search

ggerganov / llama.cpp

LLM inference in C/C++
MIT License
68.69k stars 9.87k forks source link

Bug: Assertion '__n < this->size()' failed. #9636

Closed Luke100000 closed 3 weeks ago

Luke100000 commented 2 months ago

What happened?

When using an embedding model via Ollamas API, Lama.cpp has an assertion error: Bug: Assertion '__n < this->size()' failed.

I tried nomic-embed-text-v1.5 and all-minilm. It works fine if 100% CPU.

7592 could be related

Name and Version

0.3.6 ollama-cuda from AUR, was not able to find the used lamacpp version.

What operating system are you seeing the problem on?

Linux

Relevant log output

Sep 25 15:40:18 hostname ollama[268657]: time=2024-09-25T15:40:18.840+02:00 level=INFO source=sched.go:710 msg="new model will fit in available VRAM in single GPU, loading" model=/var/lib/ollama/.ollama/models/blobs/sha256-970aa74c0a90ef7482477cf803618e776e173c007bf957f635f1015bfcfef0e6 gpu=GPU-e919e64e-b05e-1b0e-79fe-4d6f163c34c8 parallel=4 available=11899699200 required="1.0 GiB"
Sep 25 15:40:18 hostname ollama[268657]: time=2024-09-25T15:40:18.840+02:00 level=INFO source=memory.go:309 msg="offload to cuda" layers.requested=-1 layers.model=13 layers.offload=13 layers.split="" memory.available="[11.1 GiB]" memory.required.full="1.0 GiB" memory.required.partial="1.0 GiB" memory.required.kv="96.0 MiB" memory.required.allocations="[1.0 GiB]" memory.weights.total="312.1 MiB" memory.weights.repeating="267.4 MiB" memory.weights.nonrepeating="44.7 MiB" memory.graph.full="192.0 MiB" memory.graph.partial="192.0 MiB"
Sep 25 15:40:18 hostname ollama[268657]: time=2024-09-25T15:40:18.842+02:00 level=INFO source=server.go:393 msg="starting llama server" cmd="/tmp/ollama140604727/runners/cuda_v12/ollama_llama_server --model /var/lib/ollama/.ollama/models/blobs/sha256-970aa74c0a90ef7482477cf803618e776e173c007bf957f635f1015bfcfef0e6 --ctx-size 32768 --batch-size 512 --embedding --log-disable --n-gpu-layers 13 --parallel 4 --port 35395"
Sep 25 15:40:18 hostname ollama[268657]: time=2024-09-25T15:40:18.842+02:00 level=INFO source=sched.go:445 msg="loaded runners" count=1
Sep 25 15:40:18 hostname ollama[268657]: time=2024-09-25T15:40:18.842+02:00 level=INFO source=server.go:593 msg="waiting for llama runner to start responding"
Sep 25 15:40:18 hostname ollama[268657]: time=2024-09-25T15:40:18.842+02:00 level=INFO source=server.go:627 msg="waiting for server to become available" status="llm server error"
Sep 25 15:40:18 hostname ollama[269097]: INFO [main] build info | build=3535 commit="1e6f6554a" tid="140699372605440" timestamp=1727271618
Sep 25 15:40:18 hostname ollama[269097]: INFO [main] system info | n_threads=6 n_threads_batch=-1 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="140699372605440" timestamp=1727271618 total_threads=12
Sep 25 15:40:18 hostname ollama[269097]: INFO [main] HTTP server listening | hostname="127.0.0.1" n_threads_http="11" port="35395" tid="140699372605440" timestamp=1727271618
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: loaded meta data with 24 key-value pairs and 112 tensors from /var/lib/ollama/.ollama/models/blobs/sha256-970aa74c0a90ef7482477cf803618e776e173c007bf957f635f1015bfcfef0e6 (version GGUF V3 (latest))
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv   0:                       general.architecture str              = nomic-bert
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv   1:                               general.name str              = nomic-embed-text-v1.5
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv   2:                     nomic-bert.block_count u32              = 12
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv   3:                  nomic-bert.context_length u32              = 2048
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv   4:                nomic-bert.embedding_length u32              = 768
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv   5:             nomic-bert.feed_forward_length u32              = 3072
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv   6:            nomic-bert.attention.head_count u32              = 12
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv   7:    nomic-bert.attention.layer_norm_epsilon f32              = 0.000000
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv   8:                          general.file_type u32              = 1
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv   9:                nomic-bert.attention.causal bool             = false
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  10:                    nomic-bert.pooling_type u32              = 1
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  11:                  nomic-bert.rope.freq_base f32              = 1000.000000
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  12:            tokenizer.ggml.token_type_count u32              = 2
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  13:                tokenizer.ggml.bos_token_id u32              = 101
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  14:                tokenizer.ggml.eos_token_id u32              = 102
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  15:                       tokenizer.ggml.model str              = bert
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  16:                      tokenizer.ggml.tokens arr[str,30522]   = ["[PAD]", "[unused0]", "[unused1]", "...
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  17:                      tokenizer.ggml.scores arr[f32,30522]   = [-1000.000000, -1000.000000, -1000.00...
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  18:                  tokenizer.ggml.token_type arr[i32,30522]   = [3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  19:            tokenizer.ggml.unknown_token_id u32              = 100
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  20:          tokenizer.ggml.seperator_token_id u32              = 102
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  21:            tokenizer.ggml.padding_token_id u32              = 0
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  22:                tokenizer.ggml.cls_token_id u32              = 101
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - kv  23:               tokenizer.ggml.mask_token_id u32              = 103
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - type  f32:   51 tensors
Sep 25 15:40:18 hostname ollama[268657]: llama_model_loader: - type  f16:   61 tensors
Sep 25 15:40:18 hostname ollama[268657]: llm_load_vocab: special tokens cache size = 5
Sep 25 15:40:18 hostname ollama[268657]: llm_load_vocab: token to piece cache size = 0.2032 MB
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: format           = GGUF V3 (latest)
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: arch             = nomic-bert
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: vocab type       = WPM
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_vocab          = 30522
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_merges         = 0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: vocab_only       = 0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_ctx_train      = 2048
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_embd           = 768
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_layer          = 12
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_head           = 12
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_head_kv        = 12
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_rot            = 64
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_swa            = 0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_embd_head_k    = 64
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_embd_head_v    = 64
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_gqa            = 1
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_embd_k_gqa     = 768
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_embd_v_gqa     = 768
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: f_norm_eps       = 1.0e-12
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: f_norm_rms_eps   = 0.0e+00
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: f_clamp_kqv      = 0.0e+00
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: f_max_alibi_bias = 0.0e+00
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: f_logit_scale    = 0.0e+00
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_ff             = 3072
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_expert         = 0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_expert_used    = 0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: causal attn      = 0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: pooling type     = 1
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: rope type        = 2
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: rope scaling     = linear
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: freq_base_train  = 1000.0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: freq_scale_train = 1
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: n_ctx_orig_yarn  = 2048
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: rope_finetuned   = unknown
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: ssm_d_conv       = 0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: ssm_d_inner      = 0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: ssm_d_state      = 0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: ssm_dt_rank      = 0
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: model type       = 137M
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: model ftype      = F16
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: model params     = 136.73 M
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: model size       = 260.86 MiB (16.00 BPW)
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: general.name     = nomic-embed-text-v1.5
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: BOS token        = 101 '[CLS]'
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: EOS token        = 102 '[SEP]'
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: UNK token        = 100 '[UNK]'
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: SEP token        = 102 '[SEP]'
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: PAD token        = 0 '[PAD]'
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: CLS token        = 101 '[CLS]'
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: MASK token       = 103 '[MASK]'
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: LF token         = 0 '[PAD]'
Sep 25 15:40:18 hostname ollama[268657]: llm_load_print_meta: max token length = 21
Sep 25 15:40:18 hostname ollama[268657]: ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
Sep 25 15:40:18 hostname ollama[268657]: ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
Sep 25 15:40:18 hostname ollama[268657]: ggml_cuda_init: found 1 CUDA devices:
Sep 25 15:40:18 hostname ollama[268657]:   Device 0: NVIDIA GeForce RTX 3060, compute capability 8.6, VMM: yes
Sep 25 15:40:18 hostname ollama[268657]: llm_load_tensors: ggml ctx size =    0.10 MiB
Sep 25 15:40:19 hostname ollama[268657]: time=2024-09-25T15:40:19.093+02:00 level=INFO source=server.go:627 msg="waiting for server to become available" status="llm server loading model"
Sep 25 15:40:19 hostname ollama[268657]: llm_load_tensors: offloading 12 repeating layers to GPU
Sep 25 15:40:19 hostname ollama[268657]: llm_load_tensors: offloading non-repeating layers to GPU
Sep 25 15:40:19 hostname ollama[268657]: llm_load_tensors: offloaded 13/13 layers to GPU
Sep 25 15:40:19 hostname ollama[268657]: llm_load_tensors:        CPU buffer size =    44.72 MiB
Sep 25 15:40:19 hostname ollama[268657]: llm_load_tensors:      CUDA0 buffer size =   216.15 MiB
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model: n_ctx      = 32768
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model: n_batch    = 512
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model: n_ubatch   = 512
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model: flash_attn = 0
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model: freq_base  = 1000.0
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model: freq_scale = 1
Sep 25 15:40:19 hostname ollama[268657]: llama_kv_cache_init:      CUDA0 KV buffer size =  1152.00 MiB
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model: KV self size  = 1152.00 MiB, K (f16):  576.00 MiB, V (f16):  576.00 MiB
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model:        CPU  output buffer size =     0.00 MiB
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model:      CUDA0 compute buffer size =    22.01 MiB
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model:  CUDA_Host compute buffer size =     2.51 MiB
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model: graph nodes  = 453
Sep 25 15:40:19 hostname ollama[268657]: llama_new_context_with_model: graph splits = 2
Sep 25 15:40:19 hostname ollama[269097]: [1727271619] warming up the model with an empty run
Sep 25 15:40:19 hostname ollama[268657]: /usr/include/c++/14.2.1/bits/stl_vector.h:1130: std::vector<_Tp, _Alloc>::reference std::vector<_Tp, _Alloc>::operator[](size_type) [with _Tp = long unsigned int; _Alloc = std::allocator<long unsigned int>; reference = long unsigned int&; size_type = long unsigned int]: Assertion '__n < this->size()' failed.
Sep 25 15:40:20 hostname ollama[268657]: time=2024-09-25T15:40:20.297+02:00 level=INFO source=server.go:627 msg="waiting for server to become available" status="llm server not responding"
Sep 25 15:40:21 hostname ollama[268657]: time=2024-09-25T15:40:21.187+02:00 level=INFO source=server.go:627 msg="waiting for server to become available" status="llm server error"
Sep 25 15:40:21 hostname ollama[268657]: time=2024-09-25T15:40:21.438+02:00 level=ERROR source=sched.go:451 msg="error loading llama server" error="llama runner process has terminated: signal: aborted (core dumped)"
Sep 25 15:40:21 hostname ollama[268657]: [GIN] 2024/09/25 - 15:40:21 | 500 |  2.676901285s |       127.0.0.1 | POST     "/api/embed"
Sep 25 15:40:26 hostname ollama[268657]: time=2024-09-25T15:40:26.512+02:00 level=WARN source=sched.go:642 msg="gpu VRAM usage didn't recover within timeout" seconds=5.07381477 model=/var/lib/ollama/.ollama/models/blobs/sha256-970aa74c0a90ef7482477cf803618e776e173c007bf957f635f1015bfcfef0e6
Sep 25 15:40:26 hostname ollama[268657]: time=2024-09-25T15:40:26.761+02:00 level=WARN source=sched.go:642 msg="gpu VRAM usage didn't recover within timeout" seconds=5.323291756 model=/var/lib/ollama/.ollama/models/blobs/sha256-970aa74c0a90ef7482477cf803618e776e173c007bf957f635f1015bfcfef0e6
Sep 25 15:40:27 hostname ollama[268657]: time=2024-09-25T15:40:27.012+02:00 level=WARN source=sched.go:642 msg="gpu VRAM usage didn't recover within timeout" seconds=5.573759559 model=/var/lib/ollama/.ollama/models/blobs/sha256-970aa74c0a90ef7482477cf803618e776e173c007bf957f635f1015bfcfef0e6
github-actions[bot] commented 3 weeks ago

This issue was closed because it has been inactive for 14 days since being marked as stale.