github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 16:40:08 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #13085] Is AMD's AI Pro R9700 supported? #55174

New Issue
Closed
opened 2026-04-29 08:26:57 -05:00 by GiteaMirror · 8 comments
GiteaMirror commented 2026-04-29 08:26:57 -05:00
Owner
Copy Link

Originally created by @ChenxiWu-Lab on GitHub (Nov 14, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/13085

What is the issue?

I purchased an R9700 graphics card, thinking I could use its large VRAM to run larger models. However, after successfully installing Ollam and having a working ROCM output, I was initially able to perform inference on the GPU. Then, when I opened Dify and used the workflow, Ollam inexplicably reverted to CPU mode. After restarting Ollam, it couldn't detect the AMD GPU. Is this due to incorrect operation, a bug, or simply a lack of support?

Relevant log output

OS

Linux（ubuntu24.4.3）

GPU

AMD

CPU

AMD

Ollama version

0.12.11

Originally created by @ChenxiWu-Lab on GitHub (Nov 14, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/13085 ### What is the issue? I purchased an R9700 graphics card, thinking I could use its large VRAM to run larger models. However, after successfully installing Ollam and having a working ROCM output, I was initially able to perform inference on the GPU. Then, when I opened Dify and used the workflow, Ollam inexplicably reverted to CPU mode. After restarting Ollam, it couldn't detect the AMD GPU. Is this due to incorrect operation, a bug, or simply a lack of support? ### Relevant log output ```shell ``` ### OS Linux（ubuntu24.4.3） ### GPU AMD ### CPU AMD ### Ollama version 0.12.11
GiteaMirror added the bug label 2026-04-29 08:26:57 -05:00
GiteaMirror commented 2026-04-29 08:27:01 -05:00
Author
Owner
Copy Link

@ChenxiWu-Lab commented on GitHub (Nov 14, 2025):

I came across this discussion while looking at other issues. My situation is somewhat similar to theirs; it reverts to CPU mode. I'm currently using versions 0.12.11 and 0.12.6, and I'll try reverting to an earlier version to see if the problem persists.
https://github.com/ollama/ollama/issues/13061

<!-- gh-comment-id:3532664585 --> @ChenxiWu-Lab commented on GitHub (Nov 14, 2025): I came across this discussion while looking at other issues. My situation is somewhat similar to theirs; it reverts to CPU mode. I'm currently using versions 0.12.11 and 0.12.6, and I'll try reverting to an earlier version to see if the problem persists. [https://github.com/ollama/ollama/issues/13061](url)
GiteaMirror commented 2026-04-29 08:27:02 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Nov 14, 2025):

Server log may help in debugging.

<!-- gh-comment-id:3532923086 --> @rick-github commented on GitHub (Nov 14, 2025): [Server log](https://docs.ollama.com/troubleshooting) may help in debugging.
GiteaMirror commented 2026-04-29 08:27:03 -05:00
Author
Owner
Copy Link

@ChenxiWu-Lab commented on GitHub (Nov 14, 2025):

Thanks for the reminder. Below is the output log. I tried several versions from 0.12.0 to 0.12.11. The trigger mechanism is that the Ollama model is called from Dify; the first inference succeeds, but the second inference fails to call Ollama. Using ollama run in the command line also fails to load the model.

However, the Ollama service status shows everything is normal?

ollama.service - Ollama Service

Loaded: loaded (/etc/systemd/system/ollama.service; enabled; preset: enabled)

Drop-In: /etc/systemd/system/ollama.service.d
└─override.conf
Active: active (running) since Fri 2025-11-14 14:36:01 UTC; 16min ago
Main PID: 2898 (ollama)
Tasks: 29 (limit: 76795)
Memory: 32.4G (peak: 34.5G)
CPU: 3min 874ms
CGroup: /system.slice/ollama.service
├─ 2898 /usr/local/bin/ollama serve
└─37448 "[ollama]"

Nov 14 14:48:14 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:14 | 200 | 13.837µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 21.019µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 13.145µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 41.93µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 13.615µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:21 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:21 | 200 | 33.764µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:21 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:21 | 200 | 14.187µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:51:08 chenxilab ollama[2898]: time=2025-11-14T14:51:08.971Z level=INFO source=sched.go:438 msg="Load failed" model=/usr/share/ollama/.oll>
Nov 14 14:52:01 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:52:01 | 200 | 29.215µs | 127.0.0.1 | HEAD "/"
Nov 14 14:52:01 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:52:01 | 200 | 18.996µs | 127.0.0.1 | GET "/api/ps"

chenxi@chenxilab:~$ journalctl -u ollama --no-pager --follow --pager-end

Nov 14 14:36:01 chenxilab systemd[1]: Started ollama.service - Ollama Service.
Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.685Z level=INFO source=routes.go:1475 msg="server config" env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:4096 OLLAMA_DEBUG:INFO OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://0.0.0.0:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:/usr/share/ollama/.ollama/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:1 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_REMOTES:[ollama.com] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]"
Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.691Z level=INFO source=images.go:518 msg="total blobs: 25"
Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.692Z level=INFO source=images.go:525 msg="total unused blobs removed: 0"
Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.692Z level=INFO source=routes.go:1528 msg="Listening on [::]:11434 (version 0.12.3)"
Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.694Z level=INFO source=gpu.go:217 msg="looking for compatible GPUs"
Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.719Z level=INFO source=amd_linux.go:390 msg="amdgpu is supported" gpu=GPU-db3e3c1239ae7828 gpu_type=gfx1201
Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.724Z level=INFO source=types.go:131 msg="inference compute" id=GPU-db3e3c1239ae7828 library=rocm variant="" compute=gfx1201 driver=6.16 name=1002:7551 total="31.9 GiB" available="31.8 GiB"
Nov 14 14:36:35 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:36:35 | 200 | 1.373975ms | 127.0.0.1 | GET "/api/version"
Nov 14 14:42:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:42:15 | 200 | 38.302µs | 127.0.0.1 | HEAD "/"
Nov 14 14:42:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:42:15 | 200 | 5.668979ms | 127.0.0.1 | GET "/api/tags"
Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.203Z level=INFO source=server.go:399 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-ad7f12578413f17ce506bbf2809c3c7b8fd27bb1cf81acb3ac133a7538fc4259 --port 46493"
Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.204Z level=INFO source=server.go:672 msg="loading model" "model layers"=37 requested=-1
Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.204Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="58.1 GiB" free_swap="8.0 GiB"
Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.204Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="31.4 GiB" free="31.8 GiB" minimum="457.0 MiB" overhead="0 B"
Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.216Z level=INFO source=runner.go:1252 msg="starting ollama engine"
Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.217Z level=INFO source=runner.go:1287 msg="Server listening on 127.0.0.1:46493"
Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.227Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.277Z level=INFO source=ggml.go:131 msg="" architecture=qwen3 file_type=F16 name="Qwen3 4B Instruct 2507" description="" num_tensors=398 num_key_values=33
Nov 14 14:42:27 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
Nov 14 14:42:27 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
Nov 14 14:42:27 chenxilab ollama[2898]: ggml_cuda_init: found 1 ROCm devices:
Nov 14 14:42:27 chenxilab ollama[2898]: Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32, ID: GPU-db3e3c1239ae7828
Nov 14 14:42:27 chenxilab ollama[2898]: load_backend: loaded ROCm backend from /usr/local/lib/ollama/libggml-hip.so
Nov 14 14:42:27 chenxilab ollama[2898]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-haswell.so
Nov 14 14:42:27 chenxilab ollama[2898]: time=2025-11-14T14:42:27.773Z level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 ROCm.0.NO_VMM=1 ROCm.0.PEER_MAX_BATCH_SIZE=128 compiler=cgo(gcc)
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.041Z level=INFO source=runner.go:1171 msg=load request="{Operation:alloc LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=runner.go:1171 msg=load request="{Operation:commit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=ggml.go:487 msg="offloading 36 repeating layers to GPU"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=ggml.go:493 msg="offloading output layer to GPU"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=ggml.go:498 msg="offloaded 37/37 layers to GPU"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:310 msg="model weights" device=ROCm0 size="7.5 GiB"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:315 msg="model weights" device=CPU size="741.9 MiB"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:321 msg="kv cache" device=ROCm0 size="576.0 MiB"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:332 msg="compute graph" device=ROCm0 size="290.0 MiB"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:337 msg="compute graph" device=CPU size="5.0 MiB"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:342 msg="total memory" size="9.1 GiB"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=sched.go:470 msg="loaded runners" count=1
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=server.go:1251 msg="waiting for llama runner to start responding"
Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.109Z level=INFO source=server.go:1285 msg="waiting for server to become available" status="llm server loading model"
Nov 14 14:42:32 chenxilab ollama[2898]: time=2025-11-14T14:42:32.622Z level=INFO source=server.go:1289 msg="llama runner started in 6.42 seconds"
Nov 14 14:42:32 chenxilab ollama[2898]: rocblaslt error: Cannot read /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat: No such file or directory
Nov 14 14:42:32 chenxilab ollama[2898]: rocblaslt error: Could not load /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat
Nov 14 14:42:32 chenxilab ollama[2898]: hipBLASLt error: Heuristic Fetch Failed!
Nov 14 14:42:32 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_HIPBLASLT_ERROR environment variable is set.
Nov 14 14:42:32 chenxilab ollama[2898]: rocBLAS warning: hipBlasLT failed, falling back to tensile.
Nov 14 14:42:32 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_TENSILE_ERROR environment variable is set.
Nov 14 14:42:32 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:42:32 | 200 | 6.906026001s | 172.18.0.2 | POST "/api/chat"
Nov 14 14:42:36 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:42:36 | 200 | 20.449µs | 127.0.0.1 | HEAD "/"
Nov 14 14:42:36 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:42:36 | 200 | 77.076µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:43:04 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:43:04 | 400 | 126.309µs | 172.18.0.2 | POST "/api/chat"
Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.802Z level=INFO source=sched.go:537 msg="updated VRAM based on existing loaded models" gpu=GPU-db3e3c1239ae7828 library=rocm total="31.9 GiB" available="23.1 GiB"
Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.874Z level=INFO source=server.go:399 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-0c5b43627b54c17f10803d059a5e70dfafb797273440465273dcf54abba933ee --port 40239"
Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.874Z level=INFO source=server.go:672 msg="loading model" "model layers"=37 requested=-1
Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.874Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="56.9 GiB" free_swap="8.0 GiB"
Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.874Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="22.7 GiB" free="23.1 GiB" minimum="457.0 MiB" overhead="0 B"
Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.887Z level=INFO source=runner.go:1252 msg="starting ollama engine"
Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.887Z level=INFO source=runner.go:1287 msg="Server listening on 127.0.0.1:40239"
Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.897Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.934Z level=INFO source=ggml.go:131 msg="" architecture=qwen3 file_type=F16 name="Qwen3 4B Thinking 2507" description="" num_tensors=398 num_key_values=33
Nov 14 14:43:46 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
Nov 14 14:43:46 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
Nov 14 14:43:46 chenxilab ollama[2898]: ggml_cuda_init: found 1 ROCm devices:
Nov 14 14:43:46 chenxilab ollama[2898]: Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32, ID: GPU-db3e3c1239ae7828
Nov 14 14:43:46 chenxilab ollama[2898]: load_backend: loaded ROCm backend from /usr/local/lib/ollama/libggml-hip.so
Nov 14 14:43:46 chenxilab ollama[2898]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-haswell.so
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.416Z level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 ROCm.0.NO_VMM=1 ROCm.0.PEER_MAX_BATCH_SIZE=128 compiler=cgo(gcc)
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.657Z level=INFO source=runner.go:1171 msg=load request="{Operation:alloc LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.715Z level=INFO source=runner.go:1171 msg=load request="{Operation:commit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.715Z level=INFO source=ggml.go:487 msg="offloading 36 repeating layers to GPU"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.715Z level=INFO source=ggml.go:493 msg="offloading output layer to GPU"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.715Z level=INFO source=ggml.go:498 msg="offloaded 37/37 layers to GPU"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:310 msg="model weights" device=ROCm0 size="7.5 GiB"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:315 msg="model weights" device=CPU size="741.9 MiB"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:321 msg="kv cache" device=ROCm0 size="576.0 MiB"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:332 msg="compute graph" device=ROCm0 size="290.0 MiB"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:337 msg="compute graph" device=CPU size="5.0 MiB"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:342 msg="total memory" size="9.1 GiB"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=sched.go:470 msg="loaded runners" count=2
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=server.go:1251 msg="waiting for llama runner to start responding"
Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=server.go:1285 msg="waiting for server to become available" status="llm server loading model"
Nov 14 14:43:50 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:43:50 | 200 | 32.041µs | 127.0.0.1 | HEAD "/"
Nov 14 14:43:50 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:43:50 | 200 | 44.083µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:43:50 chenxilab ollama[2898]: time=2025-11-14T14:43:50.980Z level=INFO source=server.go:1289 msg="llama runner started in 6.11 seconds"
Nov 14 14:43:51 chenxilab ollama[2898]: rocblaslt error: Cannot read /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat: No such file or directory
Nov 14 14:43:51 chenxilab ollama[2898]: rocblaslt error: Could not load /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat
Nov 14 14:43:51 chenxilab ollama[2898]: hipBLASLt error: Heuristic Fetch Failed!
Nov 14 14:43:51 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_HIPBLASLT_ERROR environment variable is set.
Nov 14 14:43:51 chenxilab ollama[2898]: rocBLAS warning: hipBlasLT failed, falling back to tensile.
Nov 14 14:43:51 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_TENSILE_ERROR environment variable is set.
Nov 14 14:43:51 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:43:51 | 200 | 6.537297932s | 172.18.0.2 | POST "/api/chat"
Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.127Z level=INFO source=sched.go:537 msg="updated VRAM based on existing loaded models" gpu=GPU-db3e3c1239ae7828 library=rocm total="31.9 GiB" available="14.5 GiB"
Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.197Z level=INFO source=server.go:399 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-58574f2e94b99fb9e4391408b57e5aeaaaec10f6384e9a699fc2cb43a5c8eabf --port 41129"
Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.198Z level=INFO source=server.go:672 msg="loading model" "model layers"=49 requested=-1
Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.198Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="55.8 GiB" free_swap="8.0 GiB"
Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.198Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="14.0 GiB" free="14.5 GiB" minimum="457.0 MiB" overhead="0 B"
Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.210Z level=INFO source=runner.go:1252 msg="starting ollama engine"
Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.210Z level=INFO source=runner.go:1287 msg="Server listening on 127.0.0.1:41129"
Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.221Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.256Z level=INFO source=ggml.go:131 msg="" architecture=qwen3moe file_type=Q4_K_M name="Qwen3 30B A3B Thinking 2507" description="" num_tensors=579 num_key_values=33
Nov 14 14:44:39 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
Nov 14 14:44:39 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
Nov 14 14:44:39 chenxilab ollama[2898]: ggml_cuda_init: found 1 ROCm devices:
Nov 14 14:44:39 chenxilab ollama[2898]: Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32, ID: GPU-db3e3c1239ae7828
Nov 14 14:44:39 chenxilab ollama[2898]: load_backend: loaded ROCm backend from /usr/local/lib/ollama/libggml-hip.so
Nov 14 14:44:39 chenxilab ollama[2898]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-haswell.so
Nov 14 14:44:39 chenxilab ollama[2898]: time=2025-11-14T14:44:39.770Z level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 ROCm.0.NO_VMM=1 ROCm.0.PEER_MAX_BATCH_SIZE=128 compiler=cgo(gcc)
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=runner.go:1171 msg=load request="{Operation:close LoraPath:[] Parallel:0 BatchSize:0 FlashAttention:false KvSize:0 KvCacheType: NumThreads:0 GPULayers:[] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:310 msg="model weights" device=ROCm0 size="17.1 GiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:315 msg="model weights" device=CPU size="166.9 MiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:321 msg="kv cache" device=ROCm0 size="384.0 MiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:332 msg="compute graph" device=ROCm0 size="288.0 MiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:337 msg="compute graph" device=CPU size="4.0 MiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:342 msg="total memory" size="17.9 GiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.135Z level=ERROR source=server.go:425 msg="llama runner terminated" error="signal: killed"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.294Z level=INFO source=sched.go:537 msg="updated VRAM based on existing loaded models" gpu=GPU-db3e3c1239ae7828 library=rocm total="31.9 GiB" available="22.9 GiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.338Z level=INFO source=server.go:672 msg="loading model" "model layers"=49 requested=-1
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.338Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="56.6 GiB" free_swap="8.0 GiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.338Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="22.5 GiB" free="22.9 GiB" minimum="457.0 MiB" overhead="0 B"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.339Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.392Z level=INFO source=runner.go:1171 msg=load request="{Operation:alloc LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=runner.go:1171 msg=load request="{Operation:commit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=ggml.go:487 msg="offloading 48 repeating layers to GPU"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=ggml.go:493 msg="offloading output layer to GPU"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=ggml.go:498 msg="offloaded 49/49 layers to GPU"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:310 msg="model weights" device=ROCm0 size="17.1 GiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:315 msg="model weights" device=CPU size="166.9 MiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:321 msg="kv cache" device=ROCm0 size="384.0 MiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:332 msg="compute graph" device=ROCm0 size="288.0 MiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:337 msg="compute graph" device=CPU size="4.0 MiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:342 msg="total memory" size="17.9 GiB"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=sched.go:470 msg="loaded runners" count=2
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=server.go:1251 msg="waiting for llama runner to start responding"
Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=server.go:1285 msg="waiting for server to become available" status="llm server loading model"
Nov 14 14:44:43 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:43 | 200 | 30.237µs | 127.0.0.1 | HEAD "/"
Nov 14 14:44:43 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:43 | 200 | 54.513µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:44:48 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:48 | 200 | 30.818µs | 127.0.0.1 | HEAD "/"
Nov 14 14:44:48 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:48 | 200 | 50.395µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:44:48 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:48 | 200 | 30.799µs | 127.0.0.1 | HEAD "/"
Nov 14 14:44:48 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:48 | 200 | 24.075µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:44:49 chenxilab ollama[2898]: time=2025-11-14T14:44:49.488Z level=INFO source=server.go:1289 msg="llama runner started in 11.29 seconds"
Nov 14 14:44:49 chenxilab ollama[2898]: rocblaslt error: Cannot read /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat: No such file or directory
Nov 14 14:44:49 chenxilab ollama[2898]: rocblaslt error: Could not load /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat
Nov 14 14:44:49 chenxilab ollama[2898]: hipBLASLt error: Heuristic Fetch Failed!
Nov 14 14:44:49 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_HIPBLASLT_ERROR environment variable is set.
Nov 14 14:44:49 chenxilab ollama[2898]: rocBLAS warning: hipBlasLT failed, falling back to tensile.
Nov 14 14:44:49 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_TENSILE_ERROR environment variable is set.
Nov 14 14:44:49 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:49 | 200 | 24.807µs | 127.0.0.1 | HEAD "/"
Nov 14 14:44:49 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:49 | 200 | 31.489µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:44:49 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:49 | 200 | 11.714420993s | 172.18.0.2 | POST "/api/chat"
Nov 14 14:44:50 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:50 | 200 | 20.749µs | 127.0.0.1 | HEAD "/"
Nov 14 14:44:50 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:50 | 200 | 35.597µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:45:06 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:06 | 200 | 179.124805ms | 172.18.0.2 | POST "/api/chat"
Nov 14 14:45:49 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:49 | 200 | 30.467µs | 127.0.0.1 | HEAD "/"
Nov 14 14:45:49 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:49 | 200 | 39.695µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:45:50 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:50 | 200 | 9.357465045s | 172.18.0.2 | POST "/api/chat"
Nov 14 14:45:50 chenxilab ollama[2898]: time=2025-11-14T14:45:50.121Z level=ERROR source=server.go:425 msg="llama runner terminated" error="signal: killed"
Nov 14 14:45:51 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:51 | 200 | 27.472µs | 127.0.0.1 | HEAD "/"
Nov 14 14:45:51 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:51 | 200 | 33.072µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.241Z level=INFO source=sched.go:537 msg="updated VRAM based on existing loaded models" gpu=GPU-db3e3c1239ae7828 library=rocm total="31.9 GiB" available="23.1 GiB"
Nov 14 14:45:56 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:56 | 200 | 38.803µs | 127.0.0.1 | HEAD "/"
Nov 14 14:45:56 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:56 | 200 | 80.433µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.306Z level=INFO source=server.go:399 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-58574f2e94b99fb9e4391408b57e5aeaaaec10f6384e9a699fc2cb43a5c8eabf --port 39265"
Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.306Z level=INFO source=server.go:672 msg="loading model" "model layers"=49 requested=-1
Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.306Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="56.8 GiB" free_swap="8.0 GiB"
Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.306Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="22.7 GiB" free="23.1 GiB" minimum="457.0 MiB" overhead="0 B"
Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.319Z level=INFO source=runner.go:1252 msg="starting ollama engine"
Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.320Z level=INFO source=runner.go:1287 msg="Server listening on 127.0.0.1:39265"
Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.329Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.361Z level=INFO source=ggml.go:131 msg="" architecture=qwen3moe file_type=Q4_K_M name="Qwen3 30B A3B Thinking 2507" description="" num_tensors=579 num_key_values=33
Nov 14 14:45:57 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
Nov 14 14:45:57 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
Nov 14 14:45:57 chenxilab ollama[2898]: ggml_cuda_init: found 1 ROCm devices:
Nov 14 14:45:57 chenxilab ollama[2898]: Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32, ID: GPU-db3e3c1239ae7828
Nov 14 14:45:57 chenxilab ollama[2898]: load_backend: loaded ROCm backend from /usr/local/lib/ollama/libggml-hip.so
Nov 14 14:45:57 chenxilab ollama[2898]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-haswell.so
Nov 14 14:45:57 chenxilab ollama[2898]: time=2025-11-14T14:45:57.852Z level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 ROCm.0.NO_VMM=1 ROCm.0.PEER_MAX_BATCH_SIZE=128 compiler=cgo(gcc)
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.102Z level=INFO source=runner.go:1171 msg=load request="{Operation:alloc LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.209Z level=INFO source=runner.go:1171 msg=load request="{Operation:commit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=ggml.go:487 msg="offloading 48 repeating layers to GPU"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=ggml.go:493 msg="offloading output layer to GPU"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=ggml.go:498 msg="offloaded 49/49 layers to GPU"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:310 msg="model weights" device=ROCm0 size="17.1 GiB"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:315 msg="model weights" device=CPU size="166.9 MiB"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:321 msg="kv cache" device=ROCm0 size="384.0 MiB"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:332 msg="compute graph" device=ROCm0 size="288.0 MiB"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:337 msg="compute graph" device=CPU size="4.0 MiB"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:342 msg="total memory" size="17.9 GiB"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=sched.go:470 msg="loaded runners" count=2
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=server.go:1251 msg="waiting for llama runner to start responding"
Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=server.go:1285 msg="waiting for server to become available" status="llm server loading model"
Nov 14 14:46:00 chenxilab ollama[2898]: time=2025-11-14T14:46:00.219Z level=INFO source=server.go:1289 msg="llama runner started in 3.91 seconds"
Nov 14 14:46:00 chenxilab ollama[2898]: rocblaslt error: Cannot read /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat: No such file or directory
Nov 14 14:46:00 chenxilab ollama[2898]: rocblaslt error: Could not load /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat
Nov 14 14:46:00 chenxilab ollama[2898]: hipBLASLt error: Heuristic Fetch Failed!
Nov 14 14:46:00 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_HIPBLASLT_ERROR environment variable is set.
Nov 14 14:46:00 chenxilab ollama[2898]: rocBLAS warning: hipBlasLT failed, falling back to tensile.
Nov 14 14:46:00 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_TENSILE_ERROR environment variable is set.
Nov 14 14:46:08 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:46:08 | 200 | 12.756089909s | 172.18.0.2 | POST "/api/chat"
Nov 14 14:46:08 chenxilab ollama[2898]: time=2025-11-14T14:46:08.978Z level=ERROR source=server.go:425 msg="llama runner terminated" error="signal: killed"
Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.057Z level=INFO source=sched.go:537 msg="updated VRAM based on existing loaded models" gpu=GPU-db3e3c1239ae7828 library=rocm total="31.9 GiB" available="5.1 GiB"
Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.147Z level=INFO source=server.go:399 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-58574f2e94b99fb9e4391408b57e5aeaaaec10f6384e9a699fc2cb43a5c8eabf --port 36785"
Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.148Z level=INFO source=server.go:672 msg="loading model" "model layers"=49 requested=-1
Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.148Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="56.7 GiB" free_swap="8.0 GiB"
Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.148Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="4.6 GiB" free="5.1 GiB" minimum="457.0 MiB" overhead="0 B"
Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.160Z level=INFO source=runner.go:1252 msg="starting ollama engine"
Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.160Z level=INFO source=runner.go:1287 msg="Server listening on 127.0.0.1:36785"
Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.170Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}"
Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.219Z level=INFO source=ggml.go:131 msg="" architecture=qwen3moe file_type=Q4_K_M name="Qwen3 30B A3B Thinking 2507" description="" num_tensors=579 num_key_values=33
Nov 14 14:46:10 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:46:10 | 200 | 20.99µs | 127.0.0.1 | HEAD "/"
Nov 14 14:46:10 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:46:10 | 200 | 20.088µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:46:11 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:46:11 | 200 | 22.051µs | 127.0.0.1 | HEAD "/"
Nov 14 14:46:11 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:46:11 | 200 | 19.056µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:47:33 chenxilab ollama[2898]: time=2025-11-14T14:47:33.002Z level=ERROR source=server.go:425 msg="llama runner terminated" error="signal: killed"
Nov 14 14:47:37 chenxilab ollama[2898]: time=2025-11-14T14:47:37.888Z level=WARN source=sched.go:649 msg="gpu VRAM usage didn't recover within timeout" seconds=5.001477841 runner.size="9.1 GiB" runner.vram="9.1 GiB" runner.parallel=1 runner.pid=31658 runner.model=/usr/share/ollama/.ollama/models/blobs/sha256-ad7f12578413f17ce506bbf2809c3c7b8fd27bb1cf81acb3ac133a7538fc4259
Nov 14 14:47:38 chenxilab ollama[2898]: time=2025-11-14T14:47:38.138Z level=WARN source=sched.go:649 msg="gpu VRAM usage didn't recover within timeout" seconds=5.251149624 runner.size="9.1 GiB" runner.vram="9.1 GiB" runner.parallel=1 runner.pid=31658 runner.model=/usr/share/ollama/.ollama/models/blobs/sha256-ad7f12578413f17ce506bbf2809c3c7b8fd27bb1cf81acb3ac133a7538fc4259
Nov 14 14:47:38 chenxilab ollama[2898]: time=2025-11-14T14:47:38.388Z level=WARN source=sched.go:649 msg="gpu VRAM usage didn't recover within timeout" seconds=5.500749635 runner.size="9.1 GiB" runner.vram="9.1 GiB" runner.parallel=1 runner.pid=31658 runner.model=/usr/share/ollama/.ollama/models/blobs/sha256-ad7f12578413f17ce506bbf2809c3c7b8fd27bb1cf81acb3ac133a7538fc4259
Nov 14 14:48:00 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:00 | 200 | 32.13µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:00 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:00 | 200 | 19.207µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:11 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:11 | 200 | 28.454µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:11 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:11 | 200 | 64.582µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:12 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:12 | 200 | 28.635µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:12 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:12 | 200 | 12.523µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:13 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:13 | 200 | 26.34µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:13 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:13 | 200 | 48.952µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:13 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:13 | 200 | 33.693µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:13 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:13 | 200 | 10.179µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:14 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:14 | 200 | 25.849µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:14 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:14 | 200 | 13.837µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 21.019µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 13.145µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 41.93µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 13.615µs | 127.0.0.1 | GET "/api/ps"
Nov 14 14:48:21 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:21 | 200 | 33.764µs | 127.0.0.1 | HEAD "/"
Nov 14 14:48:21 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:21 | 200 | 14.187µs | 127.0.0.1 | GET "/api/ps"

<!-- gh-comment-id:3533152033 --> @ChenxiWu-Lab commented on GitHub (Nov 14, 2025): Thanks for the reminder. Below is the output log. I tried several versions from 0.12.0 to 0.12.11. The trigger mechanism is that the Ollama model is called from Dify; the first inference succeeds, but the second inference fails to call Ollama. Using `ollama run` in the command line also fails to load the model. However, the Ollama service status shows everything is normal? ollama.service - Ollama Service Loaded: loaded (/etc/systemd/system/ollama.service; enabled; preset: enabled) Drop-In: /etc/systemd/system/ollama.service.d └─override.conf Active: active (running) since Fri 2025-11-14 14:36:01 UTC; 16min ago Main PID: 2898 (ollama) Tasks: 29 (limit: 76795) Memory: 32.4G (peak: 34.5G) CPU: 3min 874ms CGroup: /system.slice/ollama.service ├─ 2898 /usr/local/bin/ollama serve └─37448 "[ollama]" Nov 14 14:48:14 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:14 | 200 | 13.837µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 21.019µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 13.145µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 41.93µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 13.615µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:21 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:21 | 200 | 33.764µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:21 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:21 | 200 | 14.187µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:51:08 chenxilab ollama[2898]: time=2025-11-14T14:51:08.971Z level=INFO source=sched.go:438 msg="Load failed" model=/usr/share/ollama/.oll> Nov 14 14:52:01 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:52:01 | 200 | 29.215µs | 127.0.0.1 | HEAD "/" Nov 14 14:52:01 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:52:01 | 200 | 18.996µs | 127.0.0.1 | GET "/api/ps" chenxi@chenxilab:~$ journalctl -u ollama --no-pager --follow --pager-end Nov 14 14:36:01 chenxilab systemd[1]: Started ollama.service - Ollama Service. Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.685Z level=INFO source=routes.go:1475 msg="server config" env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:4096 OLLAMA_DEBUG:INFO OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://0.0.0.0:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:/usr/share/ollama/.ollama/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:1 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_REMOTES:[ollama.com] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]" Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.691Z level=INFO source=images.go:518 msg="total blobs: 25" Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.692Z level=INFO source=images.go:525 msg="total unused blobs removed: 0" Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.692Z level=INFO source=routes.go:1528 msg="Listening on [::]:11434 (version 0.12.3)" Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.694Z level=INFO source=gpu.go:217 msg="looking for compatible GPUs" Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.719Z level=INFO source=amd_linux.go:390 msg="amdgpu is supported" gpu=GPU-db3e3c1239ae7828 gpu_type=gfx1201 Nov 14 14:36:01 chenxilab ollama[2898]: time=2025-11-14T14:36:01.724Z level=INFO source=types.go:131 msg="inference compute" id=GPU-db3e3c1239ae7828 library=rocm variant="" compute=gfx1201 driver=6.16 name=1002:7551 total="31.9 GiB" available="31.8 GiB" Nov 14 14:36:35 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:36:35 | 200 | 1.373975ms | 127.0.0.1 | GET "/api/version" Nov 14 14:42:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:42:15 | 200 | 38.302µs | 127.0.0.1 | HEAD "/" Nov 14 14:42:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:42:15 | 200 | 5.668979ms | 127.0.0.1 | GET "/api/tags" Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.203Z level=INFO source=server.go:399 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-ad7f12578413f17ce506bbf2809c3c7b8fd27bb1cf81acb3ac133a7538fc4259 --port 46493" Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.204Z level=INFO source=server.go:672 msg="loading model" "model layers"=37 requested=-1 Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.204Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="58.1 GiB" free_swap="8.0 GiB" Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.204Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="31.4 GiB" free="31.8 GiB" minimum="457.0 MiB" overhead="0 B" Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.216Z level=INFO source=runner.go:1252 msg="starting ollama engine" Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.217Z level=INFO source=runner.go:1287 msg="Server listening on 127.0.0.1:46493" Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.227Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:42:26 chenxilab ollama[2898]: time=2025-11-14T14:42:26.277Z level=INFO source=ggml.go:131 msg="" architecture=qwen3 file_type=F16 name="Qwen3 4B Instruct 2507" description="" num_tensors=398 num_key_values=33 Nov 14 14:42:27 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no Nov 14 14:42:27 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no Nov 14 14:42:27 chenxilab ollama[2898]: ggml_cuda_init: found 1 ROCm devices: Nov 14 14:42:27 chenxilab ollama[2898]: Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32, ID: GPU-db3e3c1239ae7828 Nov 14 14:42:27 chenxilab ollama[2898]: load_backend: loaded ROCm backend from /usr/local/lib/ollama/libggml-hip.so Nov 14 14:42:27 chenxilab ollama[2898]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-haswell.so Nov 14 14:42:27 chenxilab ollama[2898]: time=2025-11-14T14:42:27.773Z level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 ROCm.0.NO_VMM=1 ROCm.0.PEER_MAX_BATCH_SIZE=128 compiler=cgo(gcc) Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.041Z level=INFO source=runner.go:1171 msg=load request="{Operation:alloc LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=runner.go:1171 msg=load request="{Operation:commit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=ggml.go:487 msg="offloading 36 repeating layers to GPU" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=ggml.go:493 msg="offloading output layer to GPU" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=ggml.go:498 msg="offloaded 37/37 layers to GPU" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:310 msg="model weights" device=ROCm0 size="7.5 GiB" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:315 msg="model weights" device=CPU size="741.9 MiB" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:321 msg="kv cache" device=ROCm0 size="576.0 MiB" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:332 msg="compute graph" device=ROCm0 size="290.0 MiB" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:337 msg="compute graph" device=CPU size="5.0 MiB" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=backend.go:342 msg="total memory" size="9.1 GiB" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=sched.go:470 msg="loaded runners" count=1 Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.102Z level=INFO source=server.go:1251 msg="waiting for llama runner to start responding" Nov 14 14:42:28 chenxilab ollama[2898]: time=2025-11-14T14:42:28.109Z level=INFO source=server.go:1285 msg="waiting for server to become available" status="llm server loading model" Nov 14 14:42:32 chenxilab ollama[2898]: time=2025-11-14T14:42:32.622Z level=INFO source=server.go:1289 msg="llama runner started in 6.42 seconds" Nov 14 14:42:32 chenxilab ollama[2898]: rocblaslt error: Cannot read /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat: No such file or directory Nov 14 14:42:32 chenxilab ollama[2898]: rocblaslt error: Could not load /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat Nov 14 14:42:32 chenxilab ollama[2898]: hipBLASLt error: Heuristic Fetch Failed! Nov 14 14:42:32 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_HIPBLASLT_ERROR environment variable is set. Nov 14 14:42:32 chenxilab ollama[2898]: rocBLAS warning: hipBlasLT failed, falling back to tensile. Nov 14 14:42:32 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_TENSILE_ERROR environment variable is set. Nov 14 14:42:32 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:42:32 | 200 | 6.906026001s | 172.18.0.2 | POST "/api/chat" Nov 14 14:42:36 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:42:36 | 200 | 20.449µs | 127.0.0.1 | HEAD "/" Nov 14 14:42:36 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:42:36 | 200 | 77.076µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:43:04 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:43:04 | 400 | 126.309µs | 172.18.0.2 | POST "/api/chat" Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.802Z level=INFO source=sched.go:537 msg="updated VRAM based on existing loaded models" gpu=GPU-db3e3c1239ae7828 library=rocm total="31.9 GiB" available="23.1 GiB" Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.874Z level=INFO source=server.go:399 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-0c5b43627b54c17f10803d059a5e70dfafb797273440465273dcf54abba933ee --port 40239" Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.874Z level=INFO source=server.go:672 msg="loading model" "model layers"=37 requested=-1 Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.874Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="56.9 GiB" free_swap="8.0 GiB" Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.874Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="22.7 GiB" free="23.1 GiB" minimum="457.0 MiB" overhead="0 B" Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.887Z level=INFO source=runner.go:1252 msg="starting ollama engine" Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.887Z level=INFO source=runner.go:1287 msg="Server listening on 127.0.0.1:40239" Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.897Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:43:44 chenxilab ollama[2898]: time=2025-11-14T14:43:44.934Z level=INFO source=ggml.go:131 msg="" architecture=qwen3 file_type=F16 name="Qwen3 4B Thinking 2507" description="" num_tensors=398 num_key_values=33 Nov 14 14:43:46 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no Nov 14 14:43:46 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no Nov 14 14:43:46 chenxilab ollama[2898]: ggml_cuda_init: found 1 ROCm devices: Nov 14 14:43:46 chenxilab ollama[2898]: Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32, ID: GPU-db3e3c1239ae7828 Nov 14 14:43:46 chenxilab ollama[2898]: load_backend: loaded ROCm backend from /usr/local/lib/ollama/libggml-hip.so Nov 14 14:43:46 chenxilab ollama[2898]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-haswell.so Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.416Z level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 ROCm.0.NO_VMM=1 ROCm.0.PEER_MAX_BATCH_SIZE=128 compiler=cgo(gcc) Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.657Z level=INFO source=runner.go:1171 msg=load request="{Operation:alloc LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.715Z level=INFO source=runner.go:1171 msg=load request="{Operation:commit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:37[ID:GPU-db3e3c1239ae7828 Layers:37(0..36)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.715Z level=INFO source=ggml.go:487 msg="offloading 36 repeating layers to GPU" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.715Z level=INFO source=ggml.go:493 msg="offloading output layer to GPU" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.715Z level=INFO source=ggml.go:498 msg="offloaded 37/37 layers to GPU" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:310 msg="model weights" device=ROCm0 size="7.5 GiB" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:315 msg="model weights" device=CPU size="741.9 MiB" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:321 msg="kv cache" device=ROCm0 size="576.0 MiB" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:332 msg="compute graph" device=ROCm0 size="290.0 MiB" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:337 msg="compute graph" device=CPU size="5.0 MiB" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=backend.go:342 msg="total memory" size="9.1 GiB" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=sched.go:470 msg="loaded runners" count=2 Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=server.go:1251 msg="waiting for llama runner to start responding" Nov 14 14:43:46 chenxilab ollama[2898]: time=2025-11-14T14:43:46.716Z level=INFO source=server.go:1285 msg="waiting for server to become available" status="llm server loading model" Nov 14 14:43:50 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:43:50 | 200 | 32.041µs | 127.0.0.1 | HEAD "/" Nov 14 14:43:50 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:43:50 | 200 | 44.083µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:43:50 chenxilab ollama[2898]: time=2025-11-14T14:43:50.980Z level=INFO source=server.go:1289 msg="llama runner started in 6.11 seconds" Nov 14 14:43:51 chenxilab ollama[2898]: rocblaslt error: Cannot read /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat: No such file or directory Nov 14 14:43:51 chenxilab ollama[2898]: rocblaslt error: Could not load /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat Nov 14 14:43:51 chenxilab ollama[2898]: hipBLASLt error: Heuristic Fetch Failed! Nov 14 14:43:51 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_HIPBLASLT_ERROR environment variable is set. Nov 14 14:43:51 chenxilab ollama[2898]: rocBLAS warning: hipBlasLT failed, falling back to tensile. Nov 14 14:43:51 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_TENSILE_ERROR environment variable is set. Nov 14 14:43:51 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:43:51 | 200 | 6.537297932s | 172.18.0.2 | POST "/api/chat" Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.127Z level=INFO source=sched.go:537 msg="updated VRAM based on existing loaded models" gpu=GPU-db3e3c1239ae7828 library=rocm total="31.9 GiB" available="14.5 GiB" Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.197Z level=INFO source=server.go:399 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-58574f2e94b99fb9e4391408b57e5aeaaaec10f6384e9a699fc2cb43a5c8eabf --port 41129" Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.198Z level=INFO source=server.go:672 msg="loading model" "model layers"=49 requested=-1 Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.198Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="55.8 GiB" free_swap="8.0 GiB" Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.198Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="14.0 GiB" free="14.5 GiB" minimum="457.0 MiB" overhead="0 B" Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.210Z level=INFO source=runner.go:1252 msg="starting ollama engine" Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.210Z level=INFO source=runner.go:1287 msg="Server listening on 127.0.0.1:41129" Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.221Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:44:38 chenxilab ollama[2898]: time=2025-11-14T14:44:38.256Z level=INFO source=ggml.go:131 msg="" architecture=qwen3moe file_type=Q4_K_M name="Qwen3 30B A3B Thinking 2507" description="" num_tensors=579 num_key_values=33 Nov 14 14:44:39 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no Nov 14 14:44:39 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no Nov 14 14:44:39 chenxilab ollama[2898]: ggml_cuda_init: found 1 ROCm devices: Nov 14 14:44:39 chenxilab ollama[2898]: Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32, ID: GPU-db3e3c1239ae7828 Nov 14 14:44:39 chenxilab ollama[2898]: load_backend: loaded ROCm backend from /usr/local/lib/ollama/libggml-hip.so Nov 14 14:44:39 chenxilab ollama[2898]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-haswell.so Nov 14 14:44:39 chenxilab ollama[2898]: time=2025-11-14T14:44:39.770Z level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 ROCm.0.NO_VMM=1 ROCm.0.PEER_MAX_BATCH_SIZE=128 compiler=cgo(gcc) Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=runner.go:1171 msg=load request="{Operation:close LoraPath:[] Parallel:0 BatchSize:0 FlashAttention:false KvSize:0 KvCacheType: NumThreads:0 GPULayers:[] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:310 msg="model weights" device=ROCm0 size="17.1 GiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:315 msg="model weights" device=CPU size="166.9 MiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:321 msg="kv cache" device=ROCm0 size="384.0 MiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:332 msg="compute graph" device=ROCm0 size="288.0 MiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:337 msg="compute graph" device=CPU size="4.0 MiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.015Z level=INFO source=backend.go:342 msg="total memory" size="17.9 GiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.135Z level=ERROR source=server.go:425 msg="llama runner terminated" error="signal: killed" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.294Z level=INFO source=sched.go:537 msg="updated VRAM based on existing loaded models" gpu=GPU-db3e3c1239ae7828 library=rocm total="31.9 GiB" available="22.9 GiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.338Z level=INFO source=server.go:672 msg="loading model" "model layers"=49 requested=-1 Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.338Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="56.6 GiB" free_swap="8.0 GiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.338Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="22.5 GiB" free="22.9 GiB" minimum="457.0 MiB" overhead="0 B" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.339Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.392Z level=INFO source=runner.go:1171 msg=load request="{Operation:alloc LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=runner.go:1171 msg=load request="{Operation:commit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=ggml.go:487 msg="offloading 48 repeating layers to GPU" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=ggml.go:493 msg="offloading output layer to GPU" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=ggml.go:498 msg="offloaded 49/49 layers to GPU" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:310 msg="model weights" device=ROCm0 size="17.1 GiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:315 msg="model weights" device=CPU size="166.9 MiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:321 msg="kv cache" device=ROCm0 size="384.0 MiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:332 msg="compute graph" device=ROCm0 size="288.0 MiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:337 msg="compute graph" device=CPU size="4.0 MiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=backend.go:342 msg="total memory" size="17.9 GiB" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=sched.go:470 msg="loaded runners" count=2 Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=server.go:1251 msg="waiting for llama runner to start responding" Nov 14 14:44:40 chenxilab ollama[2898]: time=2025-11-14T14:44:40.460Z level=INFO source=server.go:1285 msg="waiting for server to become available" status="llm server loading model" Nov 14 14:44:43 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:43 | 200 | 30.237µs | 127.0.0.1 | HEAD "/" Nov 14 14:44:43 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:43 | 200 | 54.513µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:44:48 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:48 | 200 | 30.818µs | 127.0.0.1 | HEAD "/" Nov 14 14:44:48 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:48 | 200 | 50.395µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:44:48 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:48 | 200 | 30.799µs | 127.0.0.1 | HEAD "/" Nov 14 14:44:48 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:48 | 200 | 24.075µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:44:49 chenxilab ollama[2898]: time=2025-11-14T14:44:49.488Z level=INFO source=server.go:1289 msg="llama runner started in 11.29 seconds" Nov 14 14:44:49 chenxilab ollama[2898]: rocblaslt error: Cannot read /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat: No such file or directory Nov 14 14:44:49 chenxilab ollama[2898]: rocblaslt error: Could not load /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat Nov 14 14:44:49 chenxilab ollama[2898]: hipBLASLt error: Heuristic Fetch Failed! Nov 14 14:44:49 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_HIPBLASLT_ERROR environment variable is set. Nov 14 14:44:49 chenxilab ollama[2898]: rocBLAS warning: hipBlasLT failed, falling back to tensile. Nov 14 14:44:49 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_TENSILE_ERROR environment variable is set. Nov 14 14:44:49 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:49 | 200 | 24.807µs | 127.0.0.1 | HEAD "/" Nov 14 14:44:49 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:49 | 200 | 31.489µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:44:49 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:49 | 200 | 11.714420993s | 172.18.0.2 | POST "/api/chat" Nov 14 14:44:50 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:50 | 200 | 20.749µs | 127.0.0.1 | HEAD "/" Nov 14 14:44:50 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:44:50 | 200 | 35.597µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:45:06 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:06 | 200 | 179.124805ms | 172.18.0.2 | POST "/api/chat" Nov 14 14:45:49 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:49 | 200 | 30.467µs | 127.0.0.1 | HEAD "/" Nov 14 14:45:49 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:49 | 200 | 39.695µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:45:50 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:50 | 200 | 9.357465045s | 172.18.0.2 | POST "/api/chat" Nov 14 14:45:50 chenxilab ollama[2898]: time=2025-11-14T14:45:50.121Z level=ERROR source=server.go:425 msg="llama runner terminated" error="signal: killed" Nov 14 14:45:51 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:51 | 200 | 27.472µs | 127.0.0.1 | HEAD "/" Nov 14 14:45:51 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:51 | 200 | 33.072µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.241Z level=INFO source=sched.go:537 msg="updated VRAM based on existing loaded models" gpu=GPU-db3e3c1239ae7828 library=rocm total="31.9 GiB" available="23.1 GiB" Nov 14 14:45:56 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:56 | 200 | 38.803µs | 127.0.0.1 | HEAD "/" Nov 14 14:45:56 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:45:56 | 200 | 80.433µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.306Z level=INFO source=server.go:399 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-58574f2e94b99fb9e4391408b57e5aeaaaec10f6384e9a699fc2cb43a5c8eabf --port 39265" Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.306Z level=INFO source=server.go:672 msg="loading model" "model layers"=49 requested=-1 Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.306Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="56.8 GiB" free_swap="8.0 GiB" Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.306Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="22.7 GiB" free="23.1 GiB" minimum="457.0 MiB" overhead="0 B" Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.319Z level=INFO source=runner.go:1252 msg="starting ollama engine" Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.320Z level=INFO source=runner.go:1287 msg="Server listening on 127.0.0.1:39265" Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.329Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:45:56 chenxilab ollama[2898]: time=2025-11-14T14:45:56.361Z level=INFO source=ggml.go:131 msg="" architecture=qwen3moe file_type=Q4_K_M name="Qwen3 30B A3B Thinking 2507" description="" num_tensors=579 num_key_values=33 Nov 14 14:45:57 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no Nov 14 14:45:57 chenxilab ollama[2898]: ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no Nov 14 14:45:57 chenxilab ollama[2898]: ggml_cuda_init: found 1 ROCm devices: Nov 14 14:45:57 chenxilab ollama[2898]: Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32, ID: GPU-db3e3c1239ae7828 Nov 14 14:45:57 chenxilab ollama[2898]: load_backend: loaded ROCm backend from /usr/local/lib/ollama/libggml-hip.so Nov 14 14:45:57 chenxilab ollama[2898]: load_backend: loaded CPU backend from /usr/local/lib/ollama/libggml-cpu-haswell.so Nov 14 14:45:57 chenxilab ollama[2898]: time=2025-11-14T14:45:57.852Z level=INFO source=ggml.go:104 msg=system CPU.0.SSE3=1 CPU.0.SSSE3=1 CPU.0.AVX=1 CPU.0.AVX2=1 CPU.0.F16C=1 CPU.0.FMA=1 CPU.0.BMI2=1 CPU.0.LLAMAFILE=1 CPU.1.LLAMAFILE=1 ROCm.0.NO_VMM=1 ROCm.0.PEER_MAX_BATCH_SIZE=128 compiler=cgo(gcc) Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.102Z level=INFO source=runner.go:1171 msg=load request="{Operation:alloc LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.209Z level=INFO source=runner.go:1171 msg=load request="{Operation:commit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=ggml.go:487 msg="offloading 48 repeating layers to GPU" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=ggml.go:493 msg="offloading output layer to GPU" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=ggml.go:498 msg="offloaded 49/49 layers to GPU" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:310 msg="model weights" device=ROCm0 size="17.1 GiB" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:315 msg="model weights" device=CPU size="166.9 MiB" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:321 msg="kv cache" device=ROCm0 size="384.0 MiB" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:332 msg="compute graph" device=ROCm0 size="288.0 MiB" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:337 msg="compute graph" device=CPU size="4.0 MiB" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=backend.go:342 msg="total memory" size="17.9 GiB" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=sched.go:470 msg="loaded runners" count=2 Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=server.go:1251 msg="waiting for llama runner to start responding" Nov 14 14:45:58 chenxilab ollama[2898]: time=2025-11-14T14:45:58.210Z level=INFO source=server.go:1285 msg="waiting for server to become available" status="llm server loading model" Nov 14 14:46:00 chenxilab ollama[2898]: time=2025-11-14T14:46:00.219Z level=INFO source=server.go:1289 msg="llama runner started in 3.91 seconds" Nov 14 14:46:00 chenxilab ollama[2898]: rocblaslt error: Cannot read /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat: No such file or directory Nov 14 14:46:00 chenxilab ollama[2898]: rocblaslt error: Could not load /usr/local/lib/ollama/rocm/hipblaslt/library/TensileLibrary_lazy_gfx1201.dat Nov 14 14:46:00 chenxilab ollama[2898]: hipBLASLt error: Heuristic Fetch Failed! Nov 14 14:46:00 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_HIPBLASLT_ERROR environment variable is set. Nov 14 14:46:00 chenxilab ollama[2898]: rocBLAS warning: hipBlasLT failed, falling back to tensile. Nov 14 14:46:00 chenxilab ollama[2898]: This message will be only be displayed once, unless the ROCBLAS_VERBOSE_TENSILE_ERROR environment variable is set. Nov 14 14:46:08 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:46:08 | 200 | 12.756089909s | 172.18.0.2 | POST "/api/chat" Nov 14 14:46:08 chenxilab ollama[2898]: time=2025-11-14T14:46:08.978Z level=ERROR source=server.go:425 msg="llama runner terminated" error="signal: killed" Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.057Z level=INFO source=sched.go:537 msg="updated VRAM based on existing loaded models" gpu=GPU-db3e3c1239ae7828 library=rocm total="31.9 GiB" available="5.1 GiB" Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.147Z level=INFO source=server.go:399 msg="starting runner" cmd="/usr/local/bin/ollama runner --ollama-engine --model /usr/share/ollama/.ollama/models/blobs/sha256-58574f2e94b99fb9e4391408b57e5aeaaaec10f6384e9a699fc2cb43a5c8eabf --port 36785" Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.148Z level=INFO source=server.go:672 msg="loading model" "model layers"=49 requested=-1 Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.148Z level=INFO source=server.go:678 msg="system memory" total="62.6 GiB" free="56.7 GiB" free_swap="8.0 GiB" Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.148Z level=INFO source=server.go:686 msg="gpu memory" id=GPU-db3e3c1239ae7828 available="4.6 GiB" free="5.1 GiB" minimum="457.0 MiB" overhead="0 B" Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.160Z level=INFO source=runner.go:1252 msg="starting ollama engine" Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.160Z level=INFO source=runner.go:1287 msg="Server listening on 127.0.0.1:36785" Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.170Z level=INFO source=runner.go:1171 msg=load request="{Operation:fit LoraPath:[] Parallel:1 BatchSize:512 FlashAttention:false KvSize:4096 KvCacheType: NumThreads:24 GPULayers:49[ID:GPU-db3e3c1239ae7828 Layers:49(0..48)] MultiUserCache:false ProjectorPath: MainGPU:0 UseMmap:false}" Nov 14 14:46:09 chenxilab ollama[2898]: time=2025-11-14T14:46:09.219Z level=INFO source=ggml.go:131 msg="" architecture=qwen3moe file_type=Q4_K_M name="Qwen3 30B A3B Thinking 2507" description="" num_tensors=579 num_key_values=33 Nov 14 14:46:10 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:46:10 | 200 | 20.99µs | 127.0.0.1 | HEAD "/" Nov 14 14:46:10 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:46:10 | 200 | 20.088µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:46:11 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:46:11 | 200 | 22.051µs | 127.0.0.1 | HEAD "/" Nov 14 14:46:11 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:46:11 | 200 | 19.056µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:47:33 chenxilab ollama[2898]: time=2025-11-14T14:47:33.002Z level=ERROR source=server.go:425 msg="llama runner terminated" error="signal: killed" Nov 14 14:47:37 chenxilab ollama[2898]: time=2025-11-14T14:47:37.888Z level=WARN source=sched.go:649 msg="gpu VRAM usage didn't recover within timeout" seconds=5.001477841 runner.size="9.1 GiB" runner.vram="9.1 GiB" runner.parallel=1 runner.pid=31658 runner.model=/usr/share/ollama/.ollama/models/blobs/sha256-ad7f12578413f17ce506bbf2809c3c7b8fd27bb1cf81acb3ac133a7538fc4259 Nov 14 14:47:38 chenxilab ollama[2898]: time=2025-11-14T14:47:38.138Z level=WARN source=sched.go:649 msg="gpu VRAM usage didn't recover within timeout" seconds=5.251149624 runner.size="9.1 GiB" runner.vram="9.1 GiB" runner.parallel=1 runner.pid=31658 runner.model=/usr/share/ollama/.ollama/models/blobs/sha256-ad7f12578413f17ce506bbf2809c3c7b8fd27bb1cf81acb3ac133a7538fc4259 Nov 14 14:47:38 chenxilab ollama[2898]: time=2025-11-14T14:47:38.388Z level=WARN source=sched.go:649 msg="gpu VRAM usage didn't recover within timeout" seconds=5.500749635 runner.size="9.1 GiB" runner.vram="9.1 GiB" runner.parallel=1 runner.pid=31658 runner.model=/usr/share/ollama/.ollama/models/blobs/sha256-ad7f12578413f17ce506bbf2809c3c7b8fd27bb1cf81acb3ac133a7538fc4259 Nov 14 14:48:00 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:00 | 200 | 32.13µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:00 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:00 | 200 | 19.207µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:11 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:11 | 200 | 28.454µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:11 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:11 | 200 | 64.582µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:12 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:12 | 200 | 28.635µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:12 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:12 | 200 | 12.523µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:13 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:13 | 200 | 26.34µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:13 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:13 | 200 | 48.952µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:13 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:13 | 200 | 33.693µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:13 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:13 | 200 | 10.179µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:14 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:14 | 200 | 25.849µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:14 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:14 | 200 | 13.837µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 21.019µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 13.145µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 41.93µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:15 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:15 | 200 | 13.615µs | 127.0.0.1 | GET "/api/ps" Nov 14 14:48:21 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:21 | 200 | 33.764µs | 127.0.0.1 | HEAD "/" Nov 14 14:48:21 chenxilab ollama[2898]: [GIN] 2025/11/14 - 14:48:21 | 200 | 14.187µs | 127.0.0.1 | GET "/api/ps"
GiteaMirror commented 2026-04-29 08:27:04 -05:00
Author
Owner
Copy Link

@RockingRolli commented on GitHub (Nov 14, 2025):

It works with the Radeon Pro R9700... if you use the right kernel. See my comment here

<!-- gh-comment-id:3534620184 --> @RockingRolli commented on GitHub (Nov 14, 2025): It works with the Radeon Pro R9700... if you use the right kernel. See my comment [here](https://github.com/ollama/ollama/issues/12893#issuecomment-3534572020)
GiteaMirror commented 2026-04-29 08:27:04 -05:00
Author
Owner
Copy Link

@ChenxiWu-Lab commented on GitHub (Nov 15, 2025):

Thanks, however, I'm using Ubuntu, which seems to be different from your system. I've noticed that the official AMD driver supports kernels 6.8 or 6.14. I'm trying to change the kernel to 6.14 to see if that solves the problem. If not, I'll update to kernel 6.17.8 as you mentioned and then use the open-source driver.

It works with the Radeon Pro R9700... if you use the right kernel. See my comment here

<!-- gh-comment-id:3536222278 --> @ChenxiWu-Lab commented on GitHub (Nov 15, 2025): Thanks, however, I'm using Ubuntu, which seems to be different from your system. I've noticed that the official AMD driver supports kernels 6.8 or 6.14. I'm trying to change the kernel to 6.14 to see if that solves the problem. If not, I'll update to kernel 6.17.8 as you mentioned and then use the open-source driver. > It works with the Radeon Pro R9700... if you use the right kernel. See my comment [here](https://github.com/ollama/ollama/issues/12893#issuecomment-3534572020)
GiteaMirror commented 2026-04-29 08:27:05 -05:00
Author
Owner
Copy Link

@ChenxiWu-Lab commented on GitHub (Nov 20, 2025):

I finally got Ollama running successfully using the 6.17.8 kernel without any errors. However, this brought a new problem: installing ROCM on the 6.17.8 kernel might cause compatibility issues, making other services like VLLM unusable. Switching back to the 6.8 kernel caused Ollama to malfunction again (AMD memory scheduling). Version 0.13.0 has been released; I don't know if it will resolve the issue.

<!-- gh-comment-id:3555430950 --> @ChenxiWu-Lab commented on GitHub (Nov 20, 2025): I finally got Ollama running successfully using the 6.17.8 kernel without any errors. However, this brought a new problem: installing ROCM on the 6.17.8 kernel might cause compatibility issues, making other services like VLLM unusable. Switching back to the 6.8 kernel caused Ollama to malfunction again (AMD memory scheduling). Version 0.13.0 has been released; I don't know if it will resolve the issue.
GiteaMirror commented 2026-04-29 08:27:05 -05:00
Author
Owner
Copy Link

@JavierAbrego commented on GitHub (Dec 12, 2025):

amdgpu driver should not work with R9700 on 6.8 kernel as far as i know

<!-- gh-comment-id:3645521928 --> @JavierAbrego commented on GitHub (Dec 12, 2025): amdgpu driver should not work with R9700 on 6.8 kernel as far as i know
GiteaMirror commented 2026-04-29 08:27:05 -05:00
Author
Owner
Copy Link

@ChenxiWu-Lab commented on GitHub (Dec 17, 2025):

I later inquired about this issue on the ROCM community forum, and AMD staff responded kindly, but they weren't able to reproduce the problem.

After some adjustments, I discovered the issue might be related to the motherboard BIOS. Whenever this "failure to unload VRAM" problem occurred, rocm-smi would display the graphics card as being in a low-power state. Updating the motherboard BIOS resolved the problem. Currently, I can run Ollama without any issues on ROCM versions 6.4 and 6.x.

<!-- gh-comment-id:3663601810 --> @ChenxiWu-Lab commented on GitHub (Dec 17, 2025): I later inquired about this issue on the ROCM community forum, and AMD staff responded kindly, but they weren't able to reproduce the problem. After some adjustments, I discovered the issue might be related to the motherboard BIOS. Whenever this "failure to unload VRAM" problem occurred, rocm-smi would display the graphics card as being in a low-power state. Updating the motherboard BIOS resolved the problem. Currently, I can run Ollama without any issues on ROCM versions 6.4 and 6.x.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#55174
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?