github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #11841] the runner process fails to pick up GPUs with SLURM sbatch or srun with singularity #69917

New Issue
Closed
opened 2026-05-04 19:46:54 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-05-04 19:46:54 -05:00
Owner
Copy Link

Originally created by @hwang2006 on GitHub (Aug 10, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11841

What I observed (symptoms)
Under srun/sbatch + Singularity, Ollama would start and detect the A100 (“inference compute… library=cuda”), but when the runner spawned it would load:

load_backend: loaded CPU backend from /usr/lib/ollama/libggml-cpu-haswell.so

…and never show “loaded CUDA backend / using device CUDA0”.

As a result, generation was CPU-only → Gradio sat “waiting” forever or felt painfully slow.

When I ran the same script on a login node for testing purpose, the runner did pick CUDA (you saw “using device CUDA0 … runner started”).

$ srun -p amd_a100nv_8 --comment=pytorch --gres=gpu:1 ./ollama_gradio_run.sh
srun: job 540347 queued and waiting for resources
srun: job 540347 has been allocated resources

Starting Ollama + Gradio
Date: Sun Aug 10 18:24:57 KST 2025
Server: gpu34
SLURM Job ID: 540347
Gradio Port (requested): 7860
Ollama Port: 11434
Default Model: 0

load module-environment
🔍 Python / GPU:
/scratch/qualis/miniconda3/envs/deepseek/bin/python
Python executable: /scratch/qualis/miniconda3/envs/deepseek/bin/python
fastapi 0.116.1
gradio 5.41.1
gradio_client 1.11.0
uvicorn 0.35.0
NVIDIA A100-SXM4-80GB, 81920, 81038
Using CUDA toolkit at: /apps/cuda/12.1
🚀 Starting Ollama server…
Ollama PID: 125982
✅ Ollama API is up!
📋 Available models:

  • mistral:7b (7.2B, Q4_K_M)
  • qwen3:8b (8.2B, Q4_K_M)
  • gpt-oss:latest (20.9B, MXFP4)
  • gpt-oss:120b (116.8B, MXFP4)
  • tinyllama:latest (1B, Q4_0)
  • phi3:latest (3.8B, Q4_0)
  • gemma:latest (9B, Q4_0)
  • llama3:latest (8.0B, Q4_0)
    🌐 Starting Gradio web interface...
    Gradio PID: 126331
    ⏳ Waiting for Gradio UI at http://127.0.0.1:7860/ ...
    ... still waiting (10s)
    ... still waiting (20s)
    ... still waiting (30s)
    ... still waiting (40s)
    ... still waiting (50s)
    ... still waiting (60s)
    ... still waiting (70s)
    ... still waiting (80s)
    ... still waiting (90s)
    ... still waiting (100s)
    ... still waiting (110s)
    ... still waiting (120s)
    ... still waiting (130s)
    ... still waiting (140s)
    ... still waiting (150s)
    ... still waiting (160s)
    ... still waiting (170s)
    ... still waiting (180s)
    ... still waiting (190s)
    ... still waiting (200s)
    ... still waiting (210s)
    ... still waiting (220s)
    ... still waiting (230s)
    ... still waiting (240s)
    ... still waiting (250s)
    ... still waiting (260s)
    ... still waiting (270s)
    ... still waiting (280s)
    ... still waiting (290s)
    ... still waiting (300s)
    ... still waiting (310s)
    ... still waiting (320s)
    ... still waiting (330s)
    ... still waiting (340s)
    ... still waiting (350s)
    ... still waiting (360s)
    ... still waiting (370s)
    ... still waiting (380s)
    ... still waiting (390s)
    ... still waiting (400s)
    ... still waiting (410s)
    ... still waiting (420s)
    ... still waiting (430s)
    ... still waiting (440s)
    ... still waiting (450s)
    ... still waiting (460s)
    ... still waiting (470s)
    ... still waiting (480s)
    ... still waiting (490s)
    ... still waiting (500s)
    ... still waiting (510s)
    ... still waiting (520s)
    ... still waiting (530s)
    ... still waiting (540s)
    ... still waiting (550s)
    ✅ Gradio UI is up!
    =========================================
    🎉 All services started successfully!
    Gradio URL: http://gpu34:7860
    Local access (tunnel): http://localhost:7860 → use:
    ssh -N -L 7860:gpu34:7860 -L 11434:gpu34:11434 qualis@neuron.ksc.re.kr
    Ollama API: http://gpu34:11434
    Logs:
    Ollama: /scratch/qualis/deepseek/ollama_server_540347.log
    Gradio: /scratch/qualis/deepseek/gradio_server_540347.log
    =========================================

ollama_server_540347.log

ollama_gradio_run.sh.txt

Originally created by @hwang2006 on GitHub (Aug 10, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11841 What I observed (symptoms) Under srun/sbatch + Singularity, Ollama would start and detect the A100 (“inference compute… library=cuda”), but when the runner spawned it would load: load_backend: loaded CPU backend from /usr/lib/ollama/libggml-cpu-haswell.so …and never show “loaded CUDA backend / using device CUDA0”. As a result, generation was CPU-only → Gradio sat “waiting” forever or felt painfully slow. When I ran the same script on a login node for testing purpose, the runner did pick CUDA (you saw “using device CUDA0 … runner started”). $ srun -p amd_a100nv_8 --comment=pytorch --gres=gpu:1 ./ollama_gradio_run.sh srun: job 540347 queued and waiting for resources srun: job 540347 has been allocated resources ======================================== Starting Ollama + Gradio Date: Sun Aug 10 18:24:57 KST 2025 Server: gpu34 SLURM Job ID: 540347 Gradio Port (requested): 7860 Ollama Port: 11434 Default Model: 0 ======================================== load module-environment 🔍 Python / GPU: /scratch/qualis/miniconda3/envs/deepseek/bin/python Python executable: /scratch/qualis/miniconda3/envs/deepseek/bin/python fastapi 0.116.1 gradio 5.41.1 gradio_client 1.11.0 uvicorn 0.35.0 NVIDIA A100-SXM4-80GB, 81920, 81038 Using CUDA toolkit at: /apps/cuda/12.1 🚀 Starting Ollama server… Ollama PID: 125982 ✅ Ollama API is up! 📋 Available models: - mistral:7b (7.2B, Q4_K_M) - qwen3:8b (8.2B, Q4_K_M) - gpt-oss:latest (20.9B, MXFP4) - gpt-oss:120b (116.8B, MXFP4) - tinyllama:latest (1B, Q4_0) - phi3:latest (3.8B, Q4_0) - gemma:latest (9B, Q4_0) - llama3:latest (8.0B, Q4_0) 🌐 Starting Gradio web interface... Gradio PID: 126331 ⏳ Waiting for Gradio UI at http://127.0.0.1:7860/ ... ... still waiting (10s) ... still waiting (20s) ... still waiting (30s) ... still waiting (40s) ... still waiting (50s) ... still waiting (60s) ... still waiting (70s) ... still waiting (80s) ... still waiting (90s) ... still waiting (100s) ... still waiting (110s) ... still waiting (120s) ... still waiting (130s) ... still waiting (140s) ... still waiting (150s) ... still waiting (160s) ... still waiting (170s) ... still waiting (180s) ... still waiting (190s) ... still waiting (200s) ... still waiting (210s) ... still waiting (220s) ... still waiting (230s) ... still waiting (240s) ... still waiting (250s) ... still waiting (260s) ... still waiting (270s) ... still waiting (280s) ... still waiting (290s) ... still waiting (300s) ... still waiting (310s) ... still waiting (320s) ... still waiting (330s) ... still waiting (340s) ... still waiting (350s) ... still waiting (360s) ... still waiting (370s) ... still waiting (380s) ... still waiting (390s) ... still waiting (400s) ... still waiting (410s) ... still waiting (420s) ... still waiting (430s) ... still waiting (440s) ... still waiting (450s) ... still waiting (460s) ... still waiting (470s) ... still waiting (480s) ... still waiting (490s) ... still waiting (500s) ... still waiting (510s) ... still waiting (520s) ... still waiting (530s) ... still waiting (540s) ... still waiting (550s) ✅ Gradio UI is up! ========================================= 🎉 All services started successfully! Gradio URL: http://gpu34:7860 Local access (tunnel): http://localhost:7860 → use: ssh -N -L 7860:gpu34:7860 -L 11434:gpu34:11434 qualis@neuron.ksc.re.kr Ollama API: http://gpu34:11434 Logs: Ollama: /scratch/qualis/deepseek/ollama_server_540347.log Gradio: /scratch/qualis/deepseek/gradio_server_540347.log ========================================= [ollama_server_540347.log](https://github.com/user-attachments/files/21704738/ollama_server_540347.log) [ollama_gradio_run.sh.txt](https://github.com/user-attachments/files/21704730/ollama_gradio_run.sh.txt)
GiteaMirror added the feature request label 2026-05-04 19:46:54 -05:00
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 feature request
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#69917
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?