search

NVIDIA / TensorRT-LLM

TensorRT-LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and build TensorRT engines that contain state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT-LLM also contains components to create Python and C++ runtimes that execute those TensorRT engines.
https://nvidia.github.io/TensorRT-LLM
Apache License 2.0
8.11k stars 896 forks source link

Assertion failed: Unsupported architecture (/tensorrt_llm/cpp/tensorrt_llm/kernels/contextFusedMultiHeadAttention/fmhaRunner.cpp:89) #1515

Closed Aoempty closed 3 months ago

Aoempty commented 4 months ago

System Info

Who can help?

No response

Information

Tasks

Reproduction

  1. huggingface-cli login --token xxxxx(with actuall token used from Hugging Face)
  2. python3 convert_checkpoint.py --model_dir meta-llama/Llama-2-7b-chat-hf --output_dir llama-2-7b-ckpt
  3. trtllm-build --checkpoint_dir llama-2-7b-ckpt \ --gemm_plugin float16 \ --output_dir ./llama-2-7b-engine

Expected behavior

I am following the quick start guide, and after it should compile the Llama2 Model into a TensorRT Engine.

actual behavior

[TensorRT-LLM] TensorRT-LLM version: 0.10.0.dev2024042300 [04/28/2024-15:57:09] [TRT-LLM] [I] Set bert_attention_plugin to float16. [04/28/2024-15:57:09] [TRT-LLM] [I] Set gpt_attention_plugin to float16. [04/28/2024-15:57:09] [TRT-LLM] [I] Set gemm_plugin to float16. [04/28/2024-15:57:09] [TRT-LLM] [I] Set nccl_plugin to float16. [04/28/2024-15:57:09] [TRT-LLM] [I] Set lookup_plugin to None. [04/28/2024-15:57:09] [TRT-LLM] [I] Set lora_plugin to None. [04/28/2024-15:57:09] [TRT-LLM] [I] Set moe_plugin to float16. [04/28/2024-15:57:09] [TRT-LLM] [I] Set mamba_conv1d_plugin to float16. [04/28/2024-15:57:09] [TRT-LLM] [I] Set context_fmha to True. [04/28/2024-15:57:09] [TRT-LLM] [I] Set context_fmha_fp32_acc to False. [04/28/2024-15:57:09] [TRT-LLM] [I] Set paged_kv_cache to True. [04/28/2024-15:57:09] [TRT-LLM] [I] Set remove_input_padding to True. [04/28/2024-15:57:09] [TRT-LLM] [I] Set use_custom_all_reduce to True. [04/28/2024-15:57:09] [TRT-LLM] [I] Set multi_block_mode to False. [04/28/2024-15:57:09] [TRT-LLM] [I] Set enable_xqa to True. [04/28/2024-15:57:09] [TRT-LLM] [I] Set attention_qk_half_accumulation to False. [04/28/2024-15:57:09] [TRT-LLM] [I] Set tokens_per_block to 128. [04/28/2024-15:57:09] [TRT-LLM] [I] Set use_paged_context_fmha to False. [04/28/2024-15:57:09] [TRT-LLM] [I] Set use_fp8_context_fmha to False. [04/28/2024-15:57:09] [TRT-LLM] [I] Set use_context_fmha_for_generation to False. [04/28/2024-15:57:09] [TRT-LLM] [I] Set multiple_profiles to False. [04/28/2024-15:57:09] [TRT-LLM] [I] Set paged_state to True. [04/28/2024-15:57:09] [TRT-LLM] [I] Set streamingllm to False. [04/28/2024-15:57:09] [TRT-LLM] [W] remove_input_padding is enabled, while max_num_tokens is not set, setting to max_batch_sizemax_input_len. It may not be optimal to set max_num_tokens=max_batch_sizemax_input_len when remove_input_padding is enabled, because the number of packed input tokens are very likely to be smaller, we strongly recommend to set max_num_tokens according to your workloads. [04/28/2024-15:57:09] [TRT-LLM] [W] remove_input_padding is enabled, while opt_num_tokens is not set, setting to max_batch_size*max_beam_width.

[04/28/2024-15:57:10] [TRT-LLM] [I] Compute capability: (7, 5) [04/28/2024-15:57:10] [TRT-LLM] [I] SM count: 68 [04/28/2024-15:57:10] [TRT-LLM] [I] SM clock: 2100 MHz [04/28/2024-15:57:10] [TRT-LLM] [I] int4 TFLOPS: 584 [04/28/2024-15:57:10] [TRT-LLM] [I] int8 TFLOPS: 292 [04/28/2024-15:57:10] [TRT-LLM] [I] fp8 TFLOPS: 0 [04/28/2024-15:57:10] [TRT-LLM] [I] float16 TFLOPS: 146 [04/28/2024-15:57:10] [TRT-LLM] [I] bfloat16 TFLOPS: 0 [04/28/2024-15:57:10] [TRT-LLM] [I] float32 TFLOPS: 18 [04/28/2024-15:57:10] [TRT-LLM] [I] Total Memory: 11 GiB [04/28/2024-15:57:10] [TRT-LLM] [I] Memory clock: 7000 MHz [04/28/2024-15:57:10] [TRT-LLM] [I] Memory bus width: 352 [04/28/2024-15:57:10] [TRT-LLM] [I] Memory bandwidth: 616 GB/s [04/28/2024-15:57:10] [TRT-LLM] [I] NVLink is active: False [04/28/2024-15:57:10] [TRT-LLM] [I] PCIe speed: 2500 Mbps [04/28/2024-15:57:10] [TRT-LLM] [I] PCIe link width: 16 [04/28/2024-15:57:10] [TRT-LLM] [I] PCIe bandwidth: 5 GB/s [04/28/2024-15:57:10] [TRT] [I] [MemUsageChange] Init CUDA: CPU +13, GPU +0, now: CPU 259, GPU 157 (MiB) [04/28/2024-15:57:12] [TRT] [I] [MemUsageChange] Init builder kernel library: CPU +978, GPU +180, now: CPU 1373, GPU 337 (MiB) [04/28/2024-15:57:12] [TRT-LLM] [I] Set nccl_plugin to None. [04/28/2024-15:57:12] [TRT-LLM] [I] Set use_custom_all_reduce to True. [04/28/2024-15:57:12] [TRT] [W] IElementWiseLayer with inputs LLaMAForCausalLM/transformer/vocab_embedding/GATHER_0_output_0 and LLaMAForCausalLM/transformer/layers/0/input_layernorm/SHUFFLE_0_output_0: first input has type Half but second input has type Float. [04/28/2024-15:57:12] [TRT] [W] IElementWiseLayer with inputs LLaMAForCausalLM/transformer/layers/0/input_layernorm/REDUCE_AVG_0_output_0 and LLaMAForCausalLM/transformer/layers/0/input_layernorm/SHUFFLE_1_output_0: first input has type Half but second input has type Float. terminate called after throwing an instance of 'tensorrt_llm::common::TllmException' what(): [TensorRT-LLM][ERROR] Assertion failed: Unsupported architecture (/home/jenkins/agent/workspace/LLM/main/L0_PostMerge/tensorrt_llm/cpp/tensorrt_llm/kernels/contextFusedMultiHeadAttention/fmhaRunner.cpp:89) 1 0x7f120b494f3b tensorrt_llm::common::throwRuntimeError(char const, int, std::string const&) + 82 2 0x7f120b497194 /usr/local/lib/python3.10/dist-packages/tensorrt_llm/libs/libtensorrt_llm.so(+0x691194) [0x7f120b497194] 3 0x7f11deca9eef tensorrt_llm::plugins::GPTAttentionPluginCommon::initialize() + 415 4 0x7f11decd1e6d tensorrt_llm::plugins::GPTAttentionPlugin tensorrt_llm::plugins::GPTAttentionPluginCommon::cloneImpl() const + 573 5 0x7f12dfbd1279 /usr/local/lib/python3.10/dist-packages/tensorrt_libs/libnvinfer.so.9(+0xae3279) [0x7f12dfbd1279] 6 0x7f12dfb1f02e /usr/local/lib/python3.10/dist-packages/tensorrt_libs/libnvinfer.so.9(+0xa3102e) [0x7f12dfb1f02e] 7 0x7f128b6dfcef /usr/local/lib/python3.10/dist-packages/tensorrt_bindings/tensorrt.so(+0xdfcef) [0x7f128b6dfcef] 8 0x7f128b643443 /usr/local/lib/python3.10/dist-packages/tensorrt_bindings/tensorrt.so(+0x43443) [0x7f128b643443] 9 0x5617f827010e /usr/bin/python3(+0x15a10e) [0x5617f827010e] 10 0x5617f8266a7b _PyObject_MakeTpCall + 603 11 0x5617f827eacb /usr/bin/python3(+0x168acb) [0x5617f827eacb] 12 0x5617f825ecfa _PyEval_EvalFrameDefault + 24906 13 0x5617f82709fc _PyFunction_Vectorcall + 124 14 0x5617f827f492 PyObject_Call + 290 15 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 16 0x5617f82709fc _PyFunction_Vectorcall + 124 17 0x5617f827f492 PyObject_Call + 290 18 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 19 0x5617f827e7f1 /usr/bin/python3(+0x1687f1) [0x5617f827e7f1] 20 0x5617f827f492 PyObject_Call + 290 21 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 22 0x5617f82709fc _PyFunction_Vectorcall + 124 23 0x5617f8265cbd _PyObject_FastCallDictTstate + 365 24 0x5617f827b86c _PyObject_Call_Prepend + 92 25 0x5617f8396700 /usr/bin/python3(+0x280700) [0x5617f8396700] 26 0x5617f8266a7b _PyObject_MakeTpCall + 603 27 0x5617f8260150 _PyEval_EvalFrameDefault + 30112 28 0x5617f827e7f1 /usr/bin/python3(+0x1687f1) [0x5617f827e7f1] 29 0x5617f827f492 PyObject_Call + 290 30 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 31 0x5617f82709fc _PyFunction_Vectorcall + 124 32 0x5617f8265cbd _PyObject_FastCallDictTstate + 365 33 0x5617f827b86c _PyObject_Call_Prepend + 92 34 0x5617f8396700 /usr/bin/python3(+0x280700) [0x5617f8396700] 35 0x5617f827f42b PyObject_Call + 187 36 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 37 0x5617f827e7f1 /usr/bin/python3(+0x1687f1) [0x5617f827e7f1] 38 0x5617f825a53c _PyEval_EvalFrameDefault + 6540 39 0x5617f827e7f1 /usr/bin/python3(+0x1687f1) [0x5617f827e7f1] 40 0x5617f827f492 PyObject_Call + 290 41 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 42 0x5617f827e7f1 /usr/bin/python3(+0x1687f1) [0x5617f827e7f1] 43 0x5617f827f492 PyObject_Call + 290 44 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 45 0x5617f82709fc _PyFunction_Vectorcall + 124 46 0x5617f8265cbd _PyObject_FastCallDictTstate + 365 47 0x5617f827b86c _PyObject_Call_Prepend + 92 48 0x5617f8396700 /usr/bin/python3(+0x280700) [0x5617f8396700] 49 0x5617f827f42b PyObject_Call + 187 50 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 51 0x5617f82709fc _PyFunction_Vectorcall + 124 52 0x5617f825926d _PyEval_EvalFrameDefault + 1725 53 0x5617f82709fc _PyFunction_Vectorcall + 124 54 0x5617f827f492 PyObject_Call + 290 55 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 56 0x5617f82709fc _PyFunction_Vectorcall + 124 57 0x5617f827f492 PyObject_Call + 290 58 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 59 0x5617f82709fc _PyFunction_Vectorcall + 124 60 0x5617f827f492 PyObject_Call + 290 61 0x5617f825b5d7 _PyEval_EvalFrameDefault + 10791 62 0x5617f82709fc _PyFunction_Vectorcall + 124 63 0x5617f825926d _PyEval_EvalFrameDefault + 1725 64 0x5617f82559c6 /usr/bin/python3(+0x13f9c6) [0x5617f82559c6] 65 0x5617f834b256 PyEval_EvalCode + 134 66 0x5617f8376108 /usr/bin/python3(+0x260108) [0x5617f8376108] 67 0x5617f836f9cb /usr/bin/python3(+0x2599cb) [0x5617f836f9cb] 68 0x5617f8375e55 /usr/bin/python3(+0x25fe55) [0x5617f8375e55] 69 0x5617f8375338 _PyRun_SimpleFileObject + 424 70 0x5617f8374f83 _PyRun_AnyFileObject + 67 71 0x5617f8367a5e Py_RunMain + 702 72 0x5617f833e02d Py_BytesMain + 45 73 0x7f148ebe2d90 /lib/x86_64-linux-gnu/libc.so.6(+0x29d90) [0x7f148ebe2d90] 74 0x7f148ebe2e40 libc_start_main + 128 75 0x5617f833df25 _start + 37 [e62a70965c65:02001] Process received signal [e62a70965c65:02001] Signal: Aborted (6) [e62a70965c65:02001] Signal code: (-6) [e62a70965c65:02001] [ 0] /lib/x86_64-linux-gnu/libc.so.6(+0x42520)[0x7f148ebfb520] [e62a70965c65:02001] [ 1] /lib/x86_64-linux-gnu/libc.so.6(pthread_kill+0x12c)[0x7f148ec4f9fc] [e62a70965c65:02001] [ 2] /lib/x86_64-linux-gnu/libc.so.6(raise+0x16)[0x7f148ebfb476] [e62a70965c65:02001] [ 3] /lib/x86_64-linux-gnu/libc.so.6(abort+0xd3)[0x7f148ebe17f3] [e62a70965c65:02001] [ 4] /lib/x86_64-linux-gnu/libstdc++.so.6(+0xa2b9e)[0x7f13ec476b9e] [e62a70965c65:02001] [ 5] /lib/x86_64-linux-gnu/libstdc++.so.6(+0xae20c)[0x7f13ec48220c] [e62a70965c65:02001] [ 6] /lib/x86_64-linux-gnu/libstdc++.so.6(+0xad1e9)[0x7f13ec4811e9] [e62a70965c65:02001] [ 7] /lib/x86_64-linux-gnu/libstdc++.so.6(gxx_personality_v0+0x99)[0x7f13ec481959] [e62a70965c65:02001] [ 8] /lib/x86_64-linux-gnu/libgcc_s.so.1(+0x16884)[0x7f148e8eb884] [e62a70965c65:02001] [ 9] /lib/x86_64-linux-gnu/libgcc_s.so.1(_Unwind_Resume+0x12d)[0x7f148e8ec2dd] [e62a70965c65:02001] [10] /usr/local/lib/python3.10/dist-packages/tensorrt_llm/libs/libtensorrt_llm.so(+0x691118)[0x7f120b497118] [e62a70965c65:02001] [11] /usr/local/lib/python3.10/dist-packages/tensorrt_llm/libs/libnvinfer_plugin_tensorrt_llm.so(_ZN12tensorrt_llm7plugins24GPTAttentionPluginCommon10initializeEv+0x19f)[0x7f11deca9eef] [e62a70965c65:02001] [12] /usr/local/lib/python3.10/dist-packages/tensorrt_llm/libs/libnvinfer_plugin_tensorrt_llm.so(_ZNK12tensorrt_llm7plugins24GPTAttentionPluginCommon9cloneImplINS0_18GPTAttentionPluginEEEPT_v+0x23d)[0x7f11decd1e6d] [e62a70965c65:02001] [13] /usr/local/lib/python3.10/dist-packages/tensorrt_libs/libnvinfer.so.9(+0xae3279)[0x7f12dfbd1279] [e62a70965c65:02001] [14] /usr/local/lib/python3.10/dist-packages/tensorrt_libs/libnvinfer.so.9(+0xa3102e)[0x7f12dfb1f02e] [e62a70965c65:02001] [15] /usr/local/lib/python3.10/dist-packages/tensorrt_bindings/tensorrt.so(+0xdfcef)[0x7f128b6dfcef] [e62a70965c65:02001] [16] /usr/local/lib/python3.10/dist-packages/tensorrt_bindings/tensorrt.so(+0x43443)[0x7f128b643443] [e62a70965c65:02001] [17] /usr/bin/python3(+0x15a10e)[0x5617f827010e] [e62a70965c65:02001] [18] /usr/bin/python3(_PyObject_MakeTpCall+0x25b)[0x5617f8266a7b] [e62a70965c65:02001] [19] /usr/bin/python3(+0x168acb)[0x5617f827eacb] [e62a70965c65:02001] [20] /usr/bin/python3(_PyEval_EvalFrameDefault+0x614a)[0x5617f825ecfa] [e62a70965c65:02001] [21] /usr/bin/python3(_PyFunction_Vectorcall+0x7c)[0x5617f82709fc] [e62a70965c65:02001] [22] /usr/bin/python3(PyObject_Call+0x122)[0x5617f827f492] [e62a70965c65:02001] [23] /usr/bin/python3(_PyEval_EvalFrameDefault+0x2a27)[0x5617f825b5d7] [e62a70965c65:02001] [24] /usr/bin/python3(_PyFunction_Vectorcall+0x7c)[0x5617f82709fc] [e62a70965c65:02001] [25] /usr/bin/python3(PyObject_Call+0x122)[0x5617f827f492] [e62a70965c65:02001] [26] /usr/bin/python3(_PyEval_EvalFrameDefault+0x2a27)[0x5617f825b5d7] [e62a70965c65:02001] [27] /usr/bin/python3(+0x1687f1)[0x5617f827e7f1] [e62a70965c65:02001] [28] /usr/bin/python3(PyObject_Call+0x122)[0x5617f827f492] [e62a70965c65:02001] [29] /usr/bin/python3(_PyEval_EvalFrameDefault+0x2a27)[0x5617f825b5d7] [e62a70965c65:02001] End of error message Aborted (core dumped)

additional notes

None

byshiue commented 4 months ago

Please set --context_fmha disable during building engine because the fused mha kernel is not supported on Turing GPU, and it will fallback to unfused case.

Aoempty commented 4 months ago

Thanks, now it can convert successfully! However, when I running the command: python3 ../run.py --engine_dir llama-2-7b-engine --max_output_len 100 --tokenizer_dir meta-llama/Llama-2-7b-chat-hf --input_text "How do I count to nine in French?" I got the new Error: "root@81b5288ae872:/TensorRT-LLM/examples/llama# python3 ../run.py --engine_dir llama-2-7b-engine --max_output_len 100 --tokenizer_dir meta-llama/Llama-2-7b-chat-hf --input_text "How do I count to nine in French?" [TensorRT-LLM] TensorRT-LLM version: 0.10.0.dev2024043000 [TensorRT-LLM][INFO] Engine version 0.10.0.dev2024043000 found in the config file, assuming engine(s) built by new builder API. [TensorRT-LLM][WARNING] [json.exception.out_of_range.403] key 'cross_attention' not found [TensorRT-LLM][WARNING] Optional value for parameter cross_attention will not be set. [TensorRT-LLM][WARNING] Parameter layer_types cannot be read from json: [TensorRT-LLM][WARNING] [json.exception.out_of_range.403] key 'layer_types' not found [TensorRT-LLM][WARNING] [json.exception.type_error.302] type must be string, but is null [TensorRT-LLM][WARNING] Optional value for parameter quant_algo will not be set. [TensorRT-LLM][WARNING] [json.exception.type_error.302] type must be string, but is null [TensorRT-LLM][WARNING] Optional value for parameter kv_cache_quant_algo will not be set. [TensorRT-LLM][WARNING] [json.exception.out_of_range.403] key 'num_medusa_heads' not found [TensorRT-LLM][WARNING] Optional value for parameter num_medusa_heads will not be set. [TensorRT-LLM][WARNING] [json.exception.out_of_range.403] key 'max_draft_len' not found [TensorRT-LLM][WARNING] Optional value for parameter max_draft_len will not be set. [TensorRT-LLM][INFO] MPI size: 1, rank: 0 Traceback (most recent call last): File "/TensorRT-LLM/examples/llama/../run.py", line 564, in main(args) File "/TensorRT-LLM/examples/llama/../run.py", line 413, in main runner = runner_cls.from_dir(*runner_kwargs) File "/usr/local/lib/python3.10/dist-packages/tensorrt_llm/runtime/model_runner_cpp.py", line 186, in from_dir session = GptSession(config=session_config, RuntimeError: [TensorRT-LLM][ERROR] Assertion failed: Error opening engine file: llama-2-7b-engine/rank0.engine (/home/jenkins/agent/workspace/LLM/main/L0_MergeRequest/tensorrt_llm/cpp/tensorrt_llm/runtime/utils/sessionUtils.cpp:42) 1 0x7f9794a0fef3 tensorrt_llm::common::throwRuntimeError(char const, int, std::string const&) + 82 2 0x7f979643eb27 tensorrt_llm::runtime::utils::loadEngine(std::string const&) + 791 3 0x7f9806cffc24 /usr/local/lib/python3.10/dist-packages/tensorrt_llm/bindings.cpython-310-x86_64-linux-gnu.so(+0xbfc24) [0x7f9806cffc24] 4 0x7f9806cacb79 /usr/local/lib/python3.10/dist-packages/tensorrt_llm/bindings.cpython-310-x86_64-linux-gnu.so(+0x6cb79) [0x7f9806cacb79] 5 0x7f9806c8f6fc /usr/local/lib/python3.10/dist-packages/tensorrt_llm/bindings.cpython-310-x86_64-linux-gnu.so(+0x4f6fc) [0x7f9806c8f6fc] 6 0x5652e292f10e python3(+0x15a10e) [0x5652e292f10e] 7 0x5652e2925a7b _PyObject_MakeTpCall + 603 8 0x5652e293dacb python3(+0x168acb) [0x5652e293dacb] 9 0x5652e293e635 _PyObject_Call + 277 10 0x5652e293a087 python3(+0x165087) [0x5652e293a087] 11 0x5652e2925e2b python3(+0x150e2b) [0x5652e2925e2b] 12 0x7f9806c8ed69 /usr/local/lib/python3.10/dist-packages/tensorrt_llm/bindings.cpython-310-x86_64-linux-gnu.so(+0x4ed69) [0x7f9806c8ed69] 13 0x5652e2925a7b _PyObject_MakeTpCall + 603 14 0x5652e291f150 _PyEval_EvalFrameDefault + 30112 15 0x5652e293d7f1 python3(+0x1687f1) [0x5652e293d7f1] 16 0x5652e293e492 PyObject_Call + 290 17 0x5652e291a5d7 _PyEval_EvalFrameDefault + 10791 18 0x5652e292f9fc _PyFunction_Vectorcall + 124 19 0x5652e291826d _PyEval_EvalFrameDefault + 1725 20 0x5652e29149c6 python3(+0x13f9c6) [0x5652e29149c6] 21 0x5652e2a0a256 PyEval_EvalCode + 134 22 0x5652e2a35108 python3(+0x260108) [0x5652e2a35108] 23 0x5652e2a2e9cb python3(+0x2599cb) [0x5652e2a2e9cb] 24 0x5652e2a34e55 python3(+0x25fe55) [0x5652e2a34e55] 25 0x5652e2a34338 _PyRun_SimpleFileObject + 424 26 0x5652e2a33f83 _PyRun_AnyFileObject + 67 27 0x5652e2a26a5e Py_RunMain + 702 28 0x5652e29fd02d Py_BytesMain + 45 29 0x7f9a0b6dfd90 /lib/x86_64-linux-gnu/libc.so.6(+0x29d90) [0x7f9a0b6dfd90] 30 0x7f9a0b6dfe40 __libc_start_main + 128 31 0x5652e29fcf25 _start + 37". And I already check the engin inside only contains config.json: "{ "version": "0.10.0.dev2024043000", "pretrained_config": { "architecture": "LlamaForCausalLM", "dtype": "float16", "logits_dtype": "float32", "vocab_size": 32000, "max_position_embeddings": 4096, "hidden_size": 4096, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 32, "head_size": 128, "hidden_act": "silu", "intermediate_size": 11008, "norm_epsilon": 1e-05, "position_embedding_type": "rope_gpt_neox", "use_parallel_embedding": false, "embedding_sharding_dim": 0, "share_embedding_table": false, "mapping": { "world_size": 1, "tp_size": 1, "pp_size": 1, "gpus_per_node": 8 }, "quantization": { "quant_algo": null, "kv_cache_quant_algo": null, "group_size": 128, "smoothquant_val": null, "has_zero_point": false, "pre_quant_scale": false, "exclude_modules": [ "lm_head" ] }, "kv_dtype": "float16", "rotary_scaling": null, "moe_normalization_mode": null, "rotary_base": 10000.0, "moe_num_experts": 0, "moe_top_k": 0, "moe_tp_mode": 2, "attn_bias": false, "disable_weight_only_quant_plugin": false, "mlp_bias": false }, "build_config": { "max_input_len": 1024, "max_output_len": 1024, "max_batch_size": 1, "max_beam_width": 1, "max_num_tokens": 1024, "opt_num_tokens": 1, "max_prompt_embedding_table_size": 0, "gather_context_logits": false, "gather_generation_logits": false, "strongly_typed": false, "builder_opt": null, "profiling_verbosity": "layer_names_only", "enable_debug_output": false, "max_draft_len": 0, "use_refit": false, "input_timing_cache": null, "output_timing_cache": "model.cache", "lora_config": { "lora_dir": [], "lora_ckpt_source": "hf", "max_lora_rank": 64, "lora_target_modules": [], "trtllm_modules_to_hf_modules": {} }, "auto_parallel_config": { "world_size": 1, "gpus_per_node": 8, "cluster_key": "NVIDIA-GeForce-RTX-2080-Ti", "cluster_info": { "inter_node_bw_per_device": 25, "intra_node_bw_per_device": 5, "inter_node_latency": 10, "intra_node_latency": 10, "intra_node_sharp": false, "inter_node_sharp": true, "memory_bw": 616, "memory_budget_per_device": 11, "math_throughput": { "int4": 584, "int8": 292, "fp8": 0, "float16": 146, "bfloat16": 0, "float32": 18 }, "memory_efficiency": 1.0, "math_efficiency": 1.0, "communication_efficiency": 1.0 }, "sharding_cost_model": "alpha_beta", "comm_cost_model": "alpha_beta", "enable_pipeline_parallelism": false, "enable_shard_unbalanced_shape": false, "enable_shard_dynamic_shape": false, "enable_reduce_scatter": true, "builder_flags": null, "debug_mode": false, "infer_shape": true, "validation_mode": false, "same_buffer_io": { "past_keyvalue(\d+)": "present_keyvalue\1" }, "same_spec_io": {}, "sharded_io_allowlist": [ "past_keyvalue\d+", "present_keyvalue\d*" ], "fast_reduce": true, "fill_weights": false, "parallel_config_cache": null, "profile_cache": null, "dump_path": null, "debug_outputs": [] }, "weight_sparsity": false, "max_encoder_input_len": 1024, "use_fused_mlp": false, "plugin_config": { "bert_attention_plugin": "float16", "gpt_attention_plugin": "float16", "gemm_plugin": "float16", "smooth_quant_gemm_plugin": null, "identity_plugin": null, "layernorm_quantization_plugin": null, "rmsnorm_quantization_plugin": null, "nccl_plugin": null, "lookup_plugin": null, "lora_plugin": null, "weight_only_groupwise_quant_matmul_plugin": null, "weight_only_quant_matmul_plugin": null, "quantize_per_token_plugin": false, "quantize_tensor_plugin": false, "moe_plugin": "float16", "mamba_conv1d_plugin": "float16", "context_fmha": false, "context_fmha_fp32_acc": false, "paged_kv_cache": true, "remove_input_padding": true, "use_custom_all_reduce": true, "multi_block_mode": false, "enable_xqa": true, "attention_qk_half_accumulation": false, "tokens_per_block": 128, "use_paged_context_fmha": false, "use_fp8_context_fmha": false, "use_context_fmha_for_generation": false, "multiple_profiles": false, "paged_state": true, "streamingllm": false } }"

byshiue commented 4 months ago

It looks the TRT-LLM cannot load your engine successfully. Could you try rebuild the repo, and try the end to end workflow again?

byshiue commented 4 months ago

Also, could you share the log of building engine? It might has some issue during building engine.

nv-guomingz commented 3 months ago

The contextFMHA doesn't support geforce 2080Ti whichi sm version is sm75. https://github.com/NVIDIA/TensorRT-LLM/blob/main/cpp/tensorrt_llm/kernels/contextFusedMultiHeadAttention/fmhaRunner.cpp#L91. You may try to build the engine by disabling the context_fmha --context_fmha disable or on other supported hardware.