llama.cpp
llama.cpp copied to clipboard
Published
1 month ago
•
ggml-org
ggml-org
Bug: Failed to run qwen2-57b-a14b-instruct-fp16.
What happened?
I am trying to run Qwen2-57B-A14B-instruct, and I used llama-gguf-split to merge the gguf files from Qwen/Qwen2-57B-A14B-Instruct-GGUF. But it's aborted with terminate called after throwing an instance of 'std::length_error' what(): vector::_M_default_append Aborted (core dumped)。
Name and Version
./build/bin/llama-cli --version version: 3808 (699a0dc1) built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu
What operating system are you seeing the problem on?
Linux
Relevant log output
`(llama) root@201edf3683be:/home/llama.cpp# ./build/bin/llama-cli -m ./models/qwen2-57b-a14b-instruct-fp16.gguf -p "Beijing is the capital of" -n 64 -c 4096
build: 3808 (699a0dc1) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu (debug)
main: llama backend init
main: load the model and apply lora adapter, if any
llama_model_loader: loaded meta data with 28 key-value pairs and 479 tensors from ./models/qwen2-57b-a14b-instruct-fp16.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 = qwen2moe
llama_model_loader: - kv 1: general.name str = Qwen2-MoE-A14.2B-Chat
llama_model_loader: - kv 2: qwen2moe.block_count u32 = 28
llama_model_loader: - kv 3: qwen2moe.context_length u32 = 32768
llama_model_loader: - kv 4: qwen2moe.embedding_length u32 = 3584
llama_model_loader: - kv 5: qwen2moe.attention.head_count u32 = 28
llama_model_loader: - kv 6: qwen2moe.attention.head_count_kv u32 = 4
llama_model_loader: - kv 7: qwen2moe.rope.freq_base f32 = 1000000.000000
llama_model_loader: - kv 8: qwen2moe.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 9: qwen2moe.expert_used_count u32 = 8
llama_model_loader: - kv 10: qwen2moe.expert_count u32 = 64
llama_model_loader: - kv 11: qwen2moe.expert_feed_forward_length u32 = 2560
llama_model_loader: - kv 12: qwen2moe.feed_forward_length u32 = 20480
llama_model_loader: - kv 13: general.file_type u32 = 1
llama_model_loader: - kv 14: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 15: tokenizer.ggml.pre str = qwen2
llama_model_loader: - kv 16: tokenizer.ggml.tokens arr[str,151936] = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv 17: tokenizer.ggml.token_type arr[i32,151936] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 18: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv 19: tokenizer.ggml.eos_token_id u32 = 151643
llama_model_loader: - kv 20: tokenizer.ggml.padding_token_id u32 = 151643
llama_model_loader: - kv 21: tokenizer.ggml.bos_token_id u32 = 151643
llama_model_loader: - kv 22: tokenizer.chat_template str = {% for message in messages %}{{'<|im_...
llama_model_loader: - kv 23: tokenizer.ggml.add_bos_token bool = false
llama_model_loader: - kv 24: general.quantization_version u32 = 2
llama_model_loader: - kv 25: split.no u16 = 0
llama_model_loader: - kv 26: split.count u16 = 0
llama_model_loader: - kv 27: split.tensors.count i32 = 479
llama_model_loader: - type f32: 197 tensors
llama_model_loader: - type f16: 282 tensors
llm_load_vocab: special tokens cache size = 293
llm_load_vocab: token to piece cache size = 0.9338 MB
llm_load_print_meta: format = GGUF V3 (latest)
llm_load_print_meta: arch = qwen2moe
llm_load_print_meta: vocab type = BPE
llm_load_print_meta: n_vocab = 151936
llm_load_print_meta: n_merges = 151387
llm_load_print_meta: vocab_only = 0
llm_load_print_meta: n_ctx_train = 32768
llm_load_print_meta: n_embd = 3584
llm_load_print_meta: n_layer = 28
llm_load_print_meta: n_head = 28
llm_load_print_meta: n_head_kv = 4
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 = 7
llm_load_print_meta: n_embd_k_gqa = 512
llm_load_print_meta: n_embd_v_gqa = 512
llm_load_print_meta: f_norm_eps = 0.0e+00
llm_load_print_meta: f_norm_rms_eps = 1.0e-06
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 = 20480
llm_load_print_meta: n_expert = 64
llm_load_print_meta: n_expert_used = 8
llm_load_print_meta: causal attn = 1
llm_load_print_meta: pooling type = 0
llm_load_print_meta: rope type = 2
llm_load_print_meta: rope scaling = linear
llm_load_print_meta: freq_base_train = 1000000.0
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: ssm_dt_b_c_rms = 0
llm_load_print_meta: model type = 57B.A14B
llm_load_print_meta: model ftype = F16
llm_load_print_meta: model params = 57.41 B
llm_load_print_meta: model size = 106.94 GiB (16.00 BPW)
llm_load_print_meta: general.name = Qwen2-MoE-A14.2B-Chat
llm_load_print_meta: BOS token = 151643 '<|endoftext|>'
llm_load_print_meta: EOS token = 151643 '<|endoftext|>'
llm_load_print_meta: PAD token = 151643 '<|endoftext|>'
llm_load_print_meta: LF token = 148848 'ÄĬ'
llm_load_print_meta: EOT token = 151645 '<|im_end|>'
llm_load_print_meta: max token length = 256
llm_load_print_meta: n_ff_exp = 2560
llm_load_print_meta: n_ff_shexp = 0
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 4 CUDA devices:
Device 0: NVIDIA H100 80GB HBM3, compute capability 9.0, VMM: yes
Device 1: NVIDIA H100 80GB HBM3, compute capability 9.0, VMM: yes
Device 2: NVIDIA H100 80GB HBM3, compute capability 9.0, VMM: yes
Device 3: NVIDIA H100 80GB HBM3, compute capability 9.0, VMM: yes
llm_load_tensors: ggml ctx size = 0.20 MiB
llm_load_tensors: offloading 0 repeating layers to GPU
llm_load_tensors: offloaded 0/29 layers to GPU
llm_load_tensors: CPU buffer size = 109511.40 MiB
.............................................................................................
llama_new_context_with_model: n_ctx = 4096
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 = 1000000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init: CUDA_Host KV buffer size = 224.00 MiB
llama_new_context_with_model: KV self size = 224.00 MiB, K (f16): 112.00 MiB, V (f16): 112.00 MiB
llama_new_context_with_model: CUDA_Host output buffer size = 0.58 MiB
ggml_gallocr_reserve_n: reallocating CUDA0 buffer from size 0.00 MiB to 1349.38 MiB
ggml_gallocr_reserve_n: reallocating CUDA1 buffer from size 0.00 MiB to 0.00 MiB
ggml_gallocr_reserve_n: reallocating CUDA2 buffer from size 0.00 MiB to 0.00 MiB
ggml_gallocr_reserve_n: reallocating CUDA3 buffer from size 0.00 MiB to 0.00 MiB
ggml_gallocr_reserve_n: reallocating CUDA_Host buffer from size 0.00 MiB to 15.01 MiB
llama_new_context_with_model: CUDA0 compute buffer size = 1349.38 MiB
llama_new_context_with_model: CUDA_Host compute buffer size = 15.01 MiB
llama_new_context_with_model: graph nodes = 1910
llama_new_context_with_model: graph splits = 536
llama_init_from_gpt_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
ggml_backend_sched_alloc_splits: failed to allocate graph, reserving (backend_ids_changed = 1)
main: llama threadpool init, n_threads = 128
system_info: n_threads = 128 (n_threads_batch = 128) / 255 | AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | FMA = 1 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | RISCV_VECT = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 |
terminate called after throwing an instance of 'std::length_error'
what(): vector::_M_default_append
Aborted (core dumped)`
Cannot reproduce on a single GPU:
llama-cli --hf-repo Qwen/Qwen2-57B-A14B-Instruct-GGUF --hf-file qwen2-57b-a14b-instruct-q3_k_m.gguf -p "Beijing is the capital of" -n 64 -c 4096
Output
/home/phymbert/workspaces/llama.cpp/cmake-build-debug/bin/llama-cli --hf-repo Qwen/Qwen2-57B-A14B-Instruct-GGUF --hf-file qwen2-57b-a14b-instruct-q3_k_m.gguf -p "Beijing is the capital of" -n 64 -c 4096
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: NVIDIA GeForce RTX 3050 Laptop GPU, compute capability 8.6, VMM: yes
register_backend: registered backend CUDA (1 devices)
register_device: registered device CUDA0 (NVIDIA GeForce RTX 3050 Laptop GPU)
register_backend: registered backend CPU (1 devices)
register_device: registered device CPU (11th Gen Intel(R) Core(TM) i5-11400H @ 2.70GHz)
build: 4393 (d79d8f39) with cc (Ubuntu 13.2.0-23ubuntu4) 13.2.0 for x86_64-linux-gnu (debug)
main: llama backend init
main: load the model and apply lora adapter, if any
common_download_file: no previous model file found /home/phymbert/.cache/llama.cpp/Qwen_Qwen2-57B-A14B-Instruct-GGUF_qwen2-57b-a14b-instruct-q3_k_m.gguf
curl_perform_with_retry: Trying to download from https://huggingface.co/Qwen/Qwen2-57B-A14B-Instruct-GGUF/resolve/main/qwen2-57b-a14b-instruct-q3_k_m.gguf (attempt 1 of 3)...
common_download_file: trying to download model from https://huggingface.co/Qwen/Qwen2-57B-A14B-Instruct-GGUF/resolve/main/qwen2-57b-a14b-instruct-q3_k_m.gguf to /home/phymbert/.cache/llama.cpp/Qwen_Qwen2-57B-A14B-Instruct-GGUF_qwen2-57b-a14b-instruct-q3_k_m.gguf (server_etag:"0d637675f1adf9b07557cf8312b98396-1000", server_last_modified:Mon, 17 Jun 2024 04:40:23 GMT)...
curl_perform_with_retry: Trying to download from https://huggingface.co/Qwen/Qwen2-57B-A14B-Instruct-GGUF/resolve/main/qwen2-57b-a14b-instruct-q3_k_m.gguf (attempt 1 of 3)...
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 1146 100 1146 0 0 10544 0 --:--:-- --:--:-- --:--:-- 0
100 25.6G 100 25.6G 0 0 39.9M 0 0:10:56 0:10:56 --:--:-- 37.8M
common_download_file: file metadata saved: /home/phymbert/.cache/llama.cpp/Qwen_Qwen2-57B-A14B-Instruct-GGUF_qwen2-57b-a14b-instruct-q3_k_m.gguf.json
llama_load_model_from_file: using device CUDA0 (NVIDIA GeForce RTX 3050 Laptop GPU) - 3765 MiB free
llama_model_loader: loaded meta data with 25 key-value pairs and 479 tensors from /home/phymbert/.cache/llama.cpp/Qwen_Qwen2-57B-A14B-Instruct-GGUF_qwen2-57b-a14b-instruct-q3_k_m.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 = qwen2moe
llama_model_loader: - kv 1: general.name str = Qwen2-MoE-A14.2B-Chat
llama_model_loader: - kv 2: qwen2moe.block_count u32 = 28
llama_model_loader: - kv 3: qwen2moe.context_length u32 = 32768
llama_model_loader: - kv 4: qwen2moe.embedding_length u32 = 3584
llama_model_loader: - kv 5: qwen2moe.attention.head_count u32 = 28
llama_model_loader: - kv 6: qwen2moe.attention.head_count_kv u32 = 4
llama_model_loader: - kv 7: qwen2moe.rope.freq_base f32 = 1000000,000000
llama_model_loader: - kv 8: qwen2moe.attention.layer_norm_rms_epsilon f32 = 0,000001
llama_model_loader: - kv 9: qwen2moe.expert_used_count u32 = 8
llama_model_loader: - kv 10: qwen2moe.expert_count u32 = 64
llama_model_loader: - kv 11: qwen2moe.expert_feed_forward_length u32 = 2560
llama_model_loader: - kv 12: qwen2moe.feed_forward_length u32 = 20480
llama_model_loader: - kv 13: general.file_type u32 = 12
llama_model_loader: - kv 14: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 15: tokenizer.ggml.pre str = qwen2
llama_model_loader: - kv 16: tokenizer.ggml.tokens arr[str,151936] = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv 17: tokenizer.ggml.token_type arr[i32,151936] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 18: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv 19: tokenizer.ggml.eos_token_id u32 = 151643
llama_model_loader: - kv 20: tokenizer.ggml.padding_token_id u32 = 151643
llama_model_loader: - kv 21: tokenizer.ggml.bos_token_id u32 = 151643
llama_model_loader: - kv 22: tokenizer.chat_template str = {% for message in messages %}{{'<|im_...
llama_model_loader: - kv 23: tokenizer.ggml.add_bos_token bool = false
llama_model_loader: - kv 24: general.quantization_version u32 = 2
llama_model_loader: - type f32: 197 tensors
llama_model_loader: - type q3_K: 169 tensors
llama_model_loader: - type q4_K: 108 tensors
llama_model_loader: - type q5_K: 4 tensors
llama_model_loader: - type q6_K: 1 tensors
llm_load_vocab: special tokens cache size = 293
llm_load_vocab: token to piece cache size = 0,9338 MB
llm_load_print_meta: format = GGUF V3 (latest)
llm_load_print_meta: arch = qwen2moe
llm_load_print_meta: vocab type = BPE
llm_load_print_meta: n_vocab = 151936
llm_load_print_meta: n_merges = 151387
llm_load_print_meta: vocab_only = 0
llm_load_print_meta: n_ctx_train = 32768
llm_load_print_meta: n_embd = 3584
llm_load_print_meta: n_layer = 28
llm_load_print_meta: n_head = 28
llm_load_print_meta: n_head_kv = 4
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 = 7
llm_load_print_meta: n_embd_k_gqa = 512
llm_load_print_meta: n_embd_v_gqa = 512
llm_load_print_meta: f_norm_eps = 0,0e+00
llm_load_print_meta: f_norm_rms_eps = 1,0e-06
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 = 20480
llm_load_print_meta: n_expert = 64
llm_load_print_meta: n_expert_used = 8
llm_load_print_meta: causal attn = 1
llm_load_print_meta: pooling type = 0
llm_load_print_meta: rope type = 2
llm_load_print_meta: rope scaling = linear
llm_load_print_meta: freq_base_train = 1000000,0
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: ssm_dt_b_c_rms = 0
llm_load_print_meta: model type = 57B.A14B
llm_load_print_meta: model ftype = Q3_K - Medium
llm_load_print_meta: model params = 57,41 B
llm_load_print_meta: model size = 25,61 GiB (3,83 BPW)
llm_load_print_meta: general.name = Qwen2-MoE-A14.2B-Chat
llm_load_print_meta: BOS token = 151643 '<|endoftext|>'
llm_load_print_meta: EOS token = 151643 '<|endoftext|>'
llm_load_print_meta: EOT token = 151645 '<|im_end|>'
llm_load_print_meta: PAD token = 151643 '<|endoftext|>'
llm_load_print_meta: LF token = 148848 'ÄĬ'
llm_load_print_meta: EOG token = 151643 '<|endoftext|>'
llm_load_print_meta: EOG token = 151645 '<|im_end|>'
llm_load_print_meta: max token length = 256
llm_load_print_meta: n_ff_exp = 2560
llm_load_print_meta: n_ff_shexp = 0
llm_load_tensors: offloading 0 repeating layers to GPU
llm_load_tensors: offloaded 0/29 layers to GPU
llm_load_tensors: CPU_Mapped model buffer size = 26225,29 MiB
................................................................................................
llama_new_context_with_model: n_seq_max = 1
llama_new_context_with_model: n_ctx = 4096
llama_new_context_with_model: n_ctx_per_seq = 4096
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 = 1000000,0
llama_new_context_with_model: freq_scale = 1
llama_new_context_with_model: n_ctx_per_seq (4096) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 4096, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 28
llama_kv_cache_init: CPU KV buffer size = 224,00 MiB
llama_new_context_with_model: KV self size = 224,00 MiB, K (f16): 112,00 MiB, V (f16): 112,00 MiB
llama_new_context_with_model: CPU output buffer size = 0,58 MiB
llama_new_context_with_model: CUDA0 compute buffer size = 806,64 MiB
llama_new_context_with_model: CUDA_Host compute buffer size = 15,01 MiB
llama_new_context_with_model: graph nodes = 1910
llama_new_context_with_model: graph splits = 536 (with bs=512), 1 (with bs=1)
common_init_from_params: setting dry_penalty_last_n to ctx_size = 4096
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
main: llama threadpool init, n_threads = 6
system_info: n_threads = 6 (n_threads_batch = 6) / 12 | CUDA : ARCHS = 860 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |
sampler seed: 1924651909
sampler params:
repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = 4096
top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, temp = 0,800
mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000
sampler chain: logits -> logit-bias -> penalties -> dry -> top-k -> typical -> top-p -> min-p -> xtc -> temp-ext -> dist
generate: n_ctx = 4096, n_batch = 2048, n_predict = 64, n_keep = 0
Beijing is the capital of China, and it is a city with a long history and rich culture. It is also the political, economic, and cultural center of China. Beijing is located in the north of China, and it is a megacity with a population of more than 20 million people. The city is famous for its ancient architecture
llama_perf_sampler_print: sampling time = 138,62 ms / 70 runs ( 1,98 ms per token, 504,97 tokens per second)
llama_perf_context_print: load time = 20535,79 ms
llama_perf_context_print: prompt eval time = 9858,11 ms / 6 tokens ( 1643,02 ms per token, 0,61 tokens per second)
llama_perf_context_print: eval time = 186457,86 ms / 63 runs ( 2959,65 ms per token, 0,34 tokens per second)
llama_perf_context_print: total time = 196604,93 ms / 69 tokens
Process finished with exit code 0
So if the issue still exists, may it be related to split on multiple H100 on your output ?:
ggml_gallocr_reserve_n: reallocating CUDA0 buffer from size 0.00 MiB to 1349.38 MiB
ggml_gallocr_reserve_n: reallocating CUDA1 buffer from size 0.00 MiB to 0.00 MiB
ggml_gallocr_reserve_n: reallocating CUDA2 buffer from size 0.00 MiB to 0.00 MiB
ggml_gallocr_reserve_n: reallocating CUDA3 buffer from size 0.00 MiB to 0.00 MiB
@tang-t21 Could you dump the stack trace with gdb ? just run gdb --args ./llama-cli ... and type bt to get the call stack when it crashes.
