search

Jeffser / Alpaca

🦙 An Ollama client made with GTK4 and Adwaita
https://jeffser.com/alpaca
GNU General Public License v3.0
459 stars 45 forks source link

GFX Target Overriding #83

Open snoweuph opened 6 months ago

snoweuph commented 6 months ago

Is your feature request related to a problem? Please describe. My GPU isnt utilisied

Describe the solution you'd like The ability to Override GFX targets, like described in the ollama Docs: https://github.com/ollama/ollama/blob/main/docs/gpu.md#overrides

Describe alternatives you've considered Running Ollama Manually in a Toolbx

Additional context I got an RX 6650XT which has a target of: gfx1032, but the nearest supported one is: 1030, so the Override for me would be: HSA_OVERRIDE_GFX_VERSION="10.3.0"

Yes, I tried setting it as an Flatpak override, but that didnt work

Jeffser commented 5 months ago

Hi, sorry for not answering the issue, I was taking a break, I will work on this for today's update

snoweuph commented 5 months ago

Ohh nice, Thats pretty fast response, thanks ^^

Jeffser commented 5 months ago

By the way, now that I'm going to add overrides for the Ollama instance into the gui, is there any other override you think I should add?

Jeffser commented 5 months ago

I just added this feature, if you want any other override let me know

https://github.com/Jeffser/Alpaca/commit/fff3a68b29b267c402c5c6968bb5ef59d607502c

snoweuph commented 5 months ago

By the way, now that I'm going to add overrides for the Ollama instance into the gui, is there any other override you think I should add?

You should prolly add: CUDA_VISIBLE_DEVICES for GPU Selection on Cuda And for AMD ROCM you should add HIP_VISIBLE_DEVICES

snoweuph commented 5 months ago

I just added this feature, if you want any other override let me know

fff3a68

Cant wait to test it in the next release

snoweuph commented 5 months ago

Override doesnt seem to work; image Is there some kind of ollama log for the inside model?

Jeffser commented 5 months ago

Ollama should output everything as far as I know, do you know if Ollama has different binaries for cpu and gpu usage?

snoweuph commented 5 months ago

I dont know, but it says differnt things when running the installer

snoweuph commented 5 months ago

and they have different Docker images, for me: docker run --env HSA_OVERRIDE_GFX_VERSION=10.3.0 -d --device /dev/kfd --device /dev/dri -v ollama:/root/.ollama -p 11434:11434 --name ollama docker.io/ollama/ollama:rocm works

Jeffser commented 5 months ago

I see, it has a different tag so it's probably different

snoweuph commented 5 months ago

it is basicly the command for amd in docker from the doc + my overide

Jeffser commented 5 months ago

https://github.com/ollama/ollama/blob/main/docs/linux.md

It says here you need to install rocm, that could be it

snoweuph commented 5 months ago

I see, it has a different tag so it's probably different

I would exspect that to, or it is jsut named different, becuase it bundles the rocm library with itselve

snoweuph commented 5 months ago

https://github.com/ollama/ollama/blob/main/docs/linux.md

It says here you need to install rocm, that could be it

you will have to get all required libs inside of the flatpak, system libraries wont help

snoweuph commented 5 months ago

I thought you allready had all needed libs in the flatpak

Jeffser commented 5 months ago

I didn't know it needed ROCM, honestly I didn't even know it was a different thing, I thought the amdgpu driver would work out of the box

snoweuph commented 5 months ago

no, rocm is what nvvidia toolkit is, but for amd

Jeffser commented 5 months ago

I see, I'll work on this for next release, thanks for pointing it out, I don't have a compatible AMD GPU so I couldn't test it haha

snoweuph commented 5 months ago

also I wouldt even be able to get the libraries instaled on my system, becuase im running an immutable one, but its working inside the docker, so it should also be able to get it work with the flatpak

snoweuph commented 5 months ago

I see, I'll work on this for next release, thanks for pointing it out, I don't have a compatible AMD GPU so I couldn't test it haha

mine isnt 100% supported, but near enough and works flawlessley when using the override with the docker container

snoweuph commented 5 months ago

I see, I'll work on this for next release, thanks for pointing it out, I don't have a compatible AMD GPU so I couldn't test it haha

then good luck, just tell me when you need a tester xD

Jeffser commented 5 months ago

btw what distro do you use? I've been wanting to use an immutable distro for my laptop

sand-r commented 5 months ago

I installed rocm-flatpak and set Alpaca to use that flatpak driver, but I'm also unable to get it to use my 7900xtx. Any ideas?

The flatpak driver recognizes my gpu fine so I'm not sure why it's not being used.

$ env OCL_ICD_VENDORS=/usr/lib/x86_64-linux-gnu/GL/ROCm/OpenCL/vendors FLATPAK_GL_DRIVERS=ROCm:default flatpak run --device=all --command=/bin/bash --devel org.freedesktop.Platform.ClInfo
[📦 org.freedesktop.Platform.ClInfo ~]$ clinfo 
Number of platforms                               1
  Platform Name                                   AMD Accelerated Parallel Processing
  Platform Vendor                                 Advanced Micro Devices, Inc.
  Platform Version                                OpenCL 2.1 AMD-APP (3602.0)
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd cl_amd_event_callback 
  Platform Extensions function suffix             AMD
  Platform Host timer resolution                  1ns

  Platform Name                                   AMD Accelerated Parallel Processing
Number of devices                                 1
  Device Name                                     gfx1100
  Device Vendor                                   Advanced Micro Devices, Inc.
  Device Vendor ID                                0x1002
  Device Version                                  OpenCL 2.0 
  Driver Version                                  3602.0 (HSA1.1,LC)
  Device OpenCL C Version                         OpenCL C 2.0 
  Device Type                                     GPU
  Device Board Name (AMD)                         AMD Radeon RX 7900 XTX
  Device PCI-e ID (AMD)                           0x744c
  Device Topology (AMD)                           PCI-E, 0000:0d:00.0
  Device Profile                                  FULL_PROFILE
  Device Available                                Yes
  Compiler Available                              Yes
  Linker Available                                Yes
  Max compute units                               48
  SIMD per compute unit (AMD)                     4
  SIMD width (AMD)                                32
  SIMD instruction width (AMD)                    1
  Max clock frequency                             2304MHz
  Graphics IP (AMD)                               11.0
  Device Partition                                (core)
    Max number of sub-devices                     48
    Supported partition types                     None
    Supported affinity domains                    (n/a)
  Max work item dimensions                        3
  Max work item sizes                             1024x1024x1024
  Max work group size                             256
  Preferred work group size (AMD)                 256
  Max work group size (AMD)                       1024
  Preferred work group size multiple (kernel)     32
  Wavefront width (AMD)                           32
  Preferred / native vector sizes                 
    char                                                 4 / 4       
    short                                                2 / 2       
    int                                                  1 / 1       
    long                                                 1 / 1       
    half                                                 1 / 1        (cl_khr_fp16)
    float                                                1 / 1       
    double                                               1 / 1        (cl_khr_fp64)
  Half-precision Floating-point support           (cl_khr_fp16)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
  Single-precision Floating-point support         (core)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  Yes
  Double-precision Floating-point support         (cl_khr_fp64)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
  Address bits                                    64, Little-Endian
  Global memory size                              25753026560 (23.98GiB)
  Global free memory (AMD)                        24946688 (23.79GiB) 24946688 (23.79GiB)
  Global memory channels (AMD)                    12
  Global memory banks per channel (AMD)           4
  Global memory bank width (AMD)                  256 bytes
  Error Correction support                        No
  Max memory allocation                           21890072576 (20.39GiB)
  Unified memory for Host and Device              No
  Shared Virtual Memory (SVM) capabilities        (core)
    Coarse-grained buffer sharing                 Yes
    Fine-grained buffer sharing                   Yes
    Fine-grained system sharing                   No
    Atomics                                       No
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       1024 bits (128 bytes)
  Preferred alignment for atomics                 
    SVM                                           0 bytes
    Global                                        0 bytes
    Local                                         0 bytes
  Max size for global variable                    21890072576 (20.39GiB)
  Preferred total size of global vars             25753026560 (23.98GiB)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        32768 (32KiB)
  Global Memory cache line size                   64 bytes
  Image support                                   Yes
    Max number of samplers per kernel             16
    Max size for 1D images from buffer            134217728 pixels
    Max 1D or 2D image array size                 8192 images
    Base address alignment for 2D image buffers   256 bytes
    Pitch alignment for 2D image buffers          256 pixels
    Max 2D image size                             16384x16384 pixels
    Max 3D image size                             16384x16384x8192 pixels
    Max number of read image args                 128
    Max number of write image args                8
    Max number of read/write image args           64
  Max number of pipe args                         16
  Max active pipe reservations                    16
  Max pipe packet size                            415236096 (396MiB)
  Local memory type                               Local
  Local memory size                               65536 (64KiB)
  Local memory size per CU (AMD)                  65536 (64KiB)
  Local memory banks (AMD)                        32
  Max number of constant args                     8
  Max constant buffer size                        21890072576 (20.39GiB)
  Preferred constant buffer size (AMD)            16384 (16KiB)
  Max size of kernel argument                     1024
  Queue properties (on host)                      
    Out-of-order execution                        No
    Profiling                                     Yes
  Queue properties (on device)                    
    Out-of-order execution                        Yes
    Profiling                                     Yes
    Preferred size                                262144 (256KiB)
    Max size                                      8388608 (8MiB)
  Max queues on device                            1
  Max events on device                            1024
  Prefer user sync for interop                    Yes
  Number of P2P devices (AMD)                     0
  Profiling timer resolution                      1ns
  Profiling timer offset since Epoch (AMD)        0ns (Wed Dec 31 19:00:00 1969)
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            No
    Thread trace supported (AMD)                  No
    Number of async queues (AMD)                  8
    Max real-time compute queues (AMD)            8
    Max real-time compute units (AMD)             48
  printf() buffer size                            4194304 (4MiB)
  Built-in kernels                                (n/a)
  Device Extensions                               cl_khr_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_fp16 cl_khr_gl_sharing cl_amd_device_attribute_query cl_amd_media_ops cl_amd_media_ops2 cl_khr_image2d_from_buffer cl_khr_subgroups cl_khr_depth_images cl_amd_copy_buffer_p2p cl_amd_assembly_program 

NULL platform behavior
  clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...)  AMD Accelerated Parallel Processing
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...)   Success [AMD]
  clCreateContext(NULL, ...) [default]            Success [AMD]
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_DEFAULT)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   gfx1100
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CPU)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   gfx1100
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ACCELERATOR)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CUSTOM)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ALL)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   gfx1100

ICD loader properties
  ICD loader Name                                 OpenCL ICD Loader
  ICD loader Vendor                               OCL Icd free software
  ICD loader Version                              2.3.2
  ICD loader Profile                              OpenCL 3.0

Ollama does utilize my GPU properly when installed on the host. I'm on Fedora40.

snoweuph commented 5 months ago

I installed rocm-flatpak and set Alpaca to use that flatpak driver, but I'm also unable to get it to use my 7900xtx. Any ideas?

The flatpak driver recognizes my gpu fine so I'm not sure why it's not being used.

$ env OCL_ICD_VENDORS=/usr/lib/x86_64-linux-gnu/GL/ROCm/OpenCL/vendors FLATPAK_GL_DRIVERS=ROCm:default flatpak run --device=all --command=/bin/bash --devel org.freedesktop.Platform.ClInfo
[📦 org.freedesktop.Platform.ClInfo ~]$ clinfo 
Number of platforms                               1
  Platform Name                                   AMD Accelerated Parallel Processing
  Platform Vendor                                 Advanced Micro Devices, Inc.
  Platform Version                                OpenCL 2.1 AMD-APP (3602.0)
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd cl_amd_event_callback 
  Platform Extensions function suffix             AMD
  Platform Host timer resolution                  1ns

  Platform Name                                   AMD Accelerated Parallel Processing
Number of devices                                 1
  Device Name                                     gfx1100
  Device Vendor                                   Advanced Micro Devices, Inc.
  Device Vendor ID                                0x1002
  Device Version                                  OpenCL 2.0 
  Driver Version                                  3602.0 (HSA1.1,LC)
  Device OpenCL C Version                         OpenCL C 2.0 
  Device Type                                     GPU
  Device Board Name (AMD)                         AMD Radeon RX 7900 XTX
  Device PCI-e ID (AMD)                           0x744c
  Device Topology (AMD)                           PCI-E, 0000:0d:00.0
  Device Profile                                  FULL_PROFILE
  Device Available                                Yes
  Compiler Available                              Yes
  Linker Available                                Yes
  Max compute units                               48
  SIMD per compute unit (AMD)                     4
  SIMD width (AMD)                                32
  SIMD instruction width (AMD)                    1
  Max clock frequency                             2304MHz
  Graphics IP (AMD)                               11.0
  Device Partition                                (core)
    Max number of sub-devices                     48
    Supported partition types                     None
    Supported affinity domains                    (n/a)
  Max work item dimensions                        3
  Max work item sizes                             1024x1024x1024
  Max work group size                             256
  Preferred work group size (AMD)                 256
  Max work group size (AMD)                       1024
  Preferred work group size multiple (kernel)     32
  Wavefront width (AMD)                           32
  Preferred / native vector sizes                 
    char                                                 4 / 4       
    short                                                2 / 2       
    int                                                  1 / 1       
    long                                                 1 / 1       
    half                                                 1 / 1        (cl_khr_fp16)
    float                                                1 / 1       
    double                                               1 / 1        (cl_khr_fp64)
  Half-precision Floating-point support           (cl_khr_fp16)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
  Single-precision Floating-point support         (core)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  Yes
  Double-precision Floating-point support         (cl_khr_fp64)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
  Address bits                                    64, Little-Endian
  Global memory size                              25753026560 (23.98GiB)
  Global free memory (AMD)                        24946688 (23.79GiB) 24946688 (23.79GiB)
  Global memory channels (AMD)                    12
  Global memory banks per channel (AMD)           4
  Global memory bank width (AMD)                  256 bytes
  Error Correction support                        No
  Max memory allocation                           21890072576 (20.39GiB)
  Unified memory for Host and Device              No
  Shared Virtual Memory (SVM) capabilities        (core)
    Coarse-grained buffer sharing                 Yes
    Fine-grained buffer sharing                   Yes
    Fine-grained system sharing                   No
    Atomics                                       No
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       1024 bits (128 bytes)
  Preferred alignment for atomics                 
    SVM                                           0 bytes
    Global                                        0 bytes
    Local                                         0 bytes
  Max size for global variable                    21890072576 (20.39GiB)
  Preferred total size of global vars             25753026560 (23.98GiB)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        32768 (32KiB)
  Global Memory cache line size                   64 bytes
  Image support                                   Yes
    Max number of samplers per kernel             16
    Max size for 1D images from buffer            134217728 pixels
    Max 1D or 2D image array size                 8192 images
    Base address alignment for 2D image buffers   256 bytes
    Pitch alignment for 2D image buffers          256 pixels
    Max 2D image size                             16384x16384 pixels
    Max 3D image size                             16384x16384x8192 pixels
    Max number of read image args                 128
    Max number of write image args                8
    Max number of read/write image args           64
  Max number of pipe args                         16
  Max active pipe reservations                    16
  Max pipe packet size                            415236096 (396MiB)
  Local memory type                               Local
  Local memory size                               65536 (64KiB)
  Local memory size per CU (AMD)                  65536 (64KiB)
  Local memory banks (AMD)                        32
  Max number of constant args                     8
  Max constant buffer size                        21890072576 (20.39GiB)
  Preferred constant buffer size (AMD)            16384 (16KiB)
  Max size of kernel argument                     1024
  Queue properties (on host)                      
    Out-of-order execution                        No
    Profiling                                     Yes
  Queue properties (on device)                    
    Out-of-order execution                        Yes
    Profiling                                     Yes
    Preferred size                                262144 (256KiB)
    Max size                                      8388608 (8MiB)
  Max queues on device                            1
  Max events on device                            1024
  Prefer user sync for interop                    Yes
  Number of P2P devices (AMD)                     0
  Profiling timer resolution                      1ns
  Profiling timer offset since Epoch (AMD)        0ns (Wed Dec 31 19:00:00 1969)
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            No
    Thread trace supported (AMD)                  No
    Number of async queues (AMD)                  8
    Max real-time compute queues (AMD)            8
    Max real-time compute units (AMD)             48
  printf() buffer size                            4194304 (4MiB)
  Built-in kernels                                (n/a)
  Device Extensions                               cl_khr_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_fp16 cl_khr_gl_sharing cl_amd_device_attribute_query cl_amd_media_ops cl_amd_media_ops2 cl_khr_image2d_from_buffer cl_khr_subgroups cl_khr_depth_images cl_amd_copy_buffer_p2p cl_amd_assembly_program 

NULL platform behavior
  clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...)  AMD Accelerated Parallel Processing
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...)   Success [AMD]
  clCreateContext(NULL, ...) [default]            Success [AMD]
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_DEFAULT)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   gfx1100
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CPU)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   gfx1100
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ACCELERATOR)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CUSTOM)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ALL)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   gfx1100

ICD loader properties
  ICD loader Name                                 OpenCL ICD Loader
  ICD loader Vendor                               OCL Icd free software
  ICD loader Version                              2.3.2
  ICD loader Profile                              OpenCL 3.0

Ollama does utilize my GPU properly when installed on the host. I'm on Fedora40.

Alpaca doesnt currently use the right ollama binary, and is also not linked to the flatpak rocm binaries. (please read the full issue comments)

@Jeffser (The Creator of this project) Said, that he will work on this in the future, because it seemes like theres quite some stuff to do for AMD support it completeley.

for now you can just run ollama inside of a docker, or nativley and link to it in alpaca, for questions how to run ollama localy, or in a docker, see the ollama wiki.

snoweuph commented 5 months ago

btw what distro do you use? I've been wanting to use an immutable distro for my laptop

Im currentley utelizing vanilla Kinoite, but I gonna give Ublue a changse soon, especiially, becuase on native Kinoite WebRTC is jittery with my GPU for some reason, and KDE connect isnt working with with the firefox flatpak, so I gonna see, if with their kinoite image: https://github.com/ublue-os/main/pkgs/container/kinoite-main im gonna have a better experince (i will only have to reabase, so switching and trying out will be easy)