github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #7101] "CUDA error: an illegal memory access was encountered" during image processing via minicpm-v:latest model #4509

New Issue
Closed
opened 2026-04-12 15:26:21 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-12 15:26:21 -05:00
Owner
Copy Link

Originally created by @aqualx on GitHub (Oct 4, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/7101

What is the issue?

The crash happens while processing the png image via minicpm-v:latest (1862d7d5fee50b69f6e3007ec999145ab38f17688251495f87669eb81e9dd97c) model. It occurs only on specific png image. Other images are processed without any issues.
The same image was processed without issue via llava:latest (8dd30f6b0cb19f555f2c7a7ebda861449ea2cc76bf1f44e262931f45fc81d081)

Example of request:

POST: http://OLLAMA_IP:11434/api/generate

Body:

{
  "model": "minicpm-v:latest",
  "prompt": "Extract all visible text to markdown blocks.",
  "images": [
    "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"
  ],
  "stream": false,
  "keep_alive": 0
}

Ollama log:

ollama                       | [GIN] 2024/10/04 - 08:41:03 | 200 |  31.22349636s |  192.168.100.20 | POST     "/api/generate"
ollama                       | time=2024-10-04T08:41:23.009Z level=WARN source=sched.go:137 msg="multimodal models don't support parallel requests yet"
ollama                       | time=2024-10-04T08:41:23.168Z level=INFO source=sched.go:507 msg="updated VRAM based on existing loaded models" gpu=GPU-5d1301f5-9e77-83c1-9e2f-eff9c34008d5 library=cuda total="7.4 GiB" available="1.2 GiB"
ollama                       | time=2024-10-04T08:41:23.808Z level=INFO source=sched.go:714 msg="new model will fit in available VRAM in single GPU, loading" model=/root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0 gpu=GPU-5d1301f5-9e77-83c1-9e2f-eff9c34008d5 parallel=1 available=7857242112 required="5.8 GiB"
ollama                       | time=2024-10-04T08:41:23.808Z level=INFO source=server.go:103 msg="system memory" total="15.6 GiB" free="3.1 GiB" free_swap="0 B"
ollama                       | time=2024-10-04T08:41:23.810Z level=INFO source=memory.go:326 msg="offload to cuda" layers.requested=-1 layers.model=29 layers.offload=29 layers.split="" memory.available="[7.3 GiB]" memory.gpu_overhead="0 B" memory.required.full="5.8 GiB" memory.required.partial="5.8 GiB" memory.required.kv="112.0 MiB" memory.required.allocations="[5.8 GiB]" memory.weights.total="3.5 GiB" memory.weights.repeating="3.1 GiB" memory.weights.nonrepeating="425.3 MiB" memory.graph.full="303.2 MiB" memory.graph.partial="728.5 MiB"
ollama                       | time=2024-10-04T08:41:23.812Z level=INFO source=server.go:388 msg="starting llama server" cmd="/usr/lib/ollama/runners/cuda_v12/ollama_llama_server --model /root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0 --ctx-size 2048 --batch-size 512 --embedding --log-disable --n-gpu-layers 29 --mmproj /root/.ollama/models/blobs/sha256-f8a805e9e62085805c69c427287acefc284932eb4abfe6e1b1ce431d27e2f4e0 --no-mmap --parallel 1 --port 33077"
ollama                       | time=2024-10-04T08:41:23.813Z level=INFO source=sched.go:449 msg="loaded runners" count=1
ollama                       | time=2024-10-04T08:41:23.813Z level=INFO source=server.go:587 msg="waiting for llama runner to start responding"
ollama                       | time=2024-10-04T08:41:23.813Z level=INFO source=server.go:621 msg="waiting for server to become available" status="llm server error"
ollama                       | INFO [main] build info | build=10 commit="3f6ec33" tid="140630251470848" timestamp=1728031283
ollama                       | INFO [main] system info | n_threads=4 n_threads_batch=4 system_info="AVX = 1 | AVX_VNNI = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | FMA = 0 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 0 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 | " tid="140630251470848" timestamp=1728031283 total_threads=8
ollama                       | INFO [main] HTTP server listening | hostname="127.0.0.1" n_threads_http="7" port="33077" tid="140630251470848" timestamp=1728031283
ollama                       | ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ollama                       | ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ollama                       | ggml_cuda_init: found 1 CUDA devices:
ollama                       |   Device 0: Tesla P4, compute capability 6.1, VMM: yes
ollama                       | time=2024-10-04T08:41:24.065Z level=INFO source=server.go:621 msg="waiting for server to become available" status="llm server loading model"
ollama                       | llama_model_loader: loaded meta data with 22 key-value pairs and 339 tensors from /root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0 (version GGUF V3 (latest))
ollama                       | llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
ollama                       | llama_model_loader: - kv   0:                       general.architecture str              = qwen2
ollama                       | llama_model_loader: - kv   1:                               general.name str              = model
ollama                       | llama_model_loader: - kv   2:                          qwen2.block_count u32              = 28
ollama                       | llama_model_loader: - kv   3:                       qwen2.context_length u32              = 32768
ollama                       | llama_model_loader: - kv   4:                     qwen2.embedding_length u32              = 3584
ollama                       | llama_model_loader: - kv   5:                  qwen2.feed_forward_length u32              = 18944
ollama                       | llama_model_loader: - kv   6:                 qwen2.attention.head_count u32              = 28
ollama                       | llama_model_loader: - kv   7:              qwen2.attention.head_count_kv u32              = 4
ollama                       | llama_model_loader: - kv   8:                       qwen2.rope.freq_base f32              = 1000000.000000
ollama                       | llama_model_loader: - kv   9:     qwen2.attention.layer_norm_rms_epsilon f32              = 0.000001
ollama                       | llama_model_loader: - kv  10:                          general.file_type u32              = 2
ollama                       | llama_model_loader: - kv  11:                       tokenizer.ggml.model str              = gpt2
ollama                       | llama_model_loader: - kv  12:                         tokenizer.ggml.pre str              = qwen2
ollama                       | llama_model_loader: - kv  13:                      tokenizer.ggml.tokens arr[str,151666]  = ["!", "\"", "#", "$", "%", "&", "'", ...
ollama                       | llama_model_loader: - kv  14:                  tokenizer.ggml.token_type arr[i32,151666]  = [3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
ollama                       | llama_model_loader: - kv  15:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
ollama                       | llama_model_loader: - kv  16:                tokenizer.ggml.bos_token_id u32              = 151644
ollama                       | llama_model_loader: - kv  17:                tokenizer.ggml.eos_token_id u32              = 151645
ollama                       | llama_model_loader: - kv  18:            tokenizer.ggml.unknown_token_id u32              = 128244
ollama                       | llama_model_loader: - kv  19:            tokenizer.ggml.padding_token_id u32              = 0
ollama                       | llama_model_loader: - kv  20:                    tokenizer.chat_template str              = {% for message in messages %}{% if lo...
ollama                       | llama_model_loader: - kv  21:               general.quantization_version u32              = 2
ollama                       | llama_model_loader: - type  f32:  141 tensors
ollama                       | llama_model_loader: - type q4_0:  197 tensors
ollama                       | llama_model_loader: - type q6_K:    1 tensors
ollama                       | llm_load_vocab: special tokens cache size = 25
ollama                       | llm_load_vocab: token to piece cache size = 0.9309 MB
ollama                       | llm_load_print_meta: format           = GGUF V3 (latest)
ollama                       | llm_load_print_meta: arch             = qwen2
ollama                       | llm_load_print_meta: vocab type       = BPE
ollama                       | llm_load_print_meta: n_vocab          = 151666
ollama                       | llm_load_print_meta: n_merges         = 151387
ollama                       | llm_load_print_meta: vocab_only       = 0
ollama                       | llm_load_print_meta: n_ctx_train      = 32768
ollama                       | llm_load_print_meta: n_embd           = 3584
ollama                       | llm_load_print_meta: n_layer          = 28
ollama                       | llm_load_print_meta: n_head           = 28
ollama                       | llm_load_print_meta: n_head_kv        = 4
ollama                       | llm_load_print_meta: n_rot            = 128
ollama                       | llm_load_print_meta: n_swa            = 0
ollama                       | llm_load_print_meta: n_embd_head_k    = 128
ollama                       | llm_load_print_meta: n_embd_head_v    = 128
ollama                       | llm_load_print_meta: n_gqa            = 7
ollama                       | llm_load_print_meta: n_embd_k_gqa     = 512
ollama                       | llm_load_print_meta: n_embd_v_gqa     = 512
ollama                       | llm_load_print_meta: f_norm_eps       = 0.0e+00
ollama                       | llm_load_print_meta: f_norm_rms_eps   = 1.0e-06
ollama                       | llm_load_print_meta: f_clamp_kqv      = 0.0e+00
ollama                       | llm_load_print_meta: f_max_alibi_bias = 0.0e+00
ollama                       | llm_load_print_meta: f_logit_scale    = 0.0e+00
ollama                       | llm_load_print_meta: n_ff             = 18944
ollama                       | llm_load_print_meta: n_expert         = 0
ollama                       | llm_load_print_meta: n_expert_used    = 0
ollama                       | llm_load_print_meta: causal attn      = 1
ollama                       | llm_load_print_meta: pooling type     = 0
ollama                       | llm_load_print_meta: rope type        = 2
ollama                       | llm_load_print_meta: rope scaling     = linear
ollama                       | llm_load_print_meta: freq_base_train  = 1000000.0
ollama                       | llm_load_print_meta: freq_scale_train = 1
ollama                       | llm_load_print_meta: n_ctx_orig_yarn  = 32768
ollama                       | llm_load_print_meta: rope_finetuned   = unknown
ollama                       | llm_load_print_meta: ssm_d_conv       = 0
ollama                       | llm_load_print_meta: ssm_d_inner      = 0
ollama                       | llm_load_print_meta: ssm_d_state      = 0
ollama                       | llm_load_print_meta: ssm_dt_rank      = 0
ollama                       | llm_load_print_meta: ssm_dt_b_c_rms   = 0
ollama                       | llm_load_print_meta: model type       = ?B
ollama                       | llm_load_print_meta: model ftype      = Q4_0
ollama                       | llm_load_print_meta: model params     = 7.61 B
ollama                       | llm_load_print_meta: model size       = 4.12 GiB (4.65 BPW) 
ollama                       | llm_load_print_meta: general.name     = model
ollama                       | llm_load_print_meta: BOS token        = 151644 '<|im_start|>'
ollama                       | llm_load_print_meta: EOS token        = 151645 '<|im_end|>'
ollama                       | llm_load_print_meta: UNK token        = 128244 '<unk>'
ollama                       | llm_load_print_meta: PAD token        = 0 '!'
ollama                       | llm_load_print_meta: LF token         = 148848 'ÄĬ'
ollama                       | llm_load_print_meta: EOT token        = 151645 '<|im_end|>'
ollama                       | llm_load_print_meta: max token length = 256
ollama                       | llm_load_tensors: ggml ctx size =    0.30 MiB
ollama                       | llm_load_tensors: offloading 28 repeating layers to GPU
ollama                       | llm_load_tensors: offloading non-repeating layers to GPU
ollama                       | llm_load_tensors: offloaded 29/29 layers to GPU
ollama                       | llm_load_tensors:  CUDA_Host buffer size =   291.59 MiB
ollama                       | llm_load_tensors:      CUDA0 buffer size =  3926.95 MiB
ollama                       | llama_new_context_with_model: n_ctx      = 2048
ollama                       | llama_new_context_with_model: n_batch    = 512
ollama                       | llama_new_context_with_model: n_ubatch   = 512
ollama                       | llama_new_context_with_model: flash_attn = 0
ollama                       | llama_new_context_with_model: freq_base  = 1000000.0
ollama                       | llama_new_context_with_model: freq_scale = 1
ollama                       | llama_kv_cache_init:      CUDA0 KV buffer size =   112.00 MiB
ollama                       | llama_new_context_with_model: KV self size  =  112.00 MiB, K (f16):   56.00 MiB, V (f16):   56.00 MiB
ollama                       | llama_new_context_with_model:  CUDA_Host  output buffer size =     0.59 MiB
ollama                       | llama_new_context_with_model:      CUDA0 compute buffer size =   303.22 MiB
ollama                       | llama_new_context_with_model:  CUDA_Host compute buffer size =    11.01 MiB
ollama                       | llama_new_context_with_model: graph nodes  = 986
ollama                       | llama_new_context_with_model: graph splits = 2
ollama                       | INFO [main] model loaded | tid="140630251470848" timestamp=1728031297
ollama                       | time=2024-10-04T08:41:37.378Z level=INFO source=server.go:626 msg="llama runner started in 13.57 seconds"
ollama                       | ggml_cuda_compute_forward: SCALE failed
ollama                       | CUDA error: an illegal memory access was encountered
ollama                       |   current device: 0, in function ggml_cuda_compute_forward at /go/src/github.com/ollama/ollama/llm/llama.cpp/ggml/src/ggml-cuda.cu:2326
ollama                       |   err
ollama                       | /go/src/github.com/ollama/ollama/llm/llama.cpp/ggml/src/ggml-cuda.cu:102: CUDA error
ollama                       | [GIN] 2024/10/04 - 08:41:54 | 500 | 31.825048562s |  192.168.100.20 | POST     "/api/generate"
ollama                       | time=2024-10-04T08:41:59.776Z level=WARN source=sched.go:646 msg="gpu VRAM usage didn't recover within timeout" seconds=5.001895153 model=/root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0
ollama                       | time=2024-10-04T08:42:00.027Z level=WARN source=sched.go:646 msg="gpu VRAM usage didn't recover within timeout" seconds=5.252338468 model=/root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0
ollama                       | time=2024-10-04T08:42:00.276Z level=WARN source=sched.go:646 msg="gpu VRAM usage didn't recover within timeout" seconds=5.501732338 model=/root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0

OS

Docker

GPU

Nvidia

CPU

Intel

Ollama version

0.3.12

Originally created by @aqualx on GitHub (Oct 4, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/7101 ### What is the issue? The crash happens while processing the _png_ image via minicpm-v:latest (1862d7d5fee50b69f6e3007ec999145ab38f17688251495f87669eb81e9dd97c) model. It occurs only on specific _png_ image. Other images are processed without any issues. The same image was processed without issue via llava:latest (8dd30f6b0cb19f555f2c7a7ebda861449ea2cc76bf1f44e262931f45fc81d081) Example of request: POST: http://OLLAMA_IP:11434/api/generate Body: ``` { "model": "minicpm-v:latest", "prompt": "Extract all visible text to markdown blocks.", "images": [ "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" ], "stream": false, "keep_alive": 0 } ``` Ollama log: ``` ollama | [GIN] 2024/10/04 - 08:41:03 | 200 | 31.22349636s | 192.168.100.20 | POST "/api/generate" ollama | time=2024-10-04T08:41:23.009Z level=WARN source=sched.go:137 msg="multimodal models don't support parallel requests yet" ollama | time=2024-10-04T08:41:23.168Z level=INFO source=sched.go:507 msg="updated VRAM based on existing loaded models" gpu=GPU-5d1301f5-9e77-83c1-9e2f-eff9c34008d5 library=cuda total="7.4 GiB" available="1.2 GiB" ollama | time=2024-10-04T08:41:23.808Z level=INFO source=sched.go:714 msg="new model will fit in available VRAM in single GPU, loading" model=/root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0 gpu=GPU-5d1301f5-9e77-83c1-9e2f-eff9c34008d5 parallel=1 available=7857242112 required="5.8 GiB" ollama | time=2024-10-04T08:41:23.808Z level=INFO source=server.go:103 msg="system memory" total="15.6 GiB" free="3.1 GiB" free_swap="0 B" ollama | time=2024-10-04T08:41:23.810Z level=INFO source=memory.go:326 msg="offload to cuda" layers.requested=-1 layers.model=29 layers.offload=29 layers.split="" memory.available="[7.3 GiB]" memory.gpu_overhead="0 B" memory.required.full="5.8 GiB" memory.required.partial="5.8 GiB" memory.required.kv="112.0 MiB" memory.required.allocations="[5.8 GiB]" memory.weights.total="3.5 GiB" memory.weights.repeating="3.1 GiB" memory.weights.nonrepeating="425.3 MiB" memory.graph.full="303.2 MiB" memory.graph.partial="728.5 MiB" ollama | time=2024-10-04T08:41:23.812Z level=INFO source=server.go:388 msg="starting llama server" cmd="/usr/lib/ollama/runners/cuda_v12/ollama_llama_server --model /root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0 --ctx-size 2048 --batch-size 512 --embedding --log-disable --n-gpu-layers 29 --mmproj /root/.ollama/models/blobs/sha256-f8a805e9e62085805c69c427287acefc284932eb4abfe6e1b1ce431d27e2f4e0 --no-mmap --parallel 1 --port 33077" ollama | time=2024-10-04T08:41:23.813Z level=INFO source=sched.go:449 msg="loaded runners" count=1 ollama | time=2024-10-04T08:41:23.813Z level=INFO source=server.go:587 msg="waiting for llama runner to start responding" ollama | time=2024-10-04T08:41:23.813Z level=INFO source=server.go:621 msg="waiting for server to become available" status="llm server error" ollama | INFO [main] build info | build=10 commit="3f6ec33" tid="140630251470848" timestamp=1728031283 ollama | INFO [main] system info | n_threads=4 n_threads_batch=4 system_info="AVX = 1 | AVX_VNNI = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | FMA = 0 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 0 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 | " tid="140630251470848" timestamp=1728031283 total_threads=8 ollama | INFO [main] HTTP server listening | hostname="127.0.0.1" n_threads_http="7" port="33077" tid="140630251470848" timestamp=1728031283 ollama | ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no ollama | ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no ollama | ggml_cuda_init: found 1 CUDA devices: ollama | Device 0: Tesla P4, compute capability 6.1, VMM: yes ollama | time=2024-10-04T08:41:24.065Z level=INFO source=server.go:621 msg="waiting for server to become available" status="llm server loading model" ollama | llama_model_loader: loaded meta data with 22 key-value pairs and 339 tensors from /root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0 (version GGUF V3 (latest)) ollama | llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output. ollama | llama_model_loader: - kv 0: general.architecture str = qwen2 ollama | llama_model_loader: - kv 1: general.name str = model ollama | llama_model_loader: - kv 2: qwen2.block_count u32 = 28 ollama | llama_model_loader: - kv 3: qwen2.context_length u32 = 32768 ollama | llama_model_loader: - kv 4: qwen2.embedding_length u32 = 3584 ollama | llama_model_loader: - kv 5: qwen2.feed_forward_length u32 = 18944 ollama | llama_model_loader: - kv 6: qwen2.attention.head_count u32 = 28 ollama | llama_model_loader: - kv 7: qwen2.attention.head_count_kv u32 = 4 ollama | llama_model_loader: - kv 8: qwen2.rope.freq_base f32 = 1000000.000000 ollama | llama_model_loader: - kv 9: qwen2.attention.layer_norm_rms_epsilon f32 = 0.000001 ollama | llama_model_loader: - kv 10: general.file_type u32 = 2 ollama | llama_model_loader: - kv 11: tokenizer.ggml.model str = gpt2 ollama | llama_model_loader: - kv 12: tokenizer.ggml.pre str = qwen2 ollama | llama_model_loader: - kv 13: tokenizer.ggml.tokens arr[str,151666] = ["!", "\"", "#", "$", "%", "&", "'", ... ollama | llama_model_loader: - kv 14: tokenizer.ggml.token_type arr[i32,151666] = [3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... ollama | llama_model_loader: - kv 15: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",... ollama | llama_model_loader: - kv 16: tokenizer.ggml.bos_token_id u32 = 151644 ollama | llama_model_loader: - kv 17: tokenizer.ggml.eos_token_id u32 = 151645 ollama | llama_model_loader: - kv 18: tokenizer.ggml.unknown_token_id u32 = 128244 ollama | llama_model_loader: - kv 19: tokenizer.ggml.padding_token_id u32 = 0 ollama | llama_model_loader: - kv 20: tokenizer.chat_template str = {% for message in messages %}{% if lo... ollama | llama_model_loader: - kv 21: general.quantization_version u32 = 2 ollama | llama_model_loader: - type f32: 141 tensors ollama | llama_model_loader: - type q4_0: 197 tensors ollama | llama_model_loader: - type q6_K: 1 tensors ollama | llm_load_vocab: special tokens cache size = 25 ollama | llm_load_vocab: token to piece cache size = 0.9309 MB ollama | llm_load_print_meta: format = GGUF V3 (latest) ollama | llm_load_print_meta: arch = qwen2 ollama | llm_load_print_meta: vocab type = BPE ollama | llm_load_print_meta: n_vocab = 151666 ollama | llm_load_print_meta: n_merges = 151387 ollama | llm_load_print_meta: vocab_only = 0 ollama | llm_load_print_meta: n_ctx_train = 32768 ollama | llm_load_print_meta: n_embd = 3584 ollama | llm_load_print_meta: n_layer = 28 ollama | llm_load_print_meta: n_head = 28 ollama | llm_load_print_meta: n_head_kv = 4 ollama | llm_load_print_meta: n_rot = 128 ollama | llm_load_print_meta: n_swa = 0 ollama | llm_load_print_meta: n_embd_head_k = 128 ollama | llm_load_print_meta: n_embd_head_v = 128 ollama | llm_load_print_meta: n_gqa = 7 ollama | llm_load_print_meta: n_embd_k_gqa = 512 ollama | llm_load_print_meta: n_embd_v_gqa = 512 ollama | llm_load_print_meta: f_norm_eps = 0.0e+00 ollama | llm_load_print_meta: f_norm_rms_eps = 1.0e-06 ollama | llm_load_print_meta: f_clamp_kqv = 0.0e+00 ollama | llm_load_print_meta: f_max_alibi_bias = 0.0e+00 ollama | llm_load_print_meta: f_logit_scale = 0.0e+00 ollama | llm_load_print_meta: n_ff = 18944 ollama | llm_load_print_meta: n_expert = 0 ollama | llm_load_print_meta: n_expert_used = 0 ollama | llm_load_print_meta: causal attn = 1 ollama | llm_load_print_meta: pooling type = 0 ollama | llm_load_print_meta: rope type = 2 ollama | llm_load_print_meta: rope scaling = linear ollama | llm_load_print_meta: freq_base_train = 1000000.0 ollama | llm_load_print_meta: freq_scale_train = 1 ollama | llm_load_print_meta: n_ctx_orig_yarn = 32768 ollama | llm_load_print_meta: rope_finetuned = unknown ollama | llm_load_print_meta: ssm_d_conv = 0 ollama | llm_load_print_meta: ssm_d_inner = 0 ollama | llm_load_print_meta: ssm_d_state = 0 ollama | llm_load_print_meta: ssm_dt_rank = 0 ollama | llm_load_print_meta: ssm_dt_b_c_rms = 0 ollama | llm_load_print_meta: model type = ?B ollama | llm_load_print_meta: model ftype = Q4_0 ollama | llm_load_print_meta: model params = 7.61 B ollama | llm_load_print_meta: model size = 4.12 GiB (4.65 BPW) ollama | llm_load_print_meta: general.name = model ollama | llm_load_print_meta: BOS token = 151644 '<|im_start|>' ollama | llm_load_print_meta: EOS token = 151645 '<|im_end|>' ollama | llm_load_print_meta: UNK token = 128244 '<unk>' ollama | llm_load_print_meta: PAD token = 0 '!' ollama | llm_load_print_meta: LF token = 148848 'ÄĬ' ollama | llm_load_print_meta: EOT token = 151645 '<|im_end|>' ollama | llm_load_print_meta: max token length = 256 ollama | llm_load_tensors: ggml ctx size = 0.30 MiB ollama | llm_load_tensors: offloading 28 repeating layers to GPU ollama | llm_load_tensors: offloading non-repeating layers to GPU ollama | llm_load_tensors: offloaded 29/29 layers to GPU ollama | llm_load_tensors: CUDA_Host buffer size = 291.59 MiB ollama | llm_load_tensors: CUDA0 buffer size = 3926.95 MiB ollama | llama_new_context_with_model: n_ctx = 2048 ollama | llama_new_context_with_model: n_batch = 512 ollama | llama_new_context_with_model: n_ubatch = 512 ollama | llama_new_context_with_model: flash_attn = 0 ollama | llama_new_context_with_model: freq_base = 1000000.0 ollama | llama_new_context_with_model: freq_scale = 1 ollama | llama_kv_cache_init: CUDA0 KV buffer size = 112.00 MiB ollama | llama_new_context_with_model: KV self size = 112.00 MiB, K (f16): 56.00 MiB, V (f16): 56.00 MiB ollama | llama_new_context_with_model: CUDA_Host output buffer size = 0.59 MiB ollama | llama_new_context_with_model: CUDA0 compute buffer size = 303.22 MiB ollama | llama_new_context_with_model: CUDA_Host compute buffer size = 11.01 MiB ollama | llama_new_context_with_model: graph nodes = 986 ollama | llama_new_context_with_model: graph splits = 2 ollama | INFO [main] model loaded | tid="140630251470848" timestamp=1728031297 ollama | time=2024-10-04T08:41:37.378Z level=INFO source=server.go:626 msg="llama runner started in 13.57 seconds" ollama | ggml_cuda_compute_forward: SCALE failed ollama | CUDA error: an illegal memory access was encountered ollama | current device: 0, in function ggml_cuda_compute_forward at /go/src/github.com/ollama/ollama/llm/llama.cpp/ggml/src/ggml-cuda.cu:2326 ollama | err ollama | /go/src/github.com/ollama/ollama/llm/llama.cpp/ggml/src/ggml-cuda.cu:102: CUDA error ollama | [GIN] 2024/10/04 - 08:41:54 | 500 | 31.825048562s | 192.168.100.20 | POST "/api/generate" ollama | time=2024-10-04T08:41:59.776Z level=WARN source=sched.go:646 msg="gpu VRAM usage didn't recover within timeout" seconds=5.001895153 model=/root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0 ollama | time=2024-10-04T08:42:00.027Z level=WARN source=sched.go:646 msg="gpu VRAM usage didn't recover within timeout" seconds=5.252338468 model=/root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0 ollama | time=2024-10-04T08:42:00.276Z level=WARN source=sched.go:646 msg="gpu VRAM usage didn't recover within timeout" seconds=5.501732338 model=/root/.ollama/models/blobs/sha256-262843d4806aeb402336980badd414a72576b20b1e5d537647da15f16c4a4df0 ``` ### OS Docker ### GPU Nvidia ### CPU Intel ### Ollama version 0.3.12
GiteaMirror added the bug label 2026-04-12 15:26:21 -05:00
GiteaMirror commented 2026-04-12 15:26:24 -05:00
Author
Owner
Copy Link

@brdebr commented on GitHub (Oct 30, 2024):

I'm getting the same error recently, using the same model.

But on my case, the same png sometimes works and sometimes doesn't 🤷

image

<!-- gh-comment-id:2448030185 --> @brdebr commented on GitHub (Oct 30, 2024): I'm getting the same error recently, using the same model. But on my case, the same png sometimes works and sometimes doesn't 🤷 ![image](https://github.com/user-attachments/assets/621d6765-d076-4403-8889-ae7c3ae426f3)
GiteaMirror commented 2026-04-12 15:26:25 -05:00
Author
Owner
Copy Link

@ShivamSrng commented on GitHub (Nov 5, 2024):

I'm getting the same error recently, using the same model.

But on my case, the same png sometimes works and sometimes doesn't 🤷

image

Hey any updates on this issue ? Although I am not facing issue in your model, but in llama3.1:latest.

OS: Windows
GPU: Nvidia
CPU: Intel
Ollama version: 0.3.14
Driver Version: 552.44
CUDA Version: 12.4

<!-- gh-comment-id:2456240773 --> @ShivamSrng commented on GitHub (Nov 5, 2024): > I'm getting the same error recently, using the same model. > > But on my case, the same png sometimes works and sometimes doesn't 🤷 > > ![image](https://private-user-images.githubusercontent.com/33053368/381685140-621d6765-d076-4403-8889-ae7c3ae426f3.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.j1htQnhQtnqQITguI19cOdJ-h4lGVl8_FoyB4DNc-5U) Hey any updates on this issue ? Although I am not facing issue in your model, but in llama3.1:latest. OS: Windows GPU: Nvidia CPU: Intel Ollama version: 0.3.14 Driver Version: 552.44 CUDA Version: 12.4
GiteaMirror commented 2026-04-12 15:26:27 -05:00
Author
Owner
Copy Link

@deece commented on GitHub (Apr 24, 2025):

This might fix it: https://github.com/ollama/ollama/pull/10388

<!-- gh-comment-id:2826891989 --> @deece commented on GitHub (Apr 24, 2025): This might fix it: https://github.com/ollama/ollama/pull/10388
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#4509
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?