[GH-ISSUE #14045] glm-4.7-flash is slow and uses a lot of cpu #71237

New Issue

GiteaMirror · 2026-05-05T00:50:53-05:00

GiteaMirror commented

2026-05-05 00:50:53 -05:00

Originally created by @inforithmics on GitHub (Feb 3, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14045

Originally assigned to: @jmorganca on GitHub.

What is the issue?

I downloaded 0.15.5 rc and let it run on 7900 xtx and the model was very slow and used a lot of cpu. Memory was allocated on gpu. I reverted back to 15.4 and it was fast again.

Relevant log output

OS

Windows

GPU

AMD

CPU

Intel

Ollama version

0.5.15 rc

Originally created by @inforithmics on GitHub (Feb 3, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14045 Originally assigned to: @jmorganca on GitHub. ### What is the issue? I downloaded 0.15.5 rc and let it run on 7900 xtx and the model was very slow and used a lot of cpu. Memory was allocated on gpu. I reverted back to 15.4 and it was fast again. ### Relevant log output ```shell ``` ### OS Windows ### GPU AMD ### CPU Intel ### Ollama version 0.5.15 rc

GiteaMirror added the bug label 2026-05-05 00:50:53 -05:00

GiteaMirror commented

2026-05-05 00:50:58 -05:00

@rick-github commented on GitHub (Feb 3, 2026):

Server logs may aid in debugging.

@rick-github commented on GitHub (Feb 3, 2026): [Server logs](https://docs.ollama.com/troubleshooting) may aid in debugging.

GiteaMirror commented

2026-05-05 00:51:00 -05:00

@rick-github commented on GitHub (Feb 3, 2026):

glm-4.7-flash   0.15.3       prompt eval rate: 5.32 tokens/s eval rate: 131.31 tokens/s
glm-4.7-flash   0.15.4       prompt eval rate: 5.33 tokens/s eval rate: 132.45 tokens/s
glm-4.7-flash   0.15.5-rc0   prompt eval rate: 5.34 tokens/s eval rate: 131.33 tokens/s
glm-4.7-flash   0.15.5-rc1   prompt eval rate: 5.30 tokens/s eval rate: 21.11 tokens/s
qwen3           0.15.3       prompt eval rate: 8.20 tokens/s eval rate: 189.55 tokens/s
qwen3           0.15.4       prompt eval rate: 7.44 tokens/s eval rate: 192.03 tokens/s
qwen3           0.15.5-rc0   prompt eval rate: 7.72 tokens/s eval rate: 192.54 tokens/s
qwen3           0.15.5-rc1   prompt eval rate: 7.54 tokens/s eval rate: 163.58 tokens/s

Between rc0 and rc1 is the vendor sync (#13832), so that seems the likely cause.

@rick-github commented on GitHub (Feb 3, 2026): ``` glm-4.7-flash 0.15.3 prompt eval rate: 5.32 tokens/s eval rate: 131.31 tokens/s glm-4.7-flash 0.15.4 prompt eval rate: 5.33 tokens/s eval rate: 132.45 tokens/s glm-4.7-flash 0.15.5-rc0 prompt eval rate: 5.34 tokens/s eval rate: 131.33 tokens/s glm-4.7-flash 0.15.5-rc1 prompt eval rate: 5.30 tokens/s eval rate: 21.11 tokens/s qwen3 0.15.3 prompt eval rate: 8.20 tokens/s eval rate: 189.55 tokens/s qwen3 0.15.4 prompt eval rate: 7.44 tokens/s eval rate: 192.03 tokens/s qwen3 0.15.5-rc0 prompt eval rate: 7.72 tokens/s eval rate: 192.54 tokens/s qwen3 0.15.5-rc1 prompt eval rate: 7.54 tokens/s eval rate: 163.58 tokens/s ``` Between rc0 and rc1 is the vendor sync (#13832), so that seems the likely cause.

GiteaMirror commented

2026-05-05 00:51:01 -05:00

@inforithmics commented on GitHub (Feb 3, 2026):

When I checked out the code from the vendor sync. It was still fast. But I saw that glm 4.7 was implemented in the experimental engine. Maybe this influences it somehow try to investigate further.

@inforithmics commented on GitHub (Feb 3, 2026): When I checked out the code from the vendor sync. It was still fast. But I saw that glm 4.7 was implemented in the experimental engine. Maybe this influences it somehow try to investigate further.

GiteaMirror commented

2026-05-05 00:51:02 -05:00

@rick-github commented on GitHub (Feb 3, 2026):

Some other models are also affected to a lesser degree, eg qwen3 and lfm2.5-thinking, but not glm-4.7 or minimax-m2.1.

glm-4.7-flash        0.15.4       prompt eval rate: 5.40 tokens/s eval rate: 131.65 tokens/s
glm-4.7-flash        0.15.5-rc0   prompt eval rate: 5.33 tokens/s eval rate: 130.27 tokens/s
glm-4.7-flash        0.15.5-rc1   prompt eval rate: 5.27 tokens/s eval rate: 21.79 tokens/s
qwen3                0.15.4       prompt eval rate: 8.10 tokens/s eval rate: 192.90 tokens/s
qwen3                0.15.5-rc0   prompt eval rate: 7.54 tokens/s eval rate: 192.73 tokens/s
qwen3                0.15.5-rc1   prompt eval rate: 7.77 tokens/s eval rate: 166.06 tokens/s
lfm2.5-thinking:1.2b 0.15.4       prompt eval rate: 189.03 tokens/s eval rate: 783.76 tokens/s
lfm2.5-thinking:1.2b 0.15.5-rc0   prompt eval rate: 4117.98 tokens/s eval rate: 782.86 tokens/s
lfm2.5-thinking:1.2b 0.15.5-rc1   prompt eval rate: 4027.82 tokens/s eval rate: 574.14 tokens/s
frob/glm-4.7         0.15.4       prompt eval rate: 6.34 tokens/s eval rate: 49.45 tokens/s
frob/glm-4.7         0.15.5-rc0   prompt eval rate: 6.66 tokens/s eval rate: 49.59 tokens/s
frob/glm-4.7         0.15.5-rc1   prompt eval rate: 6.26 tokens/s eval rate: 49.76 tokens/s
frob/minimax-m2.1    0.15.4       prompt eval rate: 160.88 tokens/s eval rate: 119.91 tokens/s
frob/minimax-m2.1    0.15.5-rc0   prompt eval rate: 170.86 tokens/s eval rate: 119.78 tokens/s
frob/minimax-m2.1    0.15.5-rc1   prompt eval rate: 157.15 tokens/s eval rate: 121.02 tokens/s

@rick-github commented on GitHub (Feb 3, 2026): Some other models are also affected to a lesser degree, eg qwen3 and lfm2.5-thinking, but not glm-4.7 or minimax-m2.1. ``` glm-4.7-flash 0.15.4 prompt eval rate: 5.40 tokens/s eval rate: 131.65 tokens/s glm-4.7-flash 0.15.5-rc0 prompt eval rate: 5.33 tokens/s eval rate: 130.27 tokens/s glm-4.7-flash 0.15.5-rc1 prompt eval rate: 5.27 tokens/s eval rate: 21.79 tokens/s qwen3 0.15.4 prompt eval rate: 8.10 tokens/s eval rate: 192.90 tokens/s qwen3 0.15.5-rc0 prompt eval rate: 7.54 tokens/s eval rate: 192.73 tokens/s qwen3 0.15.5-rc1 prompt eval rate: 7.77 tokens/s eval rate: 166.06 tokens/s lfm2.5-thinking:1.2b 0.15.4 prompt eval rate: 189.03 tokens/s eval rate: 783.76 tokens/s lfm2.5-thinking:1.2b 0.15.5-rc0 prompt eval rate: 4117.98 tokens/s eval rate: 782.86 tokens/s lfm2.5-thinking:1.2b 0.15.5-rc1 prompt eval rate: 4027.82 tokens/s eval rate: 574.14 tokens/s frob/glm-4.7 0.15.4 prompt eval rate: 6.34 tokens/s eval rate: 49.45 tokens/s frob/glm-4.7 0.15.5-rc0 prompt eval rate: 6.66 tokens/s eval rate: 49.59 tokens/s frob/glm-4.7 0.15.5-rc1 prompt eval rate: 6.26 tokens/s eval rate: 49.76 tokens/s frob/minimax-m2.1 0.15.4 prompt eval rate: 160.88 tokens/s eval rate: 119.91 tokens/s frob/minimax-m2.1 0.15.5-rc0 prompt eval rate: 170.86 tokens/s eval rate: 119.78 tokens/s frob/minimax-m2.1 0.15.5-rc1 prompt eval rate: 157.15 tokens/s eval rate: 121.02 tokens/s ```

GiteaMirror commented

2026-05-05 00:51:04 -05:00

@jessegross commented on GitHub (Feb 3, 2026):

Confirmed that it is the GGML bump:

ef00199fb4e6d045e11e76baaab9049f3234939d is the first bad commit
commit ef00199fb4e6d045e11e76baaab9049f3234939d
Author: Jeffrey Morgan <jmorganca@gmail.com>
Date:   Mon Feb 2 17:31:59 2026 -0800

    Update vendor ggml code to a5bb8ba4 (#13832)
    
    Co-authored-by: Daniel Hiltgen <daniel@ollama.com>
    Co-authored-by: Gabe Goodhart <ghart@us.ibm.com>
    Co-authored-by: Shalini Salomi Bodapati <Shalini.Salomi.Bodapati@ibm.com>

@jessegross commented on GitHub (Feb 3, 2026): Confirmed that it is the GGML bump: ``` ef00199fb4e6d045e11e76baaab9049f3234939d is the first bad commit commit ef00199fb4e6d045e11e76baaab9049f3234939d Author: Jeffrey Morgan <jmorganca@gmail.com> Date: Mon Feb 2 17:31:59 2026 -0800 Update vendor ggml code to a5bb8ba4 (#13832) Co-authored-by: Daniel Hiltgen <daniel@ollama.com> Co-authored-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: Shalini Salomi Bodapati <Shalini.Salomi.Bodapati@ibm.com> ```

GiteaMirror commented

2026-05-05 00:51:06 -05:00

@inforithmics commented on GitHub (Feb 3, 2026):

Could this help?
https://github.com/ollama/ollama/pull/13597
Because how it looks something is run on the cpu and normally this means that an ggml operation isn‘t supported.

@inforithmics commented on GitHub (Feb 3, 2026): Could this help? https://github.com/ollama/ollama/pull/13597 Because how it looks something is run on the cpu and normally this means that an ggml operation isn‘t supported.

GiteaMirror commented

2026-05-05 00:51:08 -05:00

@inforithmics commented on GitHub (Feb 3, 2026):

Or maybe this caused it
https://github.com/ggml-org/llama.cpp/pull/18986

@inforithmics commented on GitHub (Feb 3, 2026): Or maybe this caused it https://github.com/ggml-org/llama.cpp/pull/18986

GiteaMirror commented

2026-05-05 00:51:09 -05:00

@rick-github commented on GitHub (Feb 3, 2026):

Doesn't appear to affect ROCm devices, above was Nvidia.

glm-4.7-flash        0.15.4-rocm       prompt eval rate: 93.72 tokens/s eval rate: 36.22 tokens/s
glm-4.7-flash        0.15.5-rc0-rocm   prompt eval rate: 82.84 tokens/s eval rate: 35.72 tokens/s
glm-4.7-flash        0.15.5-rc1-rocm   prompt eval rate: 94.07 tokens/s eval rate: 35.67 tokens/s
qwen3                0.15.4-rocm       prompt eval rate: 179.80 tokens/s eval rate: 31.69 tokens/s
qwen3                0.15.5-rc0-rocm   prompt eval rate: 172.61 tokens/s eval rate: 32.02 tokens/s
qwen3                0.15.5-rc1-rocm   prompt eval rate: 177.51 tokens/s eval rate: 32.14 tokens/s
lfm2.5-thinking:1.2b 0.15.4-rocm       prompt eval rate: 269.60 tokens/s eval rate: 170.19 tokens/s
lfm2.5-thinking:1.2b 0.15.5-rc0-rocm   prompt eval rate: 278.23 tokens/s eval rate: 168.17 tokens/s
lfm2.5-thinking:1.2b 0.15.5-rc1-rocm   prompt eval rate: 285.79 tokens/s eval rate: 166.60 tokens/s

@rick-github commented on GitHub (Feb 3, 2026): Doesn't appear to affect ROCm devices, above was Nvidia. ``` glm-4.7-flash 0.15.4-rocm prompt eval rate: 93.72 tokens/s eval rate: 36.22 tokens/s glm-4.7-flash 0.15.5-rc0-rocm prompt eval rate: 82.84 tokens/s eval rate: 35.72 tokens/s glm-4.7-flash 0.15.5-rc1-rocm prompt eval rate: 94.07 tokens/s eval rate: 35.67 tokens/s qwen3 0.15.4-rocm prompt eval rate: 179.80 tokens/s eval rate: 31.69 tokens/s qwen3 0.15.5-rc0-rocm prompt eval rate: 172.61 tokens/s eval rate: 32.02 tokens/s qwen3 0.15.5-rc1-rocm prompt eval rate: 177.51 tokens/s eval rate: 32.14 tokens/s lfm2.5-thinking:1.2b 0.15.4-rocm prompt eval rate: 269.60 tokens/s eval rate: 170.19 tokens/s lfm2.5-thinking:1.2b 0.15.5-rc0-rocm prompt eval rate: 278.23 tokens/s eval rate: 168.17 tokens/s lfm2.5-thinking:1.2b 0.15.5-rc1-rocm prompt eval rate: 285.79 tokens/s eval rate: 166.60 tokens/s ```

GiteaMirror commented

2026-05-05 00:51:11 -05:00

@jmorganca commented on GitHub (Feb 3, 2026):

Thanks @rick-github for the testing. Working on this!

@jmorganca commented on GitHub (Feb 3, 2026): Thanks @rick-github for the testing. Working on this!

GiteaMirror commented

2026-05-05 00:51:12 -05:00

@inforithmics commented on GitHub (Feb 4, 2026):

Doesn't appear to affect ROCm devices, above was Nvidia.

glm-4.7-flash        0.15.4-rocm       prompt eval rate: 93.72 tokens/s eval rate: 36.22 tokens/s
glm-4.7-flash        0.15.5-rc0-rocm   prompt eval rate: 82.84 tokens/s eval rate: 35.72 tokens/s
glm-4.7-flash        0.15.5-rc1-rocm   prompt eval rate: 94.07 tokens/s eval rate: 35.67 tokens/s
qwen3                0.15.4-rocm       prompt eval rate: 179.80 tokens/s eval rate: 31.69 tokens/s
qwen3                0.15.5-rc0-rocm   prompt eval rate: 172.61 tokens/s eval rate: 32.02 tokens/s
qwen3                0.15.5-rc1-rocm   prompt eval rate: 177.51 tokens/s eval rate: 32.14 tokens/s
lfm2.5-thinking:1.2b 0.15.4-rocm       prompt eval rate: 269.60 tokens/s eval rate: 170.19 tokens/s
lfm2.5-thinking:1.2b 0.15.5-rc0-rocm   prompt eval rate: 278.23 tokens/s eval rate: 168.17 tokens/s
lfm2.5-thinking:1.2b 0.15.5-rc1-rocm   prompt eval rate: 285.79 tokens/s eval rate: 166.60 tokens/s

Intresting How much VRAM does this ROCM Device have on a ROCM Device with 24 GB It showed this behavior too.
All Layers where allocated on the Gpu some Graph Memory was allocated on Cpu and more on Gpu.

@inforithmics commented on GitHub (Feb 4, 2026): > Doesn't appear to affect ROCm devices, above was Nvidia. > > ``` > glm-4.7-flash 0.15.4-rocm prompt eval rate: 93.72 tokens/s eval rate: 36.22 tokens/s > glm-4.7-flash 0.15.5-rc0-rocm prompt eval rate: 82.84 tokens/s eval rate: 35.72 tokens/s > glm-4.7-flash 0.15.5-rc1-rocm prompt eval rate: 94.07 tokens/s eval rate: 35.67 tokens/s > qwen3 0.15.4-rocm prompt eval rate: 179.80 tokens/s eval rate: 31.69 tokens/s > qwen3 0.15.5-rc0-rocm prompt eval rate: 172.61 tokens/s eval rate: 32.02 tokens/s > qwen3 0.15.5-rc1-rocm prompt eval rate: 177.51 tokens/s eval rate: 32.14 tokens/s > lfm2.5-thinking:1.2b 0.15.4-rocm prompt eval rate: 269.60 tokens/s eval rate: 170.19 tokens/s > lfm2.5-thinking:1.2b 0.15.5-rc0-rocm prompt eval rate: 278.23 tokens/s eval rate: 168.17 tokens/s > lfm2.5-thinking:1.2b 0.15.5-rc1-rocm prompt eval rate: 285.79 tokens/s eval rate: 166.60 tokens/s > ``` Intresting How much VRAM does this ROCM Device have on a ROCM Device with 24 GB It showed this behavior too. All Layers where allocated on the Gpu some Graph Memory was allocated on Cpu and more on Gpu.

GiteaMirror commented

2026-05-05 00:51:14 -05:00

@inforithmics commented on GitHub (Feb 4, 2026):

I compared the log outputs of Ollama 0.15.4 rocm and 0.15.5-rc1 rocm (Vulkan is enabled)

And found following interesting things.

0.15.4:
ggml_backend_vk_get_device_memory called: uuid 00000000-0300-0000-0000-000000000000time=2026-02-04T08:49:53.983+01:00 level=INFO source=device.go:267 msg="compute graph" device=CPU size="175.6 MiB"
time=2026-02-04T08:49:53.983+01:00 level=INFO source=device.go:272 msg="total memory" size="21.4 GiB"
t
0.15.5-rc1:
ggml_backend_vk_device_get_memory called: uuid 00000000-0300-0000-0000-000000000000

change:
change of method name ggml_backend_vk_device_get_memory

0.15.4:
time=2026-02-04T08:47:10.272+01:00 level=DEBUG source=server.go:974 msg="available gpu" id=0 library=ROCm "available layer vram"="23.0 GiB" backoff=0.00 minimum="457.0 MiB" overhead="0 B" graph="398.6 MiB"
time=2026-02-04T08:47:10.272+01:00 level=DEBUG source=server.go:974 msg="available gpu" id=868080a7-0400-0000-0002-000000000000 library=Vulkan "available layer vram"="63.5 GiB" backoff=0.00 minimum="457.0 MiB" overhead="0 B" graph="0 B"
0.15.5-rc1:
time=2026-02-04T08:49:53.981+01:00 level=DEBUG source=server.go:975 msg="available gpu" id=0 library=ROCm "available layer vram"="23.1 GiB" backoff=0.00 minimum="457.0 MiB" overhead="0 B" graph="311.6 MiB"
time=2026-02-04T08:49:53.981+01:00 level=DEBUG source=server.go:975 msg="available gpu" id=00000000-0300-0000-0000-000000000000 library=Vulkan "available layer vram"="23.1 GiB" backoff=0.00 minimum="457.0 MiB" overhead="0 B" graph="0 B"
time=2026-02-04T08:49:53.981+01:00 level=DEBUG source=server.go:975 msg="available gpu" id=868080a7-0400-0000-0002-000000000000 library=Vulkan "available layer vram"="63.5 GiB" backoff=0.00 minimum="457.0 MiB" overhead="0 B" graph="0 B"

change:
id=00000000-0300-0000-0000-000000000000 library=Vulkan "available layer vram"="23.1 GiB"

vulkan uses this device which should already supported by rocm (So it isn't filtered out) maybe the pci_id oder something similar isn't returned anymore. So Ollama doesn't detect the duplicate.
It shows 23.1 GiB available but rocm already allocated memory on this device.

0.15.4:
time=2026-02-04T08:47:10.273+01:00 level=INFO source=device.go:262 msg="compute graph" device=ROCm0 size="398.6 MiB"
time=2026-02-04T08:47:10.273+01:00 level=INFO source=device.go:267 msg="compute graph" device=CPU size="4.0 MiB"
0.15.5-rc1:
time=2026-02-04T08:49:53.983+01:00 level=INFO source=device.go:262 msg="compute graph" device=ROCm0 size="311.6 MiB"
time=2026-02-04T08:49:53.983+01:00 level=INFO source=device.go:267 msg="compute graph" device=CPU size="175.6 MiB"

change:
The Total compute graph size is higher and much more memory is allocated on cpu

Ollama0.15.4.log
Ollama0.15.5-rc1.log

@inforithmics commented on GitHub (Feb 4, 2026): I compared the log outputs of Ollama 0.15.4 rocm and 0.15.5-rc1 rocm (Vulkan is enabled) And found following interesting things. 0.15.4: ggml_backend_vk_get_device_memory called: uuid 00000000-0300-0000-0000-000000000000time=2026-02-04T08:49:53.983+01:00 level=INFO source=device.go:267 msg="compute graph" device=CPU size="175.6 MiB" time=2026-02-04T08:49:53.983+01:00 level=INFO source=device.go:272 msg="total memory" size="21.4 GiB" t 0.15.5-rc1: ggml_backend_vk_device_get_memory called: uuid 00000000-0300-0000-0000-000000000000 change: change of method name ggml_backend_vk_device_get_memory 0.15.4: time=2026-02-04T08:47:10.272+01:00 level=DEBUG source=server.go:974 msg="available gpu" id=0 library=ROCm "available layer vram"="23.0 GiB" backoff=0.00 minimum="457.0 MiB" overhead="0 B" graph="398.6 MiB" time=2026-02-04T08:47:10.272+01:00 level=DEBUG source=server.go:974 msg="available gpu" id=868080a7-0400-0000-0002-000000000000 library=Vulkan "available layer vram"="63.5 GiB" backoff=0.00 minimum="457.0 MiB" overhead="0 B" graph="0 B" 0.15.5-rc1: time=2026-02-04T08:49:53.981+01:00 level=DEBUG source=server.go:975 msg="available gpu" id=0 library=ROCm "available layer vram"="23.1 GiB" backoff=0.00 minimum="457.0 MiB" overhead="0 B" graph="311.6 MiB" time=2026-02-04T08:49:53.981+01:00 level=DEBUG source=server.go:975 msg="available gpu" id=00000000-0300-0000-0000-000000000000 library=Vulkan "available layer vram"="23.1 GiB" backoff=0.00 minimum="457.0 MiB" overhead="0 B" graph="0 B" time=2026-02-04T08:49:53.981+01:00 level=DEBUG source=server.go:975 msg="available gpu" id=868080a7-0400-0000-0002-000000000000 library=Vulkan "available layer vram"="63.5 GiB" backoff=0.00 minimum="457.0 MiB" overhead="0 B" graph="0 B" change: id=00000000-0300-0000-0000-000000000000 library=Vulkan "available layer vram"="23.1 GiB" 1. vulkan uses this device which should already supported by rocm (So it isn't filtered out) maybe the pci_id oder something similar isn't returned anymore. So Ollama doesn't detect the duplicate. 2. It shows 23.1 GiB available but rocm already allocated memory on this device. 0.15.4: time=2026-02-04T08:47:10.273+01:00 level=INFO source=device.go:262 msg="compute graph" device=ROCm0 size="398.6 MiB" time=2026-02-04T08:47:10.273+01:00 level=INFO source=device.go:267 msg="compute graph" device=CPU size="4.0 MiB" 0.15.5-rc1: time=2026-02-04T08:49:53.983+01:00 level=INFO source=device.go:262 msg="compute graph" device=ROCm0 size="311.6 MiB" time=2026-02-04T08:49:53.983+01:00 level=INFO source=device.go:267 msg="compute graph" device=CPU size="175.6 MiB" change: The Total compute graph size is higher and much more memory is allocated on cpu [Ollama0.15.4.log](https://github.com/user-attachments/files/25065743/Ollama0.15.4.log) [Ollama0.15.5-rc1.log](https://github.com/user-attachments/files/25065744/Ollama0.15.5-rc1.log)

GiteaMirror commented

2026-05-05 00:51:15 -05:00

@inforithmics commented on GitHub (Feb 5, 2026):

I played around with the reverted Vendor Update on this branch https://github.com/inforithmics/ollama/tree/MainBeforeRevert
And the results are now that glm.4.7-flash works now faster as before. 100 TG/s instead of 20 TG/s

Changes made.

Vendor Update to https://github.com/ggml-org/llama.cpp/pull/19324
Add top_k pull request. https://github.com/ollama/ollama/pull/13597

The Graph Memory on CPU is now again 4MB, and the model although uses less VRAM on GPU (22GB with 64000 context size)

@inforithmics commented on GitHub (Feb 5, 2026): I played around with the reverted Vendor Update on this branch https://github.com/inforithmics/ollama/tree/MainBeforeRevert And the results are now that glm.4.7-flash works now faster as before. 100 TG/s instead of 20 TG/s Changes made. Vendor Update to https://github.com/ggml-org/llama.cpp/pull/19324 Add top_k pull request. https://github.com/ollama/ollama/pull/13597 The Graph Memory on CPU is now again 4MB, and the model although uses less VRAM on GPU (22GB with 64000 context size)

Sign in to join this conversation.

Branches Tags

main

dhiltgen/ci

parth-launch-plan-gating

hoyyeva/anthropic-reference-images-path

parth-anthropic-reference-images-path

brucemacd/download-before-remove

hoyyeva/editor-config-repair

parth-mlx-decode-checkpoints

parth-launch-codex-app

hoyyeva/fix-codex-model-metadata-warning

hoyyeva/qwen

parth/hide-claude-desktop-till-release

hoyyeva/opencode-image-modality

parth-add-claude-code-autoinstall

release_v0.22.0

pdevine/manifest-list

codex/fix-codex-model-metadata-warning

pdevine/addressable-manifest

brucemacd/launch-fetch-reccomended

jmorganca/llama-compat

launch-copilot-cli

hoyyeva/opencode-thinking

release_v0.20.7

parth-auto-save-backup

parth-test

jmorganca/gemma4-audio-replacements

fix-manifest-digest-on-pull

hoyyeva/vscode-improve

brucemacd/install-server-wait

parth/update-claude-docs

brucemac/start-ap-install

pdevine/mlx-update

pdevine/qwen35_vision

drifkin/api-show-fallback

mintlify/image-generation-1773352582

hoyyeva/server-context-length-local-config

jmorganca/faster-reptition-penalties

jmorganca/convert-nemotron

parth-pi-thinking

pdevine/sampling-penalties

jmorganca/fix-create-quantization-memory

dongchen/resumable_transfer_fix

pdevine/sampling-cache-error

jessegross/mlx-usage

hoyyeva/openclaw-config

hoyyeva/app-html

pdevine/qwen3next

brucemacd/sign-sh-install

brucemacd/tui-update

brucemacd/usage-api

jmorganca/launch-empty

fix-app-dist-embed

mxyng/mlx-compile

mxyng/mlx-quant

mxyng/mlx-glm4.7

mxyng/mlx

brucemacd/simplify-model-picker

jmorganca/qwen3-concurrent

fix-glm-4.7-flash-mla-config

drifkin/qwen3-coder-opening-tag

brucemacd/usage-cli

fix-cuda12-fattn-shmem

ollama-imagegen-docs

parth/fix-multiline-inputs

brucemacd/config-docs

mxyng/model-files

mxyng/simple-execute

fix-imagegen-ollama-models

mxyng/async-upload

jmorganca/lazy-no-dtype-changes

imagegen-auto-detect-create

parth/decrease-concurrent-download-hf

fix-mlx-quantize-init

jmorganca/x-cleanup

usage

imagegen-readme

jmorganca/glm-image

mlx-gpu-cd

jmorganca/imagegen-modelfile

parth/agent-skills

parth/agent-allowlist

parth/signed-in-offline

parth/agents

parth/fix-context-chopping

improve-cloud-flow

parth/add-models-websearch

parth/prompt-renderer-mcp

jmorganca/native-settings

jmorganca/download-stream-hash

jmorganca/client2-rebased

brucemacd/oai-chat-req-multipart

jessegross/multi_chunk_reserve

grace/additional-omit-empty

grace/mistral-3-large

mxyng/tokenizer2

mxyng/tokenizer

jessegross/flash

hoyyeva/windows-nacked-app

mxyng/cleanup-attention

grace/deepseek-parser

hoyyeva/remember-unsent-prompt

parth/add-lfs-pointer-error-conversion

parth/olmo2-test2

hoyyeva/ollama-launchagent-plist

nicole/olmo-model

parth/olmo-test

mxyng/remove-embedded

parth/render-template

jmorganca/intellect-3

parth/remove-prealloc-linter

jmorganca/cmd-eval

nicole/nomic-embed-text-fix

mxyng/lint-2

hoyyeva/add-gemini-3-pro-preview

hoyyeva/load-model-list

mxyng/expand-path

mxyng/environ-2

hoyyeva/deeplink-json-encoding

parth/improve-tool-calling-tests

hoyyeva/conversation

hoyyeva/assistant-edit-response

hoyyeva/thinking

origin/brucemacd/invalid-char-i-err

parth/improve-tool-calling

jmorganca/required-omitempty

grace/qwen3-vl-tests

mxyng/iter-client

parth/docs-readme

nicole/embed-test

pdevine/integration-benchstat

parth/remove-generate-cmd

parth/add-toolcall-id

mxyng/server-tests

jmorganca/glm-4.6

jmorganca/gin-h-compat

drifkin/stable-tool-args

pdevine/qwen3-more-thinking

parth/add-websearch-client

nicole/websearch_local

jmorganca/qwen3-coder-updates

grace/deepseek-v3-migration-tests

mxyng/fix-create

jmorganca/cloud-errors

pdevine/parser-tidy

revert-12233-parth/simplify-entrypoints-runner

parth/enable-so-gpt-oss

brucemacd/qwen3vl

jmorganca/readme-simplify

parth/gpt-oss-structured-outputs

revert-12039-jmorganca/tools-braces

mxyng/embeddings

mxyng/gguf

mxyng/benchmark

mxyng/types-null

parth/move-parsing

mxyng/gemma2

jmorganca/docs

mxyng/16-bit

mxyng/create-stdin

pdevine/authorizedkeys

mxyng/quant

parth/opt-in-error-context-window

brucemacd/cache-models

brucemacd/runner-completion

jmorganca/llama-update-6

brucemacd/benchmark-list

brucemacd/partial-read-caps

parth/deepseek-r1-tools

mxyng/omit-array

parth/tool-prefix-temp

brucemacd/runner-test

jmorganca/qwen25vl

brucemacd/model-forward-test-ext

parth/python-function-parsing

jmorganca/cuda-compression-none

drifkin/num-parallel

drifkin/chat-truncation-fix

jmorganca/sync

parth/python-tools-calling

drifkin/array-head-count

brucemacd/create-no-loop

parth/server-enable-content-stream-with-tools

qwen25omni

mxyng/v3

brucemacd/ropeconfig

jmorganca/silence-tokenizer

parth/sample-so-test

parth/sampling-structured-outputs

brucemacd/doc-go-engine

parth/constrained-sampling-json

jmorganca/mistral-wip

brucemacd/mistral-small-convert

parth/sample-unmarshal-json-for-params

brucemacd/jomorganca/mistral

pdevine/bfloat16

jmorganca/mistral

brucemacd/mistral

pdevine/logging

parth/sample-correctness-fix

parth/sample-fix-sorting

jmorgan/sample-fix-sorting-extras

jmorganca/temp-0-images

brucemacd/parallel-embed-models

brucemacd/shim-grammar

jmorganca/fix-gguf-error

bmizerany/nameswork

jmorganca/faster-releases

bmizerany/validatenames

brucemacd/err-no-vocab

brucemacd/rope-config

brucemacd/err-hint

brucemacd/qwen2_5

brucemacd/logprobs

brucemacd/new_runner_graph_bench

progress-flicker

brucemacd/forward-test

brucemacd/go_qwen2

pdevine/gemma2

jmorganca/add-missing-symlink-eval

mxyng/next-debug

parth/set-context-size-openai

brucemacd/next-bpe-bench

brucemacd/next-bpe-test

brucemacd/new_runner_e2e

brucemacd/new_runner_qwen2

pdevine/convert-cohere2

brucemacd/convert-cli

parth/log-probs

mxyng/next-mlx

mxyng/cmd-history

parth/templating

parth/tokenize-detokenize

brucemacd/check-key-register

bmizerany/grammar

jmorganca/vendor-081b29bd

mxyng/func-checks

jmorganca/fix-null-format

parth/fix-default-to-warn-json

jmorganca/qwen2vl

jmorganca/no-concat

parth/cmd-cleanup-SO

brucemacd/check-key-register-structured-err

parth/openai-stream-usage

parth/fix-referencing-so

stream-tools-stop

jmorganca/degin-1

brucemacd/install-path-clean

brucemacd/push-name-validation

brucemacd/browser-key-register

jmorganca/openai-fix-first-message

jmorganca/fix-proxy

jessegross/sample

parth/disallow-streaming-tools

dhiltgen/remove_submodule

jmorganca/ga

jmorganca/mllama

pdevine/newlines

pdevine/geems-2b

jmorganca/llama-bump

mxyng/modelname-7

mxyng/gin-slog

mxyng/modelname-6

jyan/convert-prog

jyan/quant5

paligemma-support

pdevine/import-docs

jmorganca/openai-context

jyan/paligemma

jyan/p2

jyan/palitest

bmizerany/embedspeedup

jmorganca/llama-vit

brucemacd/allow-ollama

royh/ep-methods

royh/whisper

mxyng/api-models

mxyng/fix-memory

jyan/q4_4/8

jyan/ollama-v

royh/stream-tools

roy-embed-parallel

bmizerany/hrm

revert-5963-revert-5924-mxyng/llama3.1-rope

royh/embed-viz

jyan/local2

jyan/auth

jyan/local

jyan/parse-temp

jmorganca/template-mistral

jyan/reord-g

royh-openai-suffixdocs

royh-imgembed

royh-embed-parallel

jyan/quant4

royh-precision

jyan/progress

pdevine/fix-template

jyan/quant3

pdevine/ggla

mxyng/update-registry-domain

jmorganca/ggml-static

mxyng/create-context

jyan/v0.146

mxyng/layers-from-files

build_dist

bmizerany/noseek

royh-ls

royh-name

timeout

mxyng/server-timestamp

bmizerany/nosillyggufslurps

royh-params

jmorganca/llama-cpp-7c26775

royh-openai-delete

royh-show-rigid

jmorganca/enable-fa

jmorganca/no-error-template

jyan/format

royh-testdelete

bmizerany/fastverify

language_support

pdevine/ps-glitches

brucemacd/tokenize

bruce/iq-quants

bmizerany/filepathwithcoloninhost

mxyng/split-bin

bmizerany/client-registry

jmorganca/if-none-match

native

jmorganca/native

jmorganca/batch-embeddings

jmorganca/initcmake

jmorganca/mm

pdevine/showggmlinfo

modenameenforcealphanum

bmizerany/modenameenforcealphanum

jmorganca/done-reason

jmorganca/llama-cpp-8960fe8

ollama.com

bmizerany/filepathnobuild

bmizerany/types/model/defaultfix

rmdisplaylong

nogogen

bmizerany/x

modelfile-readme

bmizerany/replacecolon

jmorganca/limit

jmorganca/execstack

jmorganca/replace-assets

mxyng/tune-concurrency

jmorganca/testing

whitespace-detection

jmorganca/options

upgrade-all

scratch

cuda-search

mattw/airenamer

mattw/allmodelsonhuggingface

mattw/quantcontext

mattw/whatneedstorun

brucemacd/llama-mem-calc

mattw/faq-context

mattw/communitylinks

mattw/noprune

mattw/python-functioncalling

rename

mxyng/install

pulse

remove-first

editor

mattw/selfqueryingretrieval

cgo

mattw/howtoquant

api

matt/streamingapi

format-config

mxyng/extra-args

shell

update-nous-hermes

cp-model

upload-progress

fix-unknown-model

fix-model-names

delete-fix

insecure-registry

ls

deletemodels

progressbar

readme-updates

license-layers

skip-list

list-models

modelpath

matt/examplemodelfiles

distribution

go-opts

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: github-starred/ollama#71237