblas))
(properties '((tunable? . #true))) ;use AVX512, FMA, etc. when available
- (home-page "https://github.com/ggerganov/llama.cpp")
+ (home-page "https://github.com/ggml-org/llama.cpp")
(synopsis "Port of Facebook's LLaMA model in C/C++")
(description "This package provides a port to Facebook's LLaMA collection
of foundation language models. It requires models parameters to be downloaded
diff --git a/gnu/packages/patches/llama-cpp-vulkan-optional.patch b/gnu/packages/patches/llama-cpp-vulkan-optional.patch
deleted file mode 100644
index 43a49b6a02..0000000000
--- a/gnu/packages/patches/llama-cpp-vulkan-optional.patch
+++ /dev/null
@@ -1,38 +0,0 @@
-Author: Danny Milosavljevic <dannym@friendly-machines.com>
-Date: 2025-01-29
-License: Expat
-Subject: Make Vulkan optional
-
-See also: <https://github.com/ggerganov/llama.cpp/pull/11494>
-
-diff -ru orig/llama.cpp/ggml/include/ggml-vulkan.h llama.cpp/ggml/include/ggml-vulkan.h
---- orig/llama.cpp/ggml/include/ggml-vulkan.h 2025-01-29 10:24:10.894476682 +0100
-+++ llama.cpp/ggml/include/ggml-vulkan.h 2025-02-07 18:28:34.509509638 +0100
-@@ -10,8 +10,6 @@
- #define GGML_VK_NAME "Vulkan"
- #define GGML_VK_MAX_DEVICES 16
-
--GGML_BACKEND_API void ggml_vk_instance_init(void);
--
- // backend API
- GGML_BACKEND_API ggml_backend_t ggml_backend_vk_init(size_t dev_num);
-
-diff -ru orig/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp
---- orig/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp 2025-01-29 10:24:10.922476480 +0100
-+++ llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp 2025-01-29 22:33:19.955087552 +0100
-@@ -8174,8 +8174,13 @@
- /* .iface = */ ggml_backend_vk_reg_i,
- /* .context = */ nullptr,
- };
--
-- return ®
-+ try {
-+ ggml_vk_instance_init();
-+ return ®
-+ } catch (const vk::SystemError& e) {
-+ VK_LOG_DEBUG("ggml_vk_get_device_count() -> Error: System error: " << e.what());
-+ return nullptr;
-+ }
- }
-
- // Extension availability
--
2.48.1
Information forwarded
to guix-patches@gnu.org:
bug#76999; Package guix-patches.
(Mon, 31 Mar 2025 22:52:01 GMT) (full text, mbox, link).
Message #14 received at 76999@debbugs.gnu.org (full text, mbox, reply):
* gnu/packages/machine-learning.scm (python-gguf-llama-cpp): New variable.
Change-Id: I1c1b5f5956e3acb380b56816d180f53243b741fa
---
gnu/packages/machine-learning.scm | 15 +++++++++++++++
1 file changed, 15 insertions(+)
diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm
index 7fdf5f37ee..7cb807ae91 100644
--- a/gnu/packages/machine-learning.scm
+++ b/gnu/packages/machine-learning.scm
@@ -6544,6 +6544,21 @@ (define-public python-gguf
(description "A Python library for reading and writing GGUF & GGML format ML models.")
(license license:expat)))
+(define-public python-gguf-llama-cpp
+ (package/inherit python-gguf
+ (version "0.16.0")
+ (source (package-source llama-cpp))
+ (propagated-inputs (list python-numpy python-pyyaml python-sentencepiece
+ python-tqdm))
+ (native-inputs (list python-poetry-core))
+ (arguments
+ (substitute-keyword-arguments (package-arguments python-gguf)
+ ((#:phases phases #~%standard-phases)
+ #~(modify-phases #$phases
+ (add-after 'unpack 'chdir
+ (lambda _
+ (chdir "gguf-py")))))))))
+
(define-public python-gymnasium
(package
(name "python-gymnasium")
base-commit: e2c2f98edd5d64921678c2570439dedfe662b1f8
--
2.49.0
Information forwarded
to guix-patches@gnu.org:
bug#76999; Package guix-patches.
(Mon, 31 Mar 2025 22:52:02 GMT) (full text, mbox, link).
Message #17 received at 76999@debbugs.gnu.org (full text, mbox, reply):
* gnu/packages/machine-learning.scm (llama-cpp): Update to 0.0.0-b5013.
[inputs]: Add curl, glslang, and python-gguf-llama-cpp.
[native-inputs]: bash -> bash-minimal.
[source, homepage]: Update URL.
[python-scripts]: Rely on upstream to install them. Delete phase.
[fix-tests]: Fix an additional test.
* gnu/packages/patches/llama-cpp-vulkan-optional.patch: Delete.
* gnu/local.mk: Unregister patch.
Change-Id: Ic297534cd142cb83e3964eae21b4eb807b74e9bc
---
gnu/local.mk | 1 -
gnu/packages/machine-learning.scm | 47 +++++++------------
.../patches/llama-cpp-vulkan-optional.patch | 38 ---------------
3 files changed, 17 insertions(+), 69 deletions(-)
delete mode 100644 gnu/packages/patches/llama-cpp-vulkan-optional.patch
diff --git a/gnu/local.mk b/gnu/local.mk
index f03fcb14fc..00b1a7a959 100644
--- a/gnu/local.mk
+++ b/gnu/local.mk
@@ -1845,7 +1845,6 @@ dist_patch_DATA = \
%D%/packages/patches/libmhash-hmac-fix-uaf.patch \
%D%/packages/patches/libmodbus-disable-networking-test.patch \
%D%/packages/patches/lib-tl-for-telegram-memcpy.patch \
- %D%/packages/patches/llama-cpp-vulkan-optional.patch \
%D%/packages/patches/llhttp-ponyfill-object-fromentries.patch \
%D%/packages/patches/lvm2-no-systemd.patch \
%D%/packages/patches/maturin-no-cross-compile.patch \
diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm
index 7cb807ae91..84be26cf35 100644
--- a/gnu/packages/machine-learning.scm
+++ b/gnu/packages/machine-learning.scm
@@ -78,6 +78,7 @@ (define-module (gnu packages machine-learning)
#:use-module (gnu packages cmake)
#:use-module (gnu packages cpp)
#:use-module (gnu packages cran)
+ #:use-module (gnu packages curl)
#:use-module (gnu packages databases)
#:use-module (gnu packages dejagnu)
#:use-module (gnu packages documentation)
@@ -634,7 +635,7 @@ (define-public guile-aiscm-next
(deprecated-package "guile-aiscm-next" guile-aiscm))
(define-public llama-cpp
- (let ((tag "b4549"))
+ (let ((tag "b5013"))
(package
(name "llama-cpp")
(version (string-append "0.0.0-" tag))
@@ -642,19 +643,19 @@ (define-public llama-cpp
(origin
(method git-fetch)
(uri (git-reference
- (url "https://github.com/ggerganov/llama.cpp")
+ (url "https://github.com/ggml-org/llama.cpp")
(commit tag)))
(file-name (git-file-name name tag))
(sha256
- (base32 "1xf2579q0r8nv06kj8padi6w9cv30w58vdys65nq8yzm3dy452a1"))
- (patches
- (search-patches "llama-cpp-vulkan-optional.patch"))))
+ (base32 "0s73dz871x53dr366lkzq19f677bwgma2ri8m5vhbfa9p8yp4p3r"))))
(build-system cmake-build-system)
(arguments
(list
#:configure-flags
- #~(list "-DBUILD_SHARED_LIBS=ON"
+ #~(list #$(string-append "-DGGML_BUILD_NUMBER=" tag)
+ "-DBUILD_SHARED_LIBS=ON"
"-DGGML_VULKAN=ON"
+ "-DLLAMA_CURL=ON"
"-DGGML_BLAS=ON"
"-DGGML_BLAS_VENDOR=OpenBLAS"
(string-append "-DBLAS_INCLUDE_DIRS="
@@ -684,32 +685,17 @@ (define-public llama-cpp
(substitute* "ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp"
(("\"/bin/sh\"")
(string-append "\"" (search-input-file inputs "/bin/sh") "\"")))))
- (add-after 'unpack 'disable-unrunable-tests
+ (add-after 'unpack 'fix-tests
(lambda _
;; test-eval-callback downloads ML model from network, cannot
;; run in Guix build environment
(substitute* '("examples/eval-callback/CMakeLists.txt")
(("COMMAND llama-eval-callback")
- "COMMAND true llama-eval-callback"))))
- (add-before 'install 'install-python-scripts
- (lambda _
- (let ((bin (string-append #$output "/bin/")))
- (define (make-script script)
- (let ((suffix (if (string-suffix? ".py" script) "" ".py")))
- (call-with-input-file
- (string-append "../source/" script suffix)
- (lambda (input)
- (call-with-output-file (string-append bin script)
- (lambda (output)
- (format output "#!~a/bin/python3\n~a"
- #$(this-package-input "python")
- (get-string-all input))))))
- (chmod (string-append bin script) #o555)))
- (mkdir-p bin)
- (make-script "convert_hf_to_gguf")
- (make-script "convert_llama_ggml_to_gguf")
- (make-script "convert_hf_to_gguf_update.py"))))
- (add-after 'install-python-scripts 'wrap-python-scripts
+ "COMMAND true llama-eval-callback"))
+ ;; Help it find the test files it needs
+ (substitute* "tests/test-chat.cpp"
+ (("\"\\.\\./\"") "\"../source/\""))))
+ (add-after 'install 'wrap-python-scripts
(assoc-ref python:%standard-phases 'wrap))
(add-after 'install 'remove-tests
(lambda* (#:key outputs #:allow-other-keys)
@@ -717,12 +703,13 @@ (define-public llama-cpp
(string-append (assoc-ref outputs "out")
"/bin")
"^test-")))))))
- (inputs (list python vulkan-headers vulkan-loader))
- (native-inputs (list pkg-config shaderc bash))
+ (inputs (list curl glslang python python-gguf-llama-cpp
+ vulkan-headers vulkan-loader))
+ (native-inputs (list pkg-config shaderc bash-minimal))
(propagated-inputs
(list python-numpy python-pytorch python-sentencepiece openblas))
(properties '((tunable? . #true))) ;use AVX512, FMA, etc. when available
- (home-page "https://github.com/ggerganov/llama.cpp")
+ (home-page "https://github.com/ggml-org/llama.cpp")
(synopsis "Port of Facebook's LLaMA model in C/C++")
(description "This package provides a port to Facebook's LLaMA collection
of foundation language models. It requires models parameters to be downloaded
diff --git a/gnu/packages/patches/llama-cpp-vulkan-optional.patch b/gnu/packages/patches/llama-cpp-vulkan-optional.patch
deleted file mode 100644
index 43a49b6a02..0000000000
--- a/gnu/packages/patches/llama-cpp-vulkan-optional.patch
+++ /dev/null
@@ -1,38 +0,0 @@
-Author: Danny Milosavljevic <dannym@friendly-machines.com>
-Date: 2025-01-29
-License: Expat
-Subject: Make Vulkan optional
-
-See also: <https://github.com/ggerganov/llama.cpp/pull/11494>
-
-diff -ru orig/llama.cpp/ggml/include/ggml-vulkan.h llama.cpp/ggml/include/ggml-vulkan.h
---- orig/llama.cpp/ggml/include/ggml-vulkan.h 2025-01-29 10:24:10.894476682 +0100
-+++ llama.cpp/ggml/include/ggml-vulkan.h 2025-02-07 18:28:34.509509638 +0100
-@@ -10,8 +10,6 @@
- #define GGML_VK_NAME "Vulkan"
- #define GGML_VK_MAX_DEVICES 16
-
--GGML_BACKEND_API void ggml_vk_instance_init(void);
--
- // backend API
- GGML_BACKEND_API ggml_backend_t ggml_backend_vk_init(size_t dev_num);
-
-diff -ru orig/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp
---- orig/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp 2025-01-29 10:24:10.922476480 +0100
-+++ llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp 2025-01-29 22:33:19.955087552 +0100
-@@ -8174,8 +8174,13 @@
- /* .iface = */ ggml_backend_vk_reg_i,
- /* .context = */ nullptr,
- };
--
-- return ®
-+ try {
-+ ggml_vk_instance_init();
-+ return ®
-+ } catch (const vk::SystemError& e) {
-+ VK_LOG_DEBUG("ggml_vk_get_device_count() -> Error: System error: " << e.what());
-+ return nullptr;
-+ }
- }
-
- // Extension availability
--
2.49.0
Information forwarded
to guix-patches@gnu.org:
bug#76999; Package guix-patches.
(Sun, 06 Apr 2025 16:16:01 GMT) (full text, mbox, link).
Message #20 received at 76999@debbugs.gnu.org (full text, mbox, reply):
[Message part 1 (text/plain, inline)]
Morgan Smith <Morgan.J.Smith@outlook.com> writes: > * gnu/packages/machine-learning.scm (python-gguf-llama-cpp): New variable. > > Change-Id: I1c1b5f5956e3acb380b56816d180f53243b741fa > --- > gnu/packages/machine-learning.scm | 15 +++++++++++++++ > 1 file changed, 15 insertions(+) > > diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm > index 246b004156..ee5feb58fc 100644 > --- a/gnu/packages/machine-learning.scm > +++ b/gnu/packages/machine-learning.scm > @@ -6490,6 +6490,21 @@ (define-public python-gguf > (description "A Python library for reading and writing GGUF & GGML format ML models.") > (license license:expat))) > > +(define-public python-gguf-llama-cpp > + (package/inherit python-gguf > + (version "0.16.0") > + (source (package-source llama-cpp)) > + (propagated-inputs (list python-numpy python-pyyaml python-sentencepiece > + python-tqdm)) > + (native-inputs (list python-poetry-core)) > + (arguments > + (substitute-keyword-arguments (package-arguments python-gguf) > + ((#:phases phases #~%standard-phases) > + #~(modify-phases #$phases > + (add-after 'unpack 'chdir > + (lambda _ > + (chdir "gguf-py"))))))))) > + > (define-public python-gymnasium > (package > (name "python-gymnasium") Can python-gguf be updated rather than adding this package?
[signature.asc (application/pgp-signature, inline)]
Information forwarded
to guix-patches@gnu.org:
bug#76999; Package guix-patches.
(Mon, 07 Apr 2025 23:11:02 GMT) (full text, mbox, link).
Message #23 received at 76999@debbugs.gnu.org (full text, mbox, reply):
Christopher Baines <mail@cbaines.net> writes: > > Can python-gguf be updated rather than adding this package? I had a tricky time tracking down the canonical source for the package. I'm fairly certain now that it is actually in the llama-cpp package. I assumed it was a bundling situation but it doesn't seem that way. So I will send a patch where I simply update python-gguf.
Information forwarded
to guix-patches@gnu.org:
bug#76999; Package guix-patches.
(Mon, 07 Apr 2025 23:12:01 GMT) (full text, mbox, link).
Message #26 received at 76999@debbugs.gnu.org (full text, mbox, reply):
* gnu/packages/machine-learning.scm (llama-cpp): Update to 0.0.0-b5013.
[inputs]: Add curl, glslang, and python-gguf.
[native-inputs]: bash -> bash-minimal.
[source, homepage]: Update URL.
[python-scripts]: Rely on upstream to install them. Delete phase.
[fix-tests]: Fix an additional test.
* gnu/packages/patches/llama-cpp-vulkan-optional.patch: Delete.
* gnu/local.mk: Unregister patch.
Change-Id: Ic297534cd142cb83e3964eae21b4eb807b74e9bc
---
gnu/local.mk | 1 -
gnu/packages/machine-learning.scm | 47 +++++++------------
.../patches/llama-cpp-vulkan-optional.patch | 38 ---------------
3 files changed, 17 insertions(+), 69 deletions(-)
delete mode 100644 gnu/packages/patches/llama-cpp-vulkan-optional.patch
diff --git a/gnu/local.mk b/gnu/local.mk
index 6dc4b4f61b..65c21c2f0d 100644
--- a/gnu/local.mk
+++ b/gnu/local.mk
@@ -1843,7 +1843,6 @@ dist_patch_DATA = \
%D%/packages/patches/libmhash-hmac-fix-uaf.patch \
%D%/packages/patches/libmodbus-disable-networking-test.patch \
%D%/packages/patches/lib-tl-for-telegram-memcpy.patch \
- %D%/packages/patches/llama-cpp-vulkan-optional.patch \
%D%/packages/patches/llhttp-ponyfill-object-fromentries.patch \
%D%/packages/patches/lvm2-no-systemd.patch \
%D%/packages/patches/maturin-no-cross-compile.patch \
diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm
index bd7a4fd81b..0b9ee4fa39 100644
--- a/gnu/packages/machine-learning.scm
+++ b/gnu/packages/machine-learning.scm
@@ -84,6 +84,7 @@ (define-module (gnu packages machine-learning)
#:use-module (gnu packages crates-io)
#:use-module (gnu packages crates-tls)
#:use-module (gnu packages crates-web)
+ #:use-module (gnu packages curl)
#:use-module (gnu packages databases)
#:use-module (gnu packages dejagnu)
#:use-module (gnu packages documentation)
@@ -640,7 +641,7 @@ (define-public guile-aiscm-next
(deprecated-package "guile-aiscm-next" guile-aiscm))
(define-public llama-cpp
- (let ((tag "b4549"))
+ (let ((tag "b5013"))
(package
(name "llama-cpp")
(version (string-append "0.0.0-" tag))
@@ -648,19 +649,19 @@ (define-public llama-cpp
(origin
(method git-fetch)
(uri (git-reference
- (url "https://github.com/ggerganov/llama.cpp")
+ (url "https://github.com/ggml-org/llama.cpp")
(commit tag)))
(file-name (git-file-name name tag))
(sha256
- (base32 "1xf2579q0r8nv06kj8padi6w9cv30w58vdys65nq8yzm3dy452a1"))
- (patches
- (search-patches "llama-cpp-vulkan-optional.patch"))))
+ (base32 "0s73dz871x53dr366lkzq19f677bwgma2ri8m5vhbfa9p8yp4p3r"))))
(build-system cmake-build-system)
(arguments
(list
#:configure-flags
- #~(list "-DBUILD_SHARED_LIBS=ON"
+ #~(list #$(string-append "-DGGML_BUILD_NUMBER=" tag)
+ "-DBUILD_SHARED_LIBS=ON"
"-DGGML_VULKAN=ON"
+ "-DLLAMA_CURL=ON"
"-DGGML_BLAS=ON"
"-DGGML_BLAS_VENDOR=OpenBLAS"
(string-append "-DBLAS_INCLUDE_DIRS="
@@ -690,32 +691,17 @@ (define-public llama-cpp
(substitute* "ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp"
(("\"/bin/sh\"")
(string-append "\"" (search-input-file inputs "/bin/sh") "\"")))))
- (add-after 'unpack 'disable-unrunable-tests
+ (add-after 'unpack 'fix-tests
(lambda _
;; test-eval-callback downloads ML model from network, cannot
;; run in Guix build environment
(substitute* '("examples/eval-callback/CMakeLists.txt")
(("COMMAND llama-eval-callback")
- "COMMAND true llama-eval-callback"))))
- (add-before 'install 'install-python-scripts
- (lambda _
- (let ((bin (string-append #$output "/bin/")))
- (define (make-script script)
- (let ((suffix (if (string-suffix? ".py" script) "" ".py")))
- (call-with-input-file
- (string-append "../source/" script suffix)
- (lambda (input)
- (call-with-output-file (string-append bin script)
- (lambda (output)
- (format output "#!~a/bin/python3\n~a"
- #$(this-package-input "python")
- (get-string-all input))))))
- (chmod (string-append bin script) #o555)))
- (mkdir-p bin)
- (make-script "convert_hf_to_gguf")
- (make-script "convert_llama_ggml_to_gguf")
- (make-script "convert_hf_to_gguf_update.py"))))
- (add-after 'install-python-scripts 'wrap-python-scripts
+ "COMMAND true llama-eval-callback"))
+ ;; Help it find the test files it needs
+ (substitute* "tests/test-chat.cpp"
+ (("\"\\.\\./\"") "\"../source/\""))))
+ (add-after 'install 'wrap-python-scripts
(assoc-ref python:%standard-phases 'wrap))
(add-after 'install 'remove-tests
(lambda* (#:key outputs #:allow-other-keys)
@@ -723,12 +709,13 @@ (define-public llama-cpp
(string-append (assoc-ref outputs "out")
"/bin")
"^test-")))))))
- (inputs (list python vulkan-headers vulkan-loader))
- (native-inputs (list pkg-config shaderc bash))
+ (inputs (list curl glslang python python-gguf
+ vulkan-headers vulkan-loader))
+ (native-inputs (list pkg-config shaderc bash-minimal))
(propagated-inputs
(list python-numpy python-pytorch python-sentencepiece openblas))
(properties '((tunable? . #true))) ;use AVX512, FMA, etc. when available
- (home-page "https://github.com/ggerganov/llama.cpp")
+ (home-page "https://github.com/ggml-org/llama.cpp")
(synopsis "Port of Facebook's LLaMA model in C/C++")
(description "This package provides a port to Facebook's LLaMA collection
of foundation language models. It requires models parameters to be downloaded
diff --git a/gnu/packages/patches/llama-cpp-vulkan-optional.patch b/gnu/packages/patches/llama-cpp-vulkan-optional.patch
deleted file mode 100644
index 43a49b6a02..0000000000
--- a/gnu/packages/patches/llama-cpp-vulkan-optional.patch
+++ /dev/null
@@ -1,38 +0,0 @@
-Author: Danny Milosavljevic <dannym@friendly-machines.com>
-Date: 2025-01-29
-License: Expat
-Subject: Make Vulkan optional
-
-See also: <https://github.com/ggerganov/llama.cpp/pull/11494>
-
-diff -ru orig/llama.cpp/ggml/include/ggml-vulkan.h llama.cpp/ggml/include/ggml-vulkan.h
---- orig/llama.cpp/ggml/include/ggml-vulkan.h 2025-01-29 10:24:10.894476682 +0100
-+++ llama.cpp/ggml/include/ggml-vulkan.h 2025-02-07 18:28:34.509509638 +0100
-@@ -10,8 +10,6 @@
- #define GGML_VK_NAME "Vulkan"
- #define GGML_VK_MAX_DEVICES 16
-
--GGML_BACKEND_API void ggml_vk_instance_init(void);
--
- // backend API
- GGML_BACKEND_API ggml_backend_t ggml_backend_vk_init(size_t dev_num);
-
-diff -ru orig/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp
---- orig/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp 2025-01-29 10:24:10.922476480 +0100
-+++ llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp 2025-01-29 22:33:19.955087552 +0100
-@@ -8174,8 +8174,13 @@
- /* .iface = */ ggml_backend_vk_reg_i,
- /* .context = */ nullptr,
- };
--
-- return ®
-+ try {
-+ ggml_vk_instance_init();
-+ return ®
-+ } catch (const vk::SystemError& e) {
-+ VK_LOG_DEBUG("ggml_vk_get_device_count() -> Error: System error: " << e.what());
-+ return nullptr;
-+ }
- }
-
- // Extension availability
base-commit: 666a6cfd88b3e5106a9180e06ea128db8084be0e
prerequisite-patch-id: 1e2c478cf648ee8c9a3b1af55543e1b96ff24ec7
--
2.49.0
Information forwarded
to guix-patches@gnu.org:
bug#76999; Package guix-patches.
(Mon, 07 Apr 2025 23:21:01 GMT) (full text, mbox, link).
Message #29 received at 76999@debbugs.gnu.org (full text, mbox, reply):
* gnu/packages/machine-learning.scm (python-gguf): Update to 0.16.0.
Change origin to git repository. Run tests.
Change-Id: I1c1b5f5956e3acb380b56816d180f53243b741fa
---
gnu/packages/machine-learning.scm | 47 +++++++++++++++++++------------
1 file changed, 29 insertions(+), 18 deletions(-)
diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm
index 3e68af3476..bd7a4fd81b 100644
--- a/gnu/packages/machine-learning.scm
+++ b/gnu/packages/machine-learning.scm
@@ -7041,24 +7041,35 @@ (define-public oneapi-dnnl-for-r-torch
(base32 "1zyw5rd8x346bb7gac9a7x3saviw3zvp6aqz2z1l9sv163vmjfz6"))))))
(define-public python-gguf
- (package
- (name "python-gguf")
- (version "0.6.0")
- (source
- (origin
- (method url-fetch)
- (uri (pypi-uri "gguf" version))
- (sha256
- (base32 "0rbyc2h3kpqnrvbyjvv8a69l577jv55a31l12jnw21m1lamjxqmj"))))
- (build-system pyproject-build-system)
- (arguments
- (list #:tests? #false))
- (inputs (list poetry python-pytest))
- (propagated-inputs (list python-numpy))
- (home-page "https://ggml.ai")
- (synopsis "Read and write ML models in GGUF for GGML")
- (description "A Python library for reading and writing GGUF & GGML format ML models.")
- (license license:expat)))
+ ;; They didn't tag the commit
+ (let ((commit "69050a11be0ae3e01329f11371ecb6850bdaded5"))
+ (package
+ (name "python-gguf")
+ (version "0.16.0")
+ (source
+ (origin
+ (method git-fetch)
+ (uri (git-reference
+ (url "https://github.com/ggml-org/llama.cpp")
+ (commit commit)))
+ (file-name (git-file-name name commit))
+ (sha256
+ (base32 "1563mbrjykwpsbhghhzi4h1qv9qy74gq5vq4xhs58zk0jp20c7zz"))))
+ (build-system pyproject-build-system)
+ (arguments
+ (list
+ #:phases
+ #~(modify-phases %standard-phases
+ (add-after 'unpack 'chdir
+ (lambda _
+ (chdir "gguf-py"))))))
+ (propagated-inputs (list python-numpy python-pyyaml python-sentencepiece
+ python-tqdm))
+ (native-inputs (list python-poetry-core python-pytest))
+ (home-page "https://ggml.ai")
+ (synopsis "Read and write ML models in GGUF for GGML")
+ (description "A Python library for reading and writing GGUF & GGML format ML models.")
+ (license license:expat))))
(define-public python-gymnasium
(package
base-commit: 666a6cfd88b3e5106a9180e06ea128db8084be0e
--
2.49.0
Information forwarded
to guix-patches@gnu.org:
bug#76999; Package guix-patches.
(Mon, 07 Apr 2025 23:21:02 GMT) (full text, mbox, link).
Message #32 received at 76999@debbugs.gnu.org (full text, mbox, reply):
* gnu/packages/machine-learning.scm (llama-cpp): Update to 0.0.0-b5013.
[inputs]: Add curl, glslang, and python-gguf.
[native-inputs]: bash -> bash-minimal.
[source, homepage]: Update URL.
[python-scripts]: Rely on upstream to install them. Delete phase.
[fix-tests]: Fix an additional test.
* gnu/packages/patches/llama-cpp-vulkan-optional.patch: Delete.
* gnu/local.mk: Unregister patch.
Change-Id: Ic297534cd142cb83e3964eae21b4eb807b74e9bc
---
gnu/local.mk | 1 -
gnu/packages/machine-learning.scm | 47 +++++++------------
.../patches/llama-cpp-vulkan-optional.patch | 38 ---------------
3 files changed, 17 insertions(+), 69 deletions(-)
delete mode 100644 gnu/packages/patches/llama-cpp-vulkan-optional.patch
diff --git a/gnu/local.mk b/gnu/local.mk
index 6dc4b4f61b..65c21c2f0d 100644
--- a/gnu/local.mk
+++ b/gnu/local.mk
@@ -1843,7 +1843,6 @@ dist_patch_DATA = \
%D%/packages/patches/libmhash-hmac-fix-uaf.patch \
%D%/packages/patches/libmodbus-disable-networking-test.patch \
%D%/packages/patches/lib-tl-for-telegram-memcpy.patch \
- %D%/packages/patches/llama-cpp-vulkan-optional.patch \
%D%/packages/patches/llhttp-ponyfill-object-fromentries.patch \
%D%/packages/patches/lvm2-no-systemd.patch \
%D%/packages/patches/maturin-no-cross-compile.patch \
diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm
index bd7a4fd81b..0b9ee4fa39 100644
--- a/gnu/packages/machine-learning.scm
+++ b/gnu/packages/machine-learning.scm
@@ -84,6 +84,7 @@ (define-module (gnu packages machine-learning)
#:use-module (gnu packages crates-io)
#:use-module (gnu packages crates-tls)
#:use-module (gnu packages crates-web)
+ #:use-module (gnu packages curl)
#:use-module (gnu packages databases)
#:use-module (gnu packages dejagnu)
#:use-module (gnu packages documentation)
@@ -640,7 +641,7 @@ (define-public guile-aiscm-next
(deprecated-package "guile-aiscm-next" guile-aiscm))
(define-public llama-cpp
- (let ((tag "b4549"))
+ (let ((tag "b5013"))
(package
(name "llama-cpp")
(version (string-append "0.0.0-" tag))
@@ -648,19 +649,19 @@ (define-public llama-cpp
(origin
(method git-fetch)
(uri (git-reference
- (url "https://github.com/ggerganov/llama.cpp")
+ (url "https://github.com/ggml-org/llama.cpp")
(commit tag)))
(file-name (git-file-name name tag))
(sha256
- (base32 "1xf2579q0r8nv06kj8padi6w9cv30w58vdys65nq8yzm3dy452a1"))
- (patches
- (search-patches "llama-cpp-vulkan-optional.patch"))))
+ (base32 "0s73dz871x53dr366lkzq19f677bwgma2ri8m5vhbfa9p8yp4p3r"))))
(build-system cmake-build-system)
(arguments
(list
#:configure-flags
- #~(list "-DBUILD_SHARED_LIBS=ON"
+ #~(list #$(string-append "-DGGML_BUILD_NUMBER=" tag)
+ "-DBUILD_SHARED_LIBS=ON"
"-DGGML_VULKAN=ON"
+ "-DLLAMA_CURL=ON"
"-DGGML_BLAS=ON"
"-DGGML_BLAS_VENDOR=OpenBLAS"
(string-append "-DBLAS_INCLUDE_DIRS="
@@ -690,32 +691,17 @@ (define-public llama-cpp
(substitute* "ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp"
(("\"/bin/sh\"")
(string-append "\"" (search-input-file inputs "/bin/sh") "\"")))))
- (add-after 'unpack 'disable-unrunable-tests
+ (add-after 'unpack 'fix-tests
(lambda _
;; test-eval-callback downloads ML model from network, cannot
;; run in Guix build environment
(substitute* '("examples/eval-callback/CMakeLists.txt")
(("COMMAND llama-eval-callback")
- "COMMAND true llama-eval-callback"))))
- (add-before 'install 'install-python-scripts
- (lambda _
- (let ((bin (string-append #$output "/bin/")))
- (define (make-script script)
- (let ((suffix (if (string-suffix? ".py" script) "" ".py")))
- (call-with-input-file
- (string-append "../source/" script suffix)
- (lambda (input)
- (call-with-output-file (string-append bin script)
- (lambda (output)
- (format output "#!~a/bin/python3\n~a"
- #$(this-package-input "python")
- (get-string-all input))))))
- (chmod (string-append bin script) #o555)))
- (mkdir-p bin)
- (make-script "convert_hf_to_gguf")
- (make-script "convert_llama_ggml_to_gguf")
- (make-script "convert_hf_to_gguf_update.py"))))
- (add-after 'install-python-scripts 'wrap-python-scripts
+ "COMMAND true llama-eval-callback"))
+ ;; Help it find the test files it needs
+ (substitute* "tests/test-chat.cpp"
+ (("\"\\.\\./\"") "\"../source/\""))))
+ (add-after 'install 'wrap-python-scripts
(assoc-ref python:%standard-phases 'wrap))
(add-after 'install 'remove-tests
(lambda* (#:key outputs #:allow-other-keys)
@@ -723,12 +709,13 @@ (define-public llama-cpp
(string-append (assoc-ref outputs "out")
"/bin")
"^test-")))))))
- (inputs (list python vulkan-headers vulkan-loader))
- (native-inputs (list pkg-config shaderc bash))
+ (inputs (list curl glslang python python-gguf
+ vulkan-headers vulkan-loader))
+ (native-inputs (list pkg-config shaderc bash-minimal))
(propagated-inputs
(list python-numpy python-pytorch python-sentencepiece openblas))
(properties '((tunable? . #true))) ;use AVX512, FMA, etc. when available
- (home-page "https://github.com/ggerganov/llama.cpp")
+ (home-page "https://github.com/ggml-org/llama.cpp")
(synopsis "Port of Facebook's LLaMA model in C/C++")
(description "This package provides a port to Facebook's LLaMA collection
of foundation language models. It requires models parameters to be downloaded
diff --git a/gnu/packages/patches/llama-cpp-vulkan-optional.patch b/gnu/packages/patches/llama-cpp-vulkan-optional.patch
deleted file mode 100644
index 43a49b6a02..0000000000
--- a/gnu/packages/patches/llama-cpp-vulkan-optional.patch
+++ /dev/null
@@ -1,38 +0,0 @@
-Author: Danny Milosavljevic <dannym@friendly-machines.com>
-Date: 2025-01-29
-License: Expat
-Subject: Make Vulkan optional
-
-See also: <https://github.com/ggerganov/llama.cpp/pull/11494>
-
-diff -ru orig/llama.cpp/ggml/include/ggml-vulkan.h llama.cpp/ggml/include/ggml-vulkan.h
---- orig/llama.cpp/ggml/include/ggml-vulkan.h 2025-01-29 10:24:10.894476682 +0100
-+++ llama.cpp/ggml/include/ggml-vulkan.h 2025-02-07 18:28:34.509509638 +0100
-@@ -10,8 +10,6 @@
- #define GGML_VK_NAME "Vulkan"
- #define GGML_VK_MAX_DEVICES 16
-
--GGML_BACKEND_API void ggml_vk_instance_init(void);
--
- // backend API
- GGML_BACKEND_API ggml_backend_t ggml_backend_vk_init(size_t dev_num);
-
-diff -ru orig/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp
---- orig/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp 2025-01-29 10:24:10.922476480 +0100
-+++ llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp 2025-01-29 22:33:19.955087552 +0100
-@@ -8174,8 +8174,13 @@
- /* .iface = */ ggml_backend_vk_reg_i,
- /* .context = */ nullptr,
- };
--
-- return ®
-+ try {
-+ ggml_vk_instance_init();
-+ return ®
-+ } catch (const vk::SystemError& e) {
-+ VK_LOG_DEBUG("ggml_vk_get_device_count() -> Error: System error: " << e.what());
-+ return nullptr;
-+ }
- }
-
- // Extension availability
--
2.49.0
Send a report that this bug log contains spam.
Debbugs is free software and licensed under the terms of the GNU Public License version 2. The current version can be obtained from https://bugs.debian.org/debbugs-source/.
Copyright © 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson, 2005-2017 Don Armstrong, and many other contributors.