search

ggerganov / llama.cpp

LLM inference in C/C++
MIT License
61.45k stars 8.79k forks source link

Bug: [SYCL] Inference not working correctly on multiple GPUs #8294

Open ch1y0q opened 6 days ago

ch1y0q commented 6 days ago

What happened?

I am using Llama.cpp + SYCL to perform inference on a multiple GPU server. However, I get a Segmentation Fault when using multiple GPUs. The same model can produce inference output correctly with single GPU mode.

git clone https://github.com/ggerganov/llama.cpp.git
source /opt/intel/oneapi/setvars.sh
cmake -B build -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
cmake --build build --config Release -j -v

cd ~/llama.cpp/
./build/bin/llama-ls-sycl-device

## single gpu, ok
./build/bin/llama-cli -m ~/mistral-7b-v0.1.Q4_0.gguf -p "Building a website can be done in 10 simple steps:\nStep 1:" -n 400 -e -ngl 33 -s 0 -sm none -mg 0

## multiple gpus, Segmentation Fault, core dumped
./build/bin/llama-cli -m ~/mistral-7b-v0.1.Q4_0.gguf -p "Building a website can be done in 10 simple steps:\nStep 1:" -n 400 -e -ngl 33 -s 0 -sm layer

image image

Output of ./build/bin/llama-ls-sycl-device:

found 8 SYCL devices:
|  |                   |                                       |       |Max    |        |Max  |Global |                     |
|  |                   |                                       |       |compute|Max work|sub  |mem    |                     |
|ID|        Device Type|                                   Name|Version|units  |group   |group|size   |       Driver version|
|--|-------------------|---------------------------------------|-------|-------|--------|-----|-------|---------------------|
| 0| [level_zero:gpu:0]|                Intel Arc A770 Graphics|    1.3|    512|    1024|   32| 16225M|            1.3.26241|
| 1| [level_zero:gpu:1]|                Intel Arc A770 Graphics|    1.3|    512|    1024|   32| 16225M|            1.3.26241|
| 2| [level_zero:gpu:2]|                 Intel UHD Graphics 770|    1.3|     32|     512|   32| 53751M|            1.3.26241|
| 3|     [opencl:gpu:0]|                Intel Arc A770 Graphics|    3.0|    512|    1024|   32| 16225M|       23.17.26241.33|
| 4|     [opencl:gpu:1]|                Intel Arc A770 Graphics|    3.0|    512|    1024|   32| 16225M|       23.17.26241.33|
| 5|     [opencl:gpu:2]|                 Intel UHD Graphics 770|    3.0|     32|     512|   32| 53751M|       23.17.26241.33|
| 6|     [opencl:cpu:0]|                   Intel Core i9-14900K|    3.0|     32|    8192|   64| 67189M|2023.16.11.0.22_160000|
| 7|     [opencl:acc:0]|            Intel FPGA Emulation Device|    1.2|     32|67108864|   64| 67189M|2023.16.11.0.22_160000|

Name and Version

./llama-cli --version
version: 3292 (20fc3804)
built with Intel(R) oneAPI DPC++/C++ Compiler 2024.0.1 (2024.0.1.20231122) for x86_64-unknown-linux-gnu

What operating system are you seeing the problem on?

Linux

Relevant log output

Log start
main: build = 3292 (20fc3804)
main: built with Intel(R) oneAPI DPC++/C++ Compiler 2024.0.1 (2024.0.1.20231122) for x86_64-unknown-linux-gnu
main: seed  = 0
llama_model_loader: loaded meta data with 20 key-value pairs and 291 tensors from /home/arda/qiyue/mistral-7b-v0.1.Q4_0.gguf (version GGUF V2)
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv   0:                       general.architecture str              = llama
llama_model_loader: - kv   1:                               general.name str              = mistralai_mistral-7b-v0.1
llama_model_loader: - kv   2:                       llama.context_length u32              = 32768
llama_model_loader: - kv   3:                     llama.embedding_length u32              = 4096
llama_model_loader: - kv   4:                          llama.block_count u32              = 32
llama_model_loader: - kv   5:                  llama.feed_forward_length u32              = 14336
llama_model_loader: - kv   6:                 llama.rope.dimension_count u32              = 128
llama_model_loader: - kv   7:                 llama.attention.head_count u32              = 32
llama_model_loader: - kv   8:              llama.attention.head_count_kv u32              = 8
llama_model_loader: - kv   9:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  10:                       llama.rope.freq_base f32              = 10000.000000
llama_model_loader: - kv  11:                          general.file_type u32              = 2
llama_model_loader: - kv  12:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  13:                      tokenizer.ggml.tokens arr[str,32000]   = ["<unk>", "<s>", "</s>", "<0x00>", "<...
llama_model_loader: - kv  14:                      tokenizer.ggml.scores arr[f32,32000]   = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv  15:                  tokenizer.ggml.token_type arr[i32,32000]   = [2, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 6, ...
llama_model_loader: - kv  16:                tokenizer.ggml.bos_token_id u32              = 1
llama_model_loader: - kv  17:                tokenizer.ggml.eos_token_id u32              = 2
llama_model_loader: - kv  18:            tokenizer.ggml.unknown_token_id u32              = 0
llama_model_loader: - kv  19:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:   65 tensors
llama_model_loader: - type q4_0:  225 tensors
llama_model_loader: - type q6_K:    1 tensors
llm_load_vocab: special tokens cache size = 259
llm_load_vocab: token to piece cache size = 0.1637 MB
llm_load_print_meta: format           = GGUF V2
llm_load_print_meta: arch             = llama
llm_load_print_meta: vocab type       = SPM
llm_load_print_meta: n_vocab          = 32000
llm_load_print_meta: n_merges         = 0
llm_load_print_meta: n_ctx_train      = 32768
llm_load_print_meta: n_embd           = 4096
llm_load_print_meta: n_head           = 32
llm_load_print_meta: n_head_kv        = 8
llm_load_print_meta: n_layer          = 32
llm_load_print_meta: n_rot            = 128
llm_load_print_meta: n_swa            = 0
llm_load_print_meta: n_embd_head_k    = 128
llm_load_print_meta: n_embd_head_v    = 128
llm_load_print_meta: n_gqa            = 4
llm_load_print_meta: n_embd_k_gqa     = 1024
llm_load_print_meta: n_embd_v_gqa     = 1024
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
llm_load_print_meta: f_clamp_kqv      = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale    = 0.0e+00
llm_load_print_meta: n_ff             = 14336
llm_load_print_meta: n_expert         = 0
llm_load_print_meta: n_expert_used    = 0
llm_load_print_meta: causal attn      = 1
llm_load_print_meta: pooling type     = 0
llm_load_print_meta: rope type        = 0
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 10000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 32768
llm_load_print_meta: rope_finetuned   = unknown
llm_load_print_meta: ssm_d_conv       = 0
llm_load_print_meta: ssm_d_inner      = 0
llm_load_print_meta: ssm_d_state      = 0
llm_load_print_meta: ssm_dt_rank      = 0
llm_load_print_meta: model type       = 7B
llm_load_print_meta: model ftype      = Q4_0
llm_load_print_meta: model params     = 7.24 B
llm_load_print_meta: model size       = 3.83 GiB (4.54 BPW) 
llm_load_print_meta: general.name     = mistralai_mistral-7b-v0.1
llm_load_print_meta: BOS token        = 1 '<s>'
llm_load_print_meta: EOS token        = 2 '</s>'
llm_load_print_meta: UNK token        = 0 '<unk>'
llm_load_print_meta: LF token         = 13 '<0x0A>'
llm_load_print_meta: max token length = 48
ggml_sycl_init: GGML_SYCL_FORCE_MMQ:   no
ggml_sycl_init: SYCL_USE_XMX: yes
ggml_sycl_init: found 8 SYCL devices:
get_memory_info: [warning] ext_intel_free_memory is not supported (export/set ZES_ENABLE_SYSMAN=1 to support), use total memory as free memory
get_memory_info: [warning] ext_intel_free_memory is not supported (export/set ZES_ENABLE_SYSMAN=1 to support), use total memory as free memory
get_memory_info: [warning] ext_intel_free_memory is not supported (export/set ZES_ENABLE_SYSMAN=1 to support), use total memory as free memory
get_memory_info: [warning] ext_intel_free_memory is not supported (export/set ZES_ENABLE_SYSMAN=1 to support), use total memory as free memory
get_memory_info: [warning] ext_intel_free_memory is not supported (export/set ZES_ENABLE_SYSMAN=1 to support), use total memory as free memory
get_memory_info: [warning] ext_intel_free_memory is not supported (export/set ZES_ENABLE_SYSMAN=1 to support), use total memory as free memory
get_memory_info: [warning] ext_intel_free_memory is not supported (export/set ZES_ENABLE_SYSMAN=1 to support), use total memory as free memory
get_memory_info: [warning] ext_intel_free_memory is not supported (export/set ZES_ENABLE_SYSMAN=1 to support), use total memory as free memory
llm_load_tensors: ggml ctx size =    1.23 MiB
llm_load_tensors: offloading 32 repeating layers to GPU
llm_load_tensors: offloading non-repeating layers to GPU
llm_load_tensors: offloaded 33/33 layers to GPU
llm_load_tensors:      SYCL0 buffer size =   234.06 MiB
llm_load_tensors:      SYCL1 buffer size =   234.06 MiB
llm_load_tensors:      SYCL2 buffer size =   702.19 MiB
llm_load_tensors:      SYCL3 buffer size =   234.06 MiB
llm_load_tensors:      SYCL4 buffer size =   117.03 MiB
llm_load_tensors:      SYCL5 buffer size =   702.19 MiB
llm_load_tensors:      SYCL6 buffer size =   819.22 MiB
llm_load_tensors:      SYCL7 buffer size =   804.74 MiB
llm_load_tensors:        CPU buffer size =    70.31 MiB
.................................................................................................
llama_new_context_with_model: n_ctx      = 32768
llama_new_context_with_model: n_batch    = 2048
llama_new_context_with_model: n_ubatch   = 512
llama_new_context_with_model: flash_attn = 0
llama_new_context_with_model: freq_base  = 10000.0
llama_new_context_with_model: freq_scale = 1
[SYCL] call ggml_check_sycl
ggml_check_sycl: GGML_SYCL_DEBUG: 0
ggml_check_sycl: GGML_SYCL_F16: no
found 8 SYCL devices:
|  |                   |                                       |       |Max    |        |Max  |Global |                     |
|  |                   |                                       |       |compute|Max work|sub  |mem    |                     |
|ID|        Device Type|                                   Name|Version|units  |group   |group|size   |       Driver version|
|--|-------------------|---------------------------------------|-------|-------|--------|-----|-------|---------------------|
| 0| [level_zero:gpu:0]|                Intel Arc A770 Graphics|    1.3|    512|    1024|   32| 16225M|            1.3.26241|
| 1| [level_zero:gpu:1]|                Intel Arc A770 Graphics|    1.3|    512|    1024|   32| 16225M|            1.3.26241|
| 2| [level_zero:gpu:2]|                 Intel UHD Graphics 770|    1.3|     32|     512|   32| 53751M|            1.3.26241|
| 3|     [opencl:gpu:0]|                Intel Arc A770 Graphics|    3.0|    512|    1024|   32| 16225M|       23.17.26241.33|
| 4|     [opencl:gpu:1]|                Intel Arc A770 Graphics|    3.0|    512|    1024|   32| 16225M|       23.17.26241.33|
| 5|     [opencl:gpu:2]|                 Intel UHD Graphics 770|    3.0|     32|     512|   32| 53751M|       23.17.26241.33|
| 6|     [opencl:cpu:0]|                   Intel Core i9-14900K|    3.0|     32|    8192|   64| 67189M|2023.16.11.0.22_160000|
| 7|     [opencl:acc:0]|            Intel FPGA Emulation Device|    1.2|     32|67108864|   64| 67189M|2023.16.11.0.22_160000|
llama_kv_cache_init:      SYCL0 KV buffer size =   256.00 MiB
llama_kv_cache_init:      SYCL1 KV buffer size =   256.00 MiB
llama_kv_cache_init:      SYCL2 KV buffer size =   768.00 MiB
llama_kv_cache_init:      SYCL3 KV buffer size =   256.00 MiB
llama_kv_cache_init:      SYCL4 KV buffer size =   128.00 MiB
llama_kv_cache_init:      SYCL5 KV buffer size =   768.00 MiB
llama_kv_cache_init:      SYCL6 KV buffer size =   896.00 MiB
llama_kv_cache_init:      SYCL7 KV buffer size =   768.00 MiB
llama_new_context_with_model: KV self size  = 4096.00 MiB, K (f16): 2048.00 MiB, V (f16): 2048.00 MiB
llama_new_context_with_model:  SYCL_Host  output buffer size =     0.12 MiB
llama_new_context_with_model:      SYCL0 compute buffer size =  2144.00 MiB
llama_new_context_with_model:      SYCL1 compute buffer size =  2144.00 MiB
llama_new_context_with_model:      SYCL2 compute buffer size =  2144.00 MiB
llama_new_context_with_model:      SYCL3 compute buffer size =  2144.00 MiB
llama_new_context_with_model:      SYCL4 compute buffer size =  2144.00 MiB
llama_new_context_with_model:      SYCL5 compute buffer size =  2144.00 MiB
llama_new_context_with_model:      SYCL6 compute buffer size =  2144.00 MiB
llama_new_context_with_model:      SYCL7 compute buffer size =  2144.00 MiB
llama_new_context_with_model:  SYCL_Host compute buffer size =    72.01 MiB
llama_new_context_with_model: graph nodes  = 1030
llama_new_context_with_model: graph splits = 9
NeoZhangJianyu commented 5 days ago

@ch1y0q This PR https://github.com/ggerganov/llama.cpp/pull/8014 fix this issue. but it's not approved. You could use old release: https://github.com/ggerganov/llama.cpp/commit/fb76ec31a9914b7761c1727303ab30380fd4f05c or merge the PR to ggerganov/llama.cpp