github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #1566] Error: llama runner exited, you may not have enough available memory to run this model #47369

New Issue
Closed
opened 2026-04-28 03:38:14 -05:00 by GiteaMirror · 7 comments
GiteaMirror commented 2026-04-28 03:38:14 -05:00
Owner
Copy Link

Originally created by @baardove on GitHub (Dec 16, 2023).
Original GitHub issue: https://github.com/ollama/ollama/issues/1566

Hi,

When I have run a modell and try to communicate with it, I always get same response, no matter which model (or small or big)...
'
Error: llama runner exited, you may not have enough available memory to run this model
'

Any clues on this one?

My host is running ubuntu 20.04 on proxmox with approx 56 gb memory free, nvidia m40 24 gb gpu
'
free
total used free shared buff/cache available
Mem: 58212660 641572 54462900 5692 3108188 56950236
Swap: 8388604 0 8388604
'

'
nvidia-smi
Sat Dec 16 19:39:44 2023
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.129.03 Driver Version: 535.129.03 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 Tesla M40 24GB Off | 00000000:01:00.0 Off | 0 |
| N/A 37C P8 16W / 250W | 0MiB / 23040MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| No running processes found |
+---------------------------------------------------------------------------------------+
'

Seems ollama finds the gpu:
journalctl:
`
Dec 16 18:30:05 tesla ollama[2245]: 2023/12/16 18:30:05 llama.go:300: 22939 MB VRAM available, loading up to 150 GPU layers
Dec 16 18:30:05 tesla ollama[2245]: 2023/12/16 18:30:05 llama.go:436: starting llama runner
Dec 16 18:30:05 tesla ollama[2245]: 2023/12/16 18:30:05 llama.go:494: waiting for llama runner to start responding
Dec 16 18:30:05 tesla ollama[2245]: ggml_init_cublas: GGML_CUDA_FORCE_MMQ: no
Dec 16 18:30:05 tesla ollama[2245]: ggml_init_cublas: CUDA_USE_TENSOR_CORES: yes
Dec 16 18:30:05 tesla ollama[2245]: ggml_init_cublas: found 1 CUDA devices:
Dec 16 18:30:05 tesla ollama[2245]: Device 0: Tesla M40 24GB, compute capability 5.2
Dec 16 18:30:05 tesla ollama[2326]: {"timestamp":1702751405,"level":"INFO","function":"main","line":2652,"message":"build info","build":441,"commit":"948ff1>
Dec 16 18:30:05 tesla ollama[2326]: {"timestamp":1702751405,"level":"INFO","function":"main","line":2655,"message":"system info","n_threads":8,"n_threads_ba>
Dec 16 18:30:05 tesla ollama[2245]: llama_model_loader: loaded meta data with 19 key-value pairs and 291 tensors from /usr/share/ollama/.ollama/models/blobs>

Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: ggml ctx size = 0.12 MiB
Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: using CUDA for GPU acceleration
Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: mem required = 70.43 MiB
Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: offloading 32 repeating layers to GPU
Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: offloading non-repeating layers to GPU
Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: offloaded 33/33 layers to GPU
Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: VRAM used: 3577.56 MiB
`

Loading a modell works fine, but error comes when trying to communicate, happens with any modell, even the smallest.
Error: llama runner exited, you may not have enough available memory to run this model

journalctl:
'
Dec 16 18:31:50 tesla ollama[2245]: ..................................................................................................
Dec 16 18:31:50 tesla ollama[2245]: llama_new_context_with_model: n_ctx = 4096
Dec 16 18:31:50 tesla ollama[2245]: llama_new_context_with_model: freq_base = 10000.0
Dec 16 18:31:50 tesla ollama[2245]: llama_new_context_with_model: freq_scale = 1
Dec 16 18:31:51 tesla ollama[2245]: llama_kv_cache_init: VRAM kv self = 2048.00 MB
Dec 16 18:31:51 tesla ollama[2245]: llama_new_context_with_model: KV self size = 2048.00 MiB, K (f16): 1024.00 MiB, V (f16): 1024.00 MiB
Dec 16 18:31:51 tesla ollama[2245]: llama_build_graph: non-view tensors processed: 676/676
Dec 16 18:31:51 tesla ollama[2245]: llama_new_context_with_model: compute buffer total size = 291.32 MiB
Dec 16 18:31:51 tesla ollama[2245]: llama_new_context_with_model: VRAM scratch buffer: 288.00 MiB
Dec 16 18:31:51 tesla ollama[2245]: llama_new_context_with_model: total VRAM used: 5913.57 MiB (model: 3577.56 MiB, context: 2336.00 MiB)
Dec 16 18:31:51 tesla ollama[2588]: {"timestamp":1702751511,"level":"INFO","function":"main","line":3035,"message":"HTTP server listening","hostname":"127.0>
Dec 16 18:31:51 tesla ollama[2588]: {"timestamp":1702751511,"level":"INFO","function":"log_server_request","line":2596,"message":"request","remote_addr":"12>
Dec 16 18:31:51 tesla ollama[2245]: 2023/12/16 18:31:51 llama.go:508: llama runner started in 2.201689 seconds
Dec 16 18:31:51 tesla ollama[2245]: [GIN] 2023/12/16 - 18:31:51 | 200 | 2.311479662s | 127.0.0.1 | POST "/api/generate"
Dec 16 18:32:14 tesla ollama[2588]: {"timestamp":1702751534,"level":"INFO","function":"log_server_request","line":2596,"message":"request","remote_addr":"12>
Dec 16 18:32:14 tesla ollama[2245]: 2023/12/16 18:32:14 llama.go:577: loaded 0 images
Dec 16 18:32:14 tesla ollama[2245]: cuBLAS error 15 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8448
Dec 16 18:32:14 tesla ollama[2245]: current device: 0
Dec 16 18:32:14 tesla ollama[2245]: GGML_ASSERT: /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8448: !"cuBLAS error"
Dec 16 18:32:14 tesla ollama[2245]: 2023/12/16 18:32:14 llama.go:451: signal: aborted (core dumped)
Dec 16 18:32:14 tesla ollama[2245]: 2023/12/16 18:32:14 llama.go:525: llama runner stopped successfully
Dec 16 18:32:14 tesla ollama[2245]: [GIN] 2023/12/16 - 18:32:14 | 200 | 601.813679ms | 127.0.0.1 | POST "/api/generate"
'

Full log:
https://www.evernote.com/shard/s16/sh/6d2eab19-c11f-7cf4-148c-9a5cd04dc944/Zwy3R7zsW8TvzDquK5Devnpko4BPwqNquvDt4nHLGCiecB_luwmk3sH8ug

The gpu is a bit dated, so it might miss some features newer nvidia cards have. It is a affordable option to run with a lot of vram so would be nice if it was supported.

When running ComfyUI i have to start with --disable-cuda-malloc.

Regards,

Bård Ove Myhr

Originally created by @baardove on GitHub (Dec 16, 2023). Original GitHub issue: https://github.com/ollama/ollama/issues/1566 Hi, When I have run a modell and try to communicate with it, I always get same response, no matter which model (or small or big)... ' Error: llama runner exited, you may not have enough available memory to run this model ' Any clues on this one? My host is running ubuntu 20.04 on proxmox with approx 56 gb memory free, nvidia m40 24 gb gpu ' free total used free shared buff/cache available Mem: 58212660 641572 54462900 5692 3108188 56950236 Swap: 8388604 0 8388604 ' ' nvidia-smi Sat Dec 16 19:39:44 2023 +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.129.03 Driver Version: 535.129.03 CUDA Version: 12.2 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 Tesla M40 24GB Off | 00000000:01:00.0 Off | 0 | | N/A 37C P8 16W / 250W | 0MiB / 23040MiB | 0% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | No running processes found | +---------------------------------------------------------------------------------------+ ' Seems ollama finds the gpu: journalctl: ` Dec 16 18:30:05 tesla ollama[2245]: 2023/12/16 18:30:05 llama.go:300: 22939 MB VRAM available, loading up to 150 GPU layers Dec 16 18:30:05 tesla ollama[2245]: 2023/12/16 18:30:05 llama.go:436: starting llama runner Dec 16 18:30:05 tesla ollama[2245]: 2023/12/16 18:30:05 llama.go:494: waiting for llama runner to start responding Dec 16 18:30:05 tesla ollama[2245]: ggml_init_cublas: GGML_CUDA_FORCE_MMQ: no Dec 16 18:30:05 tesla ollama[2245]: ggml_init_cublas: CUDA_USE_TENSOR_CORES: yes Dec 16 18:30:05 tesla ollama[2245]: ggml_init_cublas: found 1 CUDA devices: Dec 16 18:30:05 tesla ollama[2245]: Device 0: Tesla M40 24GB, compute capability 5.2 Dec 16 18:30:05 tesla ollama[2326]: {"timestamp":1702751405,"level":"INFO","function":"main","line":2652,"message":"build info","build":441,"commit":"948ff1> Dec 16 18:30:05 tesla ollama[2326]: {"timestamp":1702751405,"level":"INFO","function":"main","line":2655,"message":"system info","n_threads":8,"n_threads_ba> Dec 16 18:30:05 tesla ollama[2245]: llama_model_loader: loaded meta data with 19 key-value pairs and 291 tensors from /usr/share/ollama/.ollama/models/blobs> --- --- Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: ggml ctx size = 0.12 MiB Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: using CUDA for GPU acceleration Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: mem required = 70.43 MiB Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: offloading 32 repeating layers to GPU Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: offloading non-repeating layers to GPU Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: offloaded 33/33 layers to GPU Dec 16 18:31:49 tesla ollama[2245]: llm_load_tensors: VRAM used: 3577.56 MiB ` Loading a modell works fine, but error comes when trying to communicate, happens with any modell, even the smallest. Error: llama runner exited, you may not have enough available memory to run this model journalctl: ' Dec 16 18:31:50 tesla ollama[2245]: .................................................................................................. Dec 16 18:31:50 tesla ollama[2245]: llama_new_context_with_model: n_ctx = 4096 Dec 16 18:31:50 tesla ollama[2245]: llama_new_context_with_model: freq_base = 10000.0 Dec 16 18:31:50 tesla ollama[2245]: llama_new_context_with_model: freq_scale = 1 Dec 16 18:31:51 tesla ollama[2245]: llama_kv_cache_init: VRAM kv self = 2048.00 MB Dec 16 18:31:51 tesla ollama[2245]: llama_new_context_with_model: KV self size = 2048.00 MiB, K (f16): 1024.00 MiB, V (f16): 1024.00 MiB Dec 16 18:31:51 tesla ollama[2245]: llama_build_graph: non-view tensors processed: 676/676 Dec 16 18:31:51 tesla ollama[2245]: llama_new_context_with_model: compute buffer total size = 291.32 MiB Dec 16 18:31:51 tesla ollama[2245]: llama_new_context_with_model: VRAM scratch buffer: 288.00 MiB Dec 16 18:31:51 tesla ollama[2245]: llama_new_context_with_model: total VRAM used: 5913.57 MiB (model: 3577.56 MiB, context: 2336.00 MiB) Dec 16 18:31:51 tesla ollama[2588]: {"timestamp":1702751511,"level":"INFO","function":"main","line":3035,"message":"HTTP server listening","hostname":"127.0> Dec 16 18:31:51 tesla ollama[2588]: {"timestamp":1702751511,"level":"INFO","function":"log_server_request","line":2596,"message":"request","remote_addr":"12> Dec 16 18:31:51 tesla ollama[2245]: 2023/12/16 18:31:51 llama.go:508: llama runner started in 2.201689 seconds Dec 16 18:31:51 tesla ollama[2245]: [GIN] 2023/12/16 - 18:31:51 | 200 | 2.311479662s | 127.0.0.1 | POST "/api/generate" Dec 16 18:32:14 tesla ollama[2588]: {"timestamp":1702751534,"level":"INFO","function":"log_server_request","line":2596,"message":"request","remote_addr":"12> Dec 16 18:32:14 tesla ollama[2245]: 2023/12/16 18:32:14 llama.go:577: loaded 0 images **Dec 16 18:32:14 tesla ollama[2245]: cuBLAS error 15 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8448** Dec 16 18:32:14 tesla ollama[2245]: current device: 0 **Dec 16 18:32:14 tesla ollama[2245]: GGML_ASSERT: /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8448: !"cuBLAS error"** Dec 16 18:32:14 tesla ollama[2245]: 2023/12/16 18:32:14 llama.go:451: signal: aborted (core dumped) Dec 16 18:32:14 tesla ollama[2245]: 2023/12/16 18:32:14 llama.go:525: llama runner stopped successfully Dec 16 18:32:14 tesla ollama[2245]: [GIN] 2023/12/16 - 18:32:14 | 200 | 601.813679ms | 127.0.0.1 | POST "/api/generate" ' Full log: https://www.evernote.com/shard/s16/sh/6d2eab19-c11f-7cf4-148c-9a5cd04dc944/Zwy3R7zsW8TvzDquK5Devnpko4BPwqNquvDt4nHLGCiecB_luwmk3sH8ug The gpu is a bit dated, so it might miss some features newer nvidia cards have. It is a affordable option to run with a lot of vram so would be nice if it was supported. When running ComfyUI i have to start with --disable-cuda-malloc. Regards, Bård Ove Myhr
GiteaMirror added the bug label 2026-04-28 03:38:14 -05:00
GiteaMirror commented 2026-04-28 03:38:16 -05:00
Author
Owner
Copy Link

@baardove commented on GitHub (Dec 16, 2023):

I kind of got it working by setting num_gpu to 40 as mentioned in another post. But it still creates error, and i suspect revert to using cpu. (it pulled a lot of memory)

No error in chat window though...

Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: n_expert = 8
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: n_expert_used = 2
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: rope scaling = linear
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: freq_base_train = 1000000.0
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: freq_scale_train = 1
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: n_yarn_orig_ctx = 32768
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: rope_finetuned = unknown
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: model type = 7B
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: model ftype = mostly Q4_0
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: model params = 46.70 B
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: model size = 24.62 GiB (4.53 BPW)
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: general.name = mistralai
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: BOS token = 1 ''
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: EOS token = 2 ''
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: UNK token = 0 ''
Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: LF token = 13 '<0x0A>'
Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: ggml ctx size = 0.39 MiB
Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: using CUDA for GPU acceleration
Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: mem required = 70.71 MiB
Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: offloading 32 repeating layers to GPU
Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: offloading non-repeating layers to GPU
Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: offloaded 33/33 layers to GPU
Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: VRAM used: 25145.55 MiB
Dec 16 21:04:47 tesla ollama[8296]: ..........................................................................................
Dec 16 21:04:47 tesla ollama[8296]: CUDA error 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8955: out of memory
Dec 16 21:04:47 tesla ollama[8296]: current device: 0
Dec 16 21:04:47 tesla ollama[8296]: GGML_ASSERT: /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8955: !"CUDA error"
Dec 16 21:04:48 tesla ollama[8296]: 2023/12/16 21:04:48 llama.go:451: 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8955: out of>
Dec 16 21:04:48 tesla ollama[8296]: current device: 0
Dec 16 21:04:48 tesla ollama[8296]: GGML_ASSERT: /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8955: !"CUDA error"
Dec 16 21:04:48 tesla ollama[8296]: 2023/12/16 21:04:48 llama.go:459: error starting llama runner: llama runner process has terminated
Dec 16 21:04:48 tesla ollama[8296]: 2023/12/16 21:04:48 llama.go:525: llama runner stopped successfully
Dec 16 21:04:48 tesla ollama[8296]: 2023/12/16 21:04:48 llama.go:436: starting llama runner
Dec 16 21:04:48 tesla ollama[8296]: 2023/12/16 21:04:48 llama.go:494: waiting for llama runner to start responding
Dec 16 21:04:48 tesla ollama[17744]: {"timestamp":1702760688,"level":"WARNING","function":"server_params_parse","line":2148,"message":"Not compiled with GP>
Dec 16 21:04:48 tesla ollama[17744]: {"timestamp":1702760688,"level":"INFO","function":"main","line":2652,"message":"build info","build":441,"commit":"948f>
Dec 16 21:04:48 tesla ollama[17744]: {"timestamp":1702760688,"level":"INFO","function":"main","line":2655,"message":"system info","n_threads":8,"n_threads_>
Dec 16 21:04:48 tesla ollama[8296]: llama_model_loader: loaded meta data with 24 key-value pairs and 995 tensors from /usr/share/ollama/.ollama/models/blob>
Dec 16 21:04:48 tesla ollama[8296]: llama_model_loader: - tensor 0: token_embd.weight q4_0 [ 4096, 32000, 1, 1 ]

<!-- gh-comment-id:1858931830 --> @baardove commented on GitHub (Dec 16, 2023): I kind of got it working by setting num_gpu to 40 as mentioned in another post. But it still creates error, and i suspect revert to using cpu. (it pulled a lot of memory) No error in chat window though... Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: n_expert = 8 Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: n_expert_used = 2 Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: rope scaling = linear Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: freq_base_train = 1000000.0 Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: freq_scale_train = 1 Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: n_yarn_orig_ctx = 32768 Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: rope_finetuned = unknown Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: model type = 7B Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: model ftype = mostly Q4_0 Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: model params = 46.70 B Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: model size = 24.62 GiB (4.53 BPW) Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: general.name = mistralai Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: BOS token = 1 '<s>' Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: EOS token = 2 '</s>' Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: UNK token = 0 '<unk>' Dec 16 21:04:43 tesla ollama[8296]: llm_load_print_meta: LF token = 13 '<0x0A>' Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: ggml ctx size = 0.39 MiB Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: using CUDA for GPU acceleration Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: mem required = 70.71 MiB Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: offloading 32 repeating layers to GPU Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: offloading non-repeating layers to GPU Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: offloaded 33/33 layers to GPU Dec 16 21:04:43 tesla ollama[8296]: llm_load_tensors: VRAM used: 25145.55 MiB Dec 16 21:04:47 tesla ollama[8296]: .......................................................................................... Dec 16 21:04:47 tesla ollama[8296]: CUDA error 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8955: out of memory Dec 16 21:04:47 tesla ollama[8296]: current device: 0 Dec 16 21:04:47 tesla ollama[8296]: GGML_ASSERT: /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8955: !"CUDA error" Dec 16 21:04:48 tesla ollama[8296]: 2023/12/16 21:04:48 llama.go:451: 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8955: out of> Dec 16 21:04:48 tesla ollama[8296]: current device: 0 Dec 16 21:04:48 tesla ollama[8296]: GGML_ASSERT: /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:8955: !"CUDA error" Dec 16 21:04:48 tesla ollama[8296]: 2023/12/16 21:04:48 llama.go:459: error starting llama runner: llama runner process has terminated Dec 16 21:04:48 tesla ollama[8296]: 2023/12/16 21:04:48 llama.go:525: llama runner stopped successfully Dec 16 21:04:48 tesla ollama[8296]: 2023/12/16 21:04:48 llama.go:436: starting llama runner Dec 16 21:04:48 tesla ollama[8296]: 2023/12/16 21:04:48 llama.go:494: waiting for llama runner to start responding Dec 16 21:04:48 tesla ollama[17744]: {"timestamp":1702760688,"level":"WARNING","function":"server_params_parse","line":2148,"message":"Not compiled with GP> Dec 16 21:04:48 tesla ollama[17744]: {"timestamp":1702760688,"level":"INFO","function":"main","line":2652,"message":"build info","build":441,"commit":"948f> Dec 16 21:04:48 tesla ollama[17744]: {"timestamp":1702760688,"level":"INFO","function":"main","line":2655,"message":"system info","n_threads":8,"n_threads_> Dec 16 21:04:48 tesla ollama[8296]: llama_model_loader: loaded meta data with 24 key-value pairs and 995 tensors from /usr/share/ollama/.ollama/models/blob> Dec 16 21:04:48 tesla ollama[8296]: llama_model_loader: - tensor 0: token_embd.weight q4_0 [ 4096, 32000, 1, 1 ]
GiteaMirror commented 2026-04-28 03:38:16 -05:00
Author
Owner
Copy Link

@phalexo commented on GitHub (Dec 16, 2023):

I am running on the Maxwell architecture.

git clone --recursive https://github.com/jmorganca/ollama.git
cd ollama/llm/llama.cpp
vi generate_linux.go

//go:generate cmake -S ggml -B ggml/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_CUDA_FORCE_MMQ=on
//go:generate cmake --build ggml/build/cuda --target server --config Release
//go:generate mv ggml/build/cuda/bin/server ggml/build/cuda/bin/ollama-runner
//go:generate cmake -S gguf -B gguf/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off -DLLAMA_CUDA_PEER_MAX_BATCH_SIZE=0 -DLLAMA_CUDA_FORCE_MMQ=on
//go:generate cmake --build gguf/build/cuda --target server --config Release
//go:generate mv gguf/build/cuda/bin/server gguf/build/cuda/bin/ollama-runner

cd ../..
go generate ./...
go build .
<!-- gh-comment-id:1858983734 --> @phalexo commented on GitHub (Dec 16, 2023): I am running on the Maxwell architecture. ```bash git clone --recursive https://github.com/jmorganca/ollama.git cd ollama/llm/llama.cpp vi generate_linux.go ``` ```go //go:generate cmake -S ggml -B ggml/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_CUDA_FORCE_MMQ=on //go:generate cmake --build ggml/build/cuda --target server --config Release //go:generate mv ggml/build/cuda/bin/server ggml/build/cuda/bin/ollama-runner //go:generate cmake -S gguf -B gguf/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off -DLLAMA_CUDA_PEER_MAX_BATCH_SIZE=0 -DLLAMA_CUDA_FORCE_MMQ=on //go:generate cmake --build gguf/build/cuda --target server --config Release //go:generate mv gguf/build/cuda/bin/server gguf/build/cuda/bin/ollama-runner ``` ```bash cd ../.. go generate ./... go build . ```
GiteaMirror commented 2026-04-28 03:38:17 -05:00
Author
Owner
Copy Link

@phalexo commented on GitHub (Dec 16, 2023):

Do you have tensor cores on your GPU? I doubt it.

<!-- gh-comment-id:1858984048 --> @phalexo commented on GitHub (Dec 16, 2023): Do you have tensor cores on your GPU? I doubt it.
GiteaMirror commented 2026-04-28 03:38:17 -05:00
Author
Owner
Copy Link

@easp commented on GitHub (Dec 17, 2023):

The capabilities of the m40 are kind of limited, plus it looks like you are using older drivers.

<!-- gh-comment-id:1859227080 --> @easp commented on GitHub (Dec 17, 2023): The capabilities of the m40 are kind of limited, plus it looks like you are using older drivers.
GiteaMirror commented 2026-04-28 03:38:18 -05:00
Author
Owner
Copy Link

@sunzh231 commented on GitHub (Dec 19, 2023):

The same error occurred on M40, and it ran normally before version 0.1.11 (including 0.1.11).

<!-- gh-comment-id:1862827092 --> @sunzh231 commented on GitHub (Dec 19, 2023): The same error occurred on M40, and it ran normally before version 0.1.11 (including 0.1.11).
GiteaMirror commented 2026-04-28 03:38:19 -05:00
Author
Owner
Copy Link

@phalexo commented on GitHub (Dec 19, 2023):

You have to rebuild with LLAMA_CUDA_FORCE_MMQ=on
performance may be a bit worse, but it would work.

On Tue, Dec 19, 2023, 9:09 AM Richard Sun @.***> wrote:

The same error occurred on M40, and it ran normally before version 0.1.11
(including 0.1.11).

—
Reply to this email directly, view it on GitHub
https://github.com/jmorganca/ollama/issues/1566#issuecomment-1862827092,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/ABDD3ZMISHXYMITQTSAP5XTYKGN2FAVCNFSM6AAAAABAXZJXUWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNRSHAZDOMBZGI
.
You are receiving this because you commented.Message ID:
@.***>

<!-- gh-comment-id:1862897481 --> @phalexo commented on GitHub (Dec 19, 2023): You have to rebuild with LLAMA_CUDA_FORCE_MMQ=on performance may be a bit worse, but it would work. On Tue, Dec 19, 2023, 9:09 AM Richard Sun ***@***.***> wrote: > The same error occurred on M40, and it ran normally before version 0.1.11 > (including 0.1.11). > > — > Reply to this email directly, view it on GitHub > <https://github.com/jmorganca/ollama/issues/1566#issuecomment-1862827092>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/ABDD3ZMISHXYMITQTSAP5XTYKGN2FAVCNFSM6AAAAABAXZJXUWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNRSHAZDOMBZGI> > . > You are receiving this because you commented.Message ID: > ***@***.***> >
GiteaMirror commented 2026-04-28 03:38:19 -05:00
Author
Owner
Copy Link

@bonswouar commented on GitHub (Jan 4, 2024):

I can confirm compilling with LLAMA_CUDA_FORCE_MMQ=on solves the issue for me too: I can now run 7B models with 10 to 20 layers on the gpu (which is an old GTX 970)

<!-- gh-comment-id:1877493475 --> @bonswouar commented on GitHub (Jan 4, 2024): I can confirm compilling with `LLAMA_CUDA_FORCE_MMQ=on` solves the issue for me too: I can now run 7B models with 10 to 20 layers on the gpu (which is an old GTX 970)
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#47369
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?