search

ggerganov / llama.cpp

LLM inference in C/C++
MIT License
67.23k stars 9.65k forks source link

Bug: When --parallel 4 is turned ON, the inferring result is apparently like fool .But when --parallel 4 is turned OFF everything is OK ? #8935

Closed hzgdeerHo closed 1 month ago

hzgdeerHo commented 2 months ago

What happened?

CMD which Works Normally:

CUDA_VISIBLE_DEVICES=0 ./llama-server -m /home/ubuntu/.cache/huggingface/hub/models--MaziyarPanahi--Meta-Llama-3.1-8B-Instruct-GGUF/snapshots/1f301d86d760b435a11a56de3863bc0121bfb98f/Meta-Llama-3.1-8B-Instruct.Q8_0.gguf --gpu-layers 33 -cb --ctx-size 16128 --flash-attn --batch-size 512 --chat-template llama3 --port 8866 --host 0.0.0.0

CMD which Works NOT Normally:

CUDA_VISIBLE_DEVICES=0 ./llama-server -m /home/ubuntu/.cache/huggingface/hub/models--MaziyarPanahi--Meta-Llama-3.1-8B-Instruct-GGUF/snapshots/1f301d86d760b435a11a56de3863bc0121bfb98f/Meta-Llama-3.1-8B-Instruct.Q8_0.gguf --gpu-layers 33 -cb --parallel 4 --ctx-size 16128 --flash-attn --batch-size 512 --chat-template llama3 --port 8866 --host 0.0.0.0

ubuntu@VM-0-16-ubuntu:~$ nvidia-smi Thu Aug 8 21:22:25 2024
+---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.183.01 Driver Version: 535.183.01 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 V100-SXM2-32GB Off | 00000000:00:08.0 Off | 0 | | N/A 34C P0 39W / 300W | 10194MiB / 32768MiB | 0% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | 0 N/A N/A 35134 C ./llama-server 10192MiB | +---------------------------------------------------------------------------------------+

Name and Version

ubuntu@VM-0-16-ubuntu:~/llama.cpp$ ^C ubuntu@VM-0-16-ubuntu:~/llama.cpp$ ./llama-cli --version version: 3549 (afd27f01) built with cc (Ubuntu 9.5.0-1ubuntu1~22.04) 9.5.0 for x86_64-linux-gnu

What operating system are you seeing the problem on?

Linux

Relevant log output

#####CMD which Works Normally:
CUDA_VISIBLE_DEVICES=0 ./llama-server -m /home/ubuntu/.cache/huggingface/hub/models--MaziyarPanahi--Meta-Llama-3.1-8B-Instruct-GGUF/snapshots/1f301d86d760b435a11a56de3863bc0121bfb98f/Meta-Llama-3.1-8B-Instruct.Q8_0.gguf  --gpu-layers 33 -cb --ctx-size 16128    --flash-attn  --batch-size 512 --chat-template llama3  --port 8866 --host 0.0.0.0     
INFO [                    main] build info | tid="140562966491136" timestamp=1723124595 build=3549 commit="afd27f01"
INFO [                    main] system info | tid="140562966491136" timestamp=1723124595 n_threads=10 n_threads_batch=-1 total_threads=10 system_info="AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 1 | AVX512_VBMI = 0 | AVX512_VNNI = 1 | AVX512_BF16 = 0 | FMA = 1 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 | "
llama_model_loader: loaded meta data with 33 key-value pairs and 291 tensors from /home/ubuntu/.cache/huggingface/hub/models--MaziyarPanahi--Meta-Llama-3.1-8B-Instruct-GGUF/snapshots/1f301d86d760b435a11a56de3863bc0121bfb98f/Meta-Llama-3.1-8B-Instruct.Q8_0.gguf (version GGUF V3 (latest))
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.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Models Meta Llama Meta Llama 3.1 8B I...
llama_model_loader: - kv   3:                           general.finetune str              = Instruct
llama_model_loader: - kv   4:                           general.basename str              = models-meta-llama-Meta-Llama-3.1
llama_model_loader: - kv   5:                         general.size_label str              = 8B
llama_model_loader: - kv   6:                            general.license str              = llama3.1
llama_model_loader: - kv   7:                               general.tags arr[str,6]       = ["facebook", "meta", "pytorch", "llam...
llama_model_loader: - kv   8:                          general.languages arr[str,8]       = ["en", "de", "fr", "it", "pt", "hi", ...
llama_model_loader: - kv   9:                          llama.block_count u32              = 32
llama_model_loader: - kv  10:                       llama.context_length u32              = 131072
llama_model_loader: - kv  11:                     llama.embedding_length u32              = 4096
llama_model_loader: - kv  12:                  llama.feed_forward_length u32              = 14336
llama_model_loader: - kv  13:                 llama.attention.head_count u32              = 32
llama_model_loader: - kv  14:              llama.attention.head_count_kv u32              = 8
llama_model_loader: - kv  15:                       llama.rope.freq_base f32              = 500000.000000
llama_model_loader: - kv  16:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  17:                          general.file_type u32              = 7
llama_model_loader: - kv  18:                           llama.vocab_size u32              = 128256
llama_model_loader: - kv  19:                 llama.rope.dimension_count u32              = 128
llama_model_loader: - kv  20:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  21:                         tokenizer.ggml.pre str              = smaug-bpe
llama_model_loader: - kv  22:                      tokenizer.ggml.tokens arr[str,128256]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  23:                  tokenizer.ggml.token_type arr[i32,128256]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  24:                      tokenizer.ggml.merges arr[str,280147]  = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "...
llama_model_loader: - kv  25:                tokenizer.ggml.bos_token_id u32              = 128000
llama_model_loader: - kv  26:                tokenizer.ggml.eos_token_id u32              = 128009
llama_model_loader: - kv  27:                    tokenizer.chat_template str              = {% set loop_messages = messages %}{% ...
llama_model_loader: - kv  28:               general.quantization_version u32              = 2
llama_model_loader: - kv  29:                      quantize.imatrix.file str              = ./Meta-Llama-3.1-8B-Instruct-GGUF_ima...
llama_model_loader: - kv  30:                   quantize.imatrix.dataset str              = group_40.txt
llama_model_loader: - kv  31:             quantize.imatrix.entries_count i32              = 224
llama_model_loader: - kv  32:              quantize.imatrix.chunks_count i32              = 68
llama_model_loader: - type  f32:   65 tensors
llama_model_loader: - type q8_0:  226 tensors
llm_load_vocab: special tokens cache size = 256
llm_load_vocab: token to piece cache size = 0.7999 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = llama
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 128256
llm_load_print_meta: n_merges         = 280147
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 131072
llm_load_print_meta: n_embd           = 4096
llm_load_print_meta: n_layer          = 32
llm_load_print_meta: n_head           = 32
llm_load_print_meta: n_head_kv        = 8
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  = 500000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 131072
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       = 8B
llm_load_print_meta: model ftype      = Q8_0
llm_load_print_meta: model params     = 8.03 B
llm_load_print_meta: model size       = 7.95 GiB (8.50 BPW) 
llm_load_print_meta: general.name     = Models Meta Llama Meta Llama 3.1 8B Instruct
llm_load_print_meta: BOS token        = 128000 '<|begin_of_text|>'
llm_load_print_meta: EOS token        = 128009 '<|eot_id|>'
llm_load_print_meta: LF token         = 128 'Ä'
llm_load_print_meta: EOT token        = 128009 '<|eot_id|>'
llm_load_print_meta: max token length = 256
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: Tesla V100-SXM2-32GB, compute capability 7.0, VMM: yes
llm_load_tensors: ggml ctx size =    0.27 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:        CPU buffer size =   532.31 MiB
llm_load_tensors:      CUDA0 buffer size =  7605.33 MiB
.........................................................................................
llama_new_context_with_model: n_ctx      = 16128
llama_new_context_with_model: n_batch    = 512
llama_new_context_with_model: n_ubatch   = 512
llama_new_context_with_model: flash_attn = 1
llama_new_context_with_model: freq_base  = 500000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init:      CUDA0 KV buffer size =  2016.00 MiB
llama_new_context_with_model: KV self size  = 2016.00 MiB, K (f16): 1008.00 MiB, V (f16): 1008.00 MiB
llama_new_context_with_model:  CUDA_Host  output buffer size =     0.98 MiB
llama_new_context_with_model:      CUDA0 compute buffer size =   258.50 MiB
llama_new_context_with_model:  CUDA_Host compute buffer size =    39.51 MiB
llama_new_context_with_model: graph nodes  = 903
llama_new_context_with_model: graph splits = 2
INFO [                    init] initializing slots | tid="140562966491136" timestamp=1723124598 n_slots=1
INFO [                    init] new slot | tid="140562966491136" timestamp=1723124598 id_slot=0 n_ctx_slot=16128
INFO [                    main] model loaded | tid="140562966491136" timestamp=1723124598
INFO [                    main] chat template | tid="140562966491136" timestamp=1723124598 chat_example="<|start_header_id|>system<|end_header_id|>\n\nYou are a helpful assistant<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nHello<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\nHi there<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nHow are you?<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" built_in=false
INFO [                    main] HTTP server listening | tid="140562966491136" timestamp=1723124598 n_threads_http="9" port="8866" hostname="0.0.0.0"
INFO [            update_slots] all slots are idle | tid="140562966491136" timestamp=1723124598
INFO [   launch_slot_with_task] slot is processing task | tid="140562966491136" timestamp=1723124772 id_slot=0 id_task=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124772 id_slot=0 id_task=0 p0=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124772 id_slot=0 id_task=0 p0=512
INFO [           print_timings] prompt eval time     =     391.14 ms /   907 tokens (    0.43 ms per token,  2318.87 tokens per second) | tid="140562966491136" timestamp=1723124773 id_slot=0 id_task=0 t_prompt_processing=391.138 n_prompt_tokens_processed=907 t_token=0.4312436604189636 n_tokens_second=2318.874668275647
INFO [           print_timings] generation eval time =    1003.99 ms /    74 runs   (   13.57 ms per token,    73.71 tokens per second) | tid="140562966491136" timestamp=1723124773 id_slot=0 id_task=0 t_token_generation=1003.985 n_decoded=74 t_token=13.567364864864865 n_tokens_second=73.70628047231781
INFO [           print_timings]           total time =    1395.12 ms | tid="140562966491136" timestamp=1723124773 id_slot=0 id_task=0 t_prompt_processing=391.138 t_token_generation=1003.985 t_total=1395.123
INFO [            update_slots] slot released | tid="140562966491136" timestamp=1723124773 id_slot=0 id_task=0 n_ctx=16128 n_past=980 n_system_tokens=0 n_cache_tokens=512 truncated=false
INFO [            update_slots] all slots are idle | tid="140562966491136" timestamp=1723124773
INFO [      log_server_request] request | tid="140561439375360" timestamp=1723124773 remote_addr="43.153.18.71" remote_port=57628 status=200 method="POST" path="/v1/chat/completions" params={}
INFO [   launch_slot_with_task] slot is processing task | tid="140562966491136" timestamp=1723124774 id_slot=0 id_task=76
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124774 id_slot=0 id_task=76 p0=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124774 id_slot=0 id_task=76 p0=512
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124774 id_slot=0 id_task=76 p0=1024
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124774 id_slot=0 id_task=76 p0=1536
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124774 id_slot=0 id_task=76 p0=2048
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124774 id_slot=0 id_task=76 p0=2560
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124775 id_slot=0 id_task=76 p0=3072
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124775 id_slot=0 id_task=76 p0=3584
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124775 id_slot=0 id_task=76 p0=4096
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124775 id_slot=0 id_task=76 p0=4608
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124776 id_slot=0 id_task=76 p0=5120
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124776 id_slot=0 id_task=76 p0=5632
INFO [           print_timings] prompt eval time     =    2921.58 ms /  5756 tokens (    0.51 ms per token,  1970.17 tokens per second) | tid="140562966491136" timestamp=1723124778 id_slot=0 id_task=76 t_prompt_processing=2921.578 n_prompt_tokens_processed=5756 t_token=0.5075708825573315 n_tokens_second=1970.168176239005
INFO [           print_timings] generation eval time =    2037.78 ms /   133 runs   (   15.32 ms per token,    65.27 tokens per second) | tid="140562966491136" timestamp=1723124778 id_slot=0 id_task=76 t_token_generation=2037.779 n_decoded=133 t_token=15.321646616541353 n_tokens_second=65.26713642647215
INFO [           print_timings]           total time =    4959.36 ms | tid="140562966491136" timestamp=1723124778 id_slot=0 id_task=76 t_prompt_processing=2921.578 t_token_generation=2037.779 t_total=4959.357
INFO [            update_slots] slot released | tid="140562966491136" timestamp=1723124778 id_slot=0 id_task=76 n_ctx=16128 n_past=5888 n_system_tokens=0 n_cache_tokens=5632 truncated=false
INFO [            update_slots] all slots are idle | tid="140562966491136" timestamp=1723124778
INFO [      log_server_request] request | tid="140558945218560" timestamp=1723124778 remote_addr="43.153.18.71" remote_port=57638 status=200 method="POST" path="/v1/chat/completions" params={}
INFO [            update_slots] all slots are idle | tid="140562966491136" timestamp=1723124778
INFO [   launch_slot_with_task] slot is processing task | tid="140562966491136" timestamp=1723124779 id_slot=0 id_task=222
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124779 id_slot=0 id_task=222 p0=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124779 id_slot=0 id_task=222 p0=512
INFO [            update_slots] kv cache rm [p0, end) | tid="140562966491136" timestamp=1723124779 id_slot=0 id_task=222 p0=1024
^C^CReceived second interrupt, terminating immediately.

####NOT Normally:

ubuntu@VM-0-16-ubuntu:~/llama.cpp$ CUDA_VISIBLE_DEVICES=0 ./llama-server -m /home/ubuntu/.cache/huggingface/hub/models--MaziyarPanahi--Meta-Llama-3.1-8B-Instruct-GGUF/snapshots/1f301d86d760b435a11a56de3863bc0121bfb98f/Meta-Llama-3.1-8B-Instruct.Q8_0.gguf  --gpu-layers 33 -cb --parallel 4 --ctx-size 16128    --flash-attn  --batch-size 512 --chat-template llama3  --port 8866 --host 0.0.0.0     
INFO [                    main] build info | tid="140411292143616" timestamp=1723125078 build=3549 commit="afd27f01"
INFO [                    main] system info | tid="140411292143616" timestamp=1723125078 n_threads=10 n_threads_batch=-1 total_threads=10 system_info="AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 1 | AVX512_VBMI = 0 | AVX512_VNNI = 1 | AVX512_BF16 = 0 | FMA = 1 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 | "
llama_model_loader: loaded meta data with 33 key-value pairs and 291 tensors from /home/ubuntu/.cache/huggingface/hub/models--MaziyarPanahi--Meta-Llama-3.1-8B-Instruct-GGUF/snapshots/1f301d86d760b435a11a56de3863bc0121bfb98f/Meta-Llama-3.1-8B-Instruct.Q8_0.gguf (version GGUF V3 (latest))
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.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Models Meta Llama Meta Llama 3.1 8B I...
llama_model_loader: - kv   3:                           general.finetune str              = Instruct
llama_model_loader: - kv   4:                           general.basename str              = models-meta-llama-Meta-Llama-3.1
llama_model_loader: - kv   5:                         general.size_label str              = 8B
llama_model_loader: - kv   6:                            general.license str              = llama3.1
llama_model_loader: - kv   7:                               general.tags arr[str,6]       = ["facebook", "meta", "pytorch", "llam...
llama_model_loader: - kv   8:                          general.languages arr[str,8]       = ["en", "de", "fr", "it", "pt", "hi", ...
llama_model_loader: - kv   9:                          llama.block_count u32              = 32
llama_model_loader: - kv  10:                       llama.context_length u32              = 131072
llama_model_loader: - kv  11:                     llama.embedding_length u32              = 4096
llama_model_loader: - kv  12:                  llama.feed_forward_length u32              = 14336
llama_model_loader: - kv  13:                 llama.attention.head_count u32              = 32
llama_model_loader: - kv  14:              llama.attention.head_count_kv u32              = 8
llama_model_loader: - kv  15:                       llama.rope.freq_base f32              = 500000.000000
llama_model_loader: - kv  16:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  17:                          general.file_type u32              = 7
llama_model_loader: - kv  18:                           llama.vocab_size u32              = 128256
llama_model_loader: - kv  19:                 llama.rope.dimension_count u32              = 128
llama_model_loader: - kv  20:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  21:                         tokenizer.ggml.pre str              = smaug-bpe
llama_model_loader: - kv  22:                      tokenizer.ggml.tokens arr[str,128256]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  23:                  tokenizer.ggml.token_type arr[i32,128256]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  24:                      tokenizer.ggml.merges arr[str,280147]  = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "...
llama_model_loader: - kv  25:                tokenizer.ggml.bos_token_id u32              = 128000
llama_model_loader: - kv  26:                tokenizer.ggml.eos_token_id u32              = 128009
llama_model_loader: - kv  27:                    tokenizer.chat_template str              = {% set loop_messages = messages %}{% ...
llama_model_loader: - kv  28:               general.quantization_version u32              = 2
llama_model_loader: - kv  29:                      quantize.imatrix.file str              = ./Meta-Llama-3.1-8B-Instruct-GGUF_ima...
llama_model_loader: - kv  30:                   quantize.imatrix.dataset str              = group_40.txt
llama_model_loader: - kv  31:             quantize.imatrix.entries_count i32              = 224
llama_model_loader: - kv  32:              quantize.imatrix.chunks_count i32              = 68
llama_model_loader: - type  f32:   65 tensors
llama_model_loader: - type q8_0:  226 tensors
llm_load_vocab: special tokens cache size = 256
llm_load_vocab: token to piece cache size = 0.7999 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = llama
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 128256
llm_load_print_meta: n_merges         = 280147
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 131072
llm_load_print_meta: n_embd           = 4096
llm_load_print_meta: n_layer          = 32
llm_load_print_meta: n_head           = 32
llm_load_print_meta: n_head_kv        = 8
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  = 500000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 131072
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       = 8B
llm_load_print_meta: model ftype      = Q8_0
llm_load_print_meta: model params     = 8.03 B
llm_load_print_meta: model size       = 7.95 GiB (8.50 BPW) 
llm_load_print_meta: general.name     = Models Meta Llama Meta Llama 3.1 8B Instruct
llm_load_print_meta: BOS token        = 128000 '<|begin_of_text|>'
llm_load_print_meta: EOS token        = 128009 '<|eot_id|>'
llm_load_print_meta: LF token         = 128 'Ä'
llm_load_print_meta: EOT token        = 128009 '<|eot_id|>'
llm_load_print_meta: max token length = 256
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: Tesla V100-SXM2-32GB, compute capability 7.0, VMM: yes
llm_load_tensors: ggml ctx size =    0.27 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:        CPU buffer size =   532.31 MiB
llm_load_tensors:      CUDA0 buffer size =  7605.33 MiB
.........................................................................................
llama_new_context_with_model: n_ctx      = 16128
llama_new_context_with_model: n_batch    = 512
llama_new_context_with_model: n_ubatch   = 512
llama_new_context_with_model: flash_attn = 1
llama_new_context_with_model: freq_base  = 500000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init:      CUDA0 KV buffer size =  2016.00 MiB
llama_new_context_with_model: KV self size  = 2016.00 MiB, K (f16): 1008.00 MiB, V (f16): 1008.00 MiB
llama_new_context_with_model:  CUDA_Host  output buffer size =     2.45 MiB
llama_new_context_with_model:      CUDA0 compute buffer size =   258.50 MiB
llama_new_context_with_model:  CUDA_Host compute buffer size =    39.51 MiB
llama_new_context_with_model: graph nodes  = 903
llama_new_context_with_model: graph splits = 2
INFO [                    init] initializing slots | tid="140411292143616" timestamp=1723125081 n_slots=4
INFO [                    init] new slot | tid="140411292143616" timestamp=1723125081 id_slot=0 n_ctx_slot=4032
INFO [                    init] new slot | tid="140411292143616" timestamp=1723125081 id_slot=1 n_ctx_slot=4032
INFO [                    init] new slot | tid="140411292143616" timestamp=1723125081 id_slot=2 n_ctx_slot=4032
INFO [                    init] new slot | tid="140411292143616" timestamp=1723125081 id_slot=3 n_ctx_slot=4032
INFO [                    main] model loaded | tid="140411292143616" timestamp=1723125081
INFO [                    main] chat template | tid="140411292143616" timestamp=1723125081 chat_example="<|start_header_id|>system<|end_header_id|>\n\nYou are a helpful assistant<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nHello<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\nHi there<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nHow are you?<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" built_in=false
INFO [                    main] HTTP server listening | tid="140411292143616" timestamp=1723125081 n_threads_http="9" port="8866" hostname="0.0.0.0"
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125081
INFO [   launch_slot_with_task] slot is processing task | tid="140411292143616" timestamp=1723125094 id_slot=0 id_task=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125094 id_slot=0 id_task=0 p0=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125094 id_slot=0 id_task=0 p0=512
INFO [           print_timings] prompt eval time     =     391.33 ms /   907 tokens (    0.43 ms per token,  2317.74 tokens per second) | tid="140411292143616" timestamp=1723125096 id_slot=0 id_task=0 t_prompt_processing=391.329 n_prompt_tokens_processed=907 t_token=0.4314542447629548 n_tokens_second=2317.7428710880104
INFO [           print_timings] generation eval time =    1014.42 ms /    74 runs   (   13.71 ms per token,    72.95 tokens per second) | tid="140411292143616" timestamp=1723125096 id_slot=0 id_task=0 t_token_generation=1014.416 n_decoded=74 t_token=13.708324324324325 n_tokens_second=72.94837620857714
INFO [           print_timings]           total time =    1405.75 ms | tid="140411292143616" timestamp=1723125096 id_slot=0 id_task=0 t_prompt_processing=391.329 t_token_generation=1014.416 t_total=1405.7450000000001
INFO [            update_slots] slot released | tid="140411292143616" timestamp=1723125096 id_slot=0 id_task=0 n_ctx=16128 n_past=980 n_system_tokens=0 n_cache_tokens=512 truncated=false
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125096
INFO [      log_server_request] request | tid="140409762209792" timestamp=1723125096 remote_addr="43.153.18.71" remote_port=36174 status=200 method="POST" path="/v1/chat/completions" params={}
INFO [   launch_slot_with_task] slot is processing task | tid="140411292143616" timestamp=1723125096 id_slot=1 id_task=76
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125096 id_slot=1 id_task=76 p0=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125096 id_slot=1 id_task=76 p0=512
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125096 id_slot=1 id_task=76 p0=1024
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125096 id_slot=1 id_task=76 p0=1536
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125097 id_slot=1 id_task=76 p0=2048
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125097 id_slot=1 id_task=76 p0=2560
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125097 id_slot=1 id_task=76 p0=3072
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125097 id_slot=1 id_task=76 p0=3584
INFO [           print_timings] prompt eval time     =    1927.38 ms /  3740 tokens (    0.52 ms per token,  1940.46 tokens per second) | tid="140411292143616" timestamp=1723125099 id_slot=1 id_task=76 t_prompt_processing=1927.375 n_prompt_tokens_processed=3740 t_token=0.5153409090909091 n_tokens_second=1940.463065049614
INFO [           print_timings] generation eval time =    1145.59 ms /    76 runs   (   15.07 ms per token,    66.34 tokens per second) | tid="140411292143616" timestamp=1723125099 id_slot=1 id_task=76 t_token_generation=1145.587 n_decoded=76 t_token=15.073513157894737 n_tokens_second=66.34153495107748
INFO [           print_timings]           total time =    3072.96 ms | tid="140411292143616" timestamp=1723125099 id_slot=1 id_task=76 t_prompt_processing=1927.375 t_token_generation=1145.587 t_total=3072.962
INFO [            update_slots] slot released | tid="140411292143616" timestamp=1723125099 id_slot=1 id_task=76 n_ctx=16128 n_past=3815 n_system_tokens=0 n_cache_tokens=3584 truncated=true
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125099
INFO [      log_server_request] request | tid="140409753817088" timestamp=1723125099 remote_addr="43.153.18.71" remote_port=36190 status=200 method="POST" path="/v1/chat/completions" params={}
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125099
INFO [   launch_slot_with_task] slot is processing task | tid="140411292143616" timestamp=1723125099 id_slot=2 id_task=161
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125099 id_slot=2 id_task=161 p0=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125099 id_slot=2 id_task=161 p0=512
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125099 id_slot=2 id_task=161 p0=1024
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125100 id_slot=2 id_task=161 p0=1536
INFO [           print_timings] prompt eval time     =    1117.78 ms /  1543 tokens (    0.72 ms per token,  1380.41 tokens per second) | tid="140411292143616" timestamp=1723125115 id_slot=2 id_task=161 t_prompt_processing=1117.784 n_prompt_tokens_processed=1543 t_token=0.7244225534672716 n_tokens_second=1380.409810840019
INFO [           print_timings] generation eval time =   14567.23 ms /   906 runs   (   16.08 ms per token,    62.19 tokens per second) | tid="140411292143616" timestamp=1723125115 id_slot=2 id_task=161 t_token_generation=14567.234 n_decoded=906 t_token=16.07862472406181 n_tokens_second=62.19437403147364
INFO [           print_timings]           total time =   15685.02 ms | tid="140411292143616" timestamp=1723125115 id_slot=2 id_task=161 t_prompt_processing=1117.784 t_token_generation=14567.234 t_total=15685.018
INFO [            update_slots] slot released | tid="140411292143616" timestamp=1723125115 id_slot=2 id_task=161 n_ctx=16128 n_past=2448 n_system_tokens=0 n_cache_tokens=1536 truncated=false
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125115
INFO [      log_server_request] request | tid="140409762209792" timestamp=1723125115 remote_addr="43.153.18.71" remote_port=36174 status=200 method="POST" path="/v1/chat/completions" params={}
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125115
INFO [   launch_slot_with_task] slot is processing task | tid="140411292143616" timestamp=1723125115 id_slot=2 id_task=1072
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125115 id_slot=2 id_task=1072 p0=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125115 id_slot=2 id_task=1072 p0=512
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125115 id_slot=2 id_task=1072 p0=1024
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125115 id_slot=2 id_task=1072 p0=1536
INFO [           print_timings] prompt eval time     =    1318.40 ms /  1781 tokens (    0.74 ms per token,  1350.88 tokens per second) | tid="140411292143616" timestamp=1723125123 id_slot=2 id_task=1072 t_prompt_processing=1318.401 n_prompt_tokens_processed=1781 t_token=0.7402588433464347 n_tokens_second=1350.8788297338974
INFO [           print_timings] generation eval time =    7254.65 ms /   450 runs   (   16.12 ms per token,    62.03 tokens per second) | tid="140411292143616" timestamp=1723125123 id_slot=2 id_task=1072 t_token_generation=7254.651 n_decoded=450 t_token=16.121446666666667 n_tokens_second=62.02917273346437
INFO [           print_timings]           total time =    8573.05 ms | tid="140411292143616" timestamp=1723125123 id_slot=2 id_task=1072 t_prompt_processing=1318.401 t_token_generation=7254.651 t_total=8573.052
INFO [            update_slots] slot released | tid="140411292143616" timestamp=1723125123 id_slot=2 id_task=1072 n_ctx=16128 n_past=2230 n_system_tokens=0 n_cache_tokens=1536 truncated=false
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125123
INFO [      log_server_request] request | tid="140409745424384" timestamp=1723125123 remote_addr="43.153.18.71" remote_port=42030 status=200 method="POST" path="/v1/chat/completions" params={}
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125123
INFO [   launch_slot_with_task] slot is processing task | tid="140411292143616" timestamp=1723125123 id_slot=3 id_task=1527
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125123 id_slot=3 id_task=1527 p0=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125124 id_slot=3 id_task=1527 p0=512
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125124 id_slot=3 id_task=1527 p0=1024
INFO [           print_timings] prompt eval time     =    1365.03 ms /  1534 tokens (    0.89 ms per token,  1123.78 tokens per second) | tid="140411292143616" timestamp=1723125133 id_slot=3 id_task=1527 t_prompt_processing=1365.034 n_prompt_tokens_processed=1534 t_token=0.8898526727509779 n_tokens_second=1123.7815321816158
INFO [           print_timings] generation eval time =    7954.98 ms /   471 runs   (   16.89 ms per token,    59.21 tokens per second) | tid="140411292143616" timestamp=1723125133 id_slot=3 id_task=1527 t_token_generation=7954.977 n_decoded=471 t_token=16.889547770700638 n_tokens_second=59.208216441103474
INFO [           print_timings]           total time =    9320.01 ms | tid="140411292143616" timestamp=1723125133 id_slot=3 id_task=1527 t_prompt_processing=1365.034 t_token_generation=7954.977 t_total=9320.011
INFO [            update_slots] slot released | tid="140411292143616" timestamp=1723125133 id_slot=3 id_task=1527 n_ctx=16128 n_past=2004 n_system_tokens=0 n_cache_tokens=1024 truncated=false
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125133
INFO [      log_server_request] request | tid="140409745424384" timestamp=1723125133 remote_addr="43.153.18.71" remote_port=42030 status=200 method="POST" path="/v1/chat/completions" params={}
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125133
INFO [   launch_slot_with_task] slot is processing task | tid="140411292143616" timestamp=1723125133 id_slot=0 id_task=2002
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125133 id_slot=0 id_task=2002 p0=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125133 id_slot=0 id_task=2002 p0=512
INFO [           print_timings] prompt eval time     =     907.58 ms /   906 tokens (    1.00 ms per token,   998.26 tokens per second) | tid="140411292143616" timestamp=1723125135 id_slot=0 id_task=2002 t_prompt_processing=907.579 n_prompt_tokens_processed=906 t_token=1.001742825607064 n_tokens_second=998.2602065495125
INFO [           print_timings] generation eval time =    1150.37 ms /    68 runs   (   16.92 ms per token,    59.11 tokens per second) | tid="140411292143616" timestamp=1723125135 id_slot=0 id_task=2002 t_token_generation=1150.367 n_decoded=68 t_token=16.91716176470588 n_tokens_second=59.11157048142028
INFO [           print_timings]           total time =    2057.95 ms | tid="140411292143616" timestamp=1723125135 id_slot=0 id_task=2002 t_prompt_processing=907.579 t_token_generation=1150.367 t_total=2057.946
INFO [            update_slots] slot released | tid="140411292143616" timestamp=1723125135 id_slot=0 id_task=2002 n_ctx=16128 n_past=973 n_system_tokens=0 n_cache_tokens=512 truncated=false
INFO [            update_slots] all slots are idle | tid="140411292143616" timestamp=1723125135
INFO [      log_server_request] request | tid="140409745424384" timestamp=1723125135 remote_addr="43.153.18.71" remote_port=42030 status=200 method="POST" path="/v1/chat/completions" params={}
INFO [   launch_slot_with_task] slot is processing task | tid="140411292143616" timestamp=1723125135 id_slot=1 id_task=2072
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125135 id_slot=1 id_task=2072 p0=0
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125135 id_slot=1 id_task=2072 p0=512
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125136 id_slot=1 id_task=2072 p0=1024
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125136 id_slot=1 id_task=2072 p0=1536
INFO [            update_slots] kv cache rm [p0, end) | tid="140411292143616" timestamp=1723125137 id_slot=1 id_task=2072 p0=2048
^C^CReceived second interrupt, terminating immediately.
ngxson commented 2 months ago

can you try with --ctx-size 16384 instead of --ctx-size 16128 ? (I'm not sure if it fixes the problem or not)

hzgdeerHo commented 2 months ago

It does not work with --ctx-size 16384 ,but If I set like this : --ctx-size 32000 ,It works, I think it is related about the truncated process is enabled .How could I disabled the truncated process. THANKS ！

ngxson commented 2 months ago

I'm not sure what you mean by "truncated process".

Keep in mind that the actual context size will be --ctx-size divided by --parallel, so for example with 16384 you have 16384 / 4096 = 4096 tokens per slot, so it's normal to increase ctx size if you set a high value for --parallel

hzgdeerHo commented 2 months ago

THANKS！

github-actions[bot] commented 1 month ago

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