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
7.4k stars 800 forks source link

Encountered an error in forward function: slice 712 exceeds buffer size 471 #1480

Open sleepwalker2017 opened 2 months ago

sleepwalker2017 commented 2 months ago

System Info

GPU A30 * 2

TensorRT-LLM version: v0.9.0

Model: vicuna 13B

Who can help?

@byshiue

Information

Tasks

Reproduction

  1. build engine
python convert_checkpoint.py --model_dir /data/weilong.yu/vicuna-13b/vicuna-13b-v1.5/ \
                              --output_dir ./tllm_checkpoint_2gpu_fp16 \
                              --dtype float16 --tp_size 2

trtllm-build --checkpoint_dir ./tllm_checkpoint_2gpu_fp16 \
            --output_dir ./tmp/llama/13B/trt_engines/fp16/2-gpu \
            --gemm_plugin float16 \
            --use_fused_mlp \
            --max_batch_size $1 \
            --max_input_len 2048 \
            --max_output_len 256 \
            --context_fmha enable \
            --paged_kv_cache enable \
            --use_paged_context_fmha enable \
            --remove_input_padding enable  --workers 2 \
            --use_fused_mlp
  1. run benchmark
mpirun -n 2 --allow-run-as-root ./gptManagerBenchmark --engine_dir ../../../examples/llama/tmp/llama/13B/trt_engines/fp16/2-gpu/ --dataset ../../../benchmarks/cpp/token-norm-dist.json --kv_cache_free_gpu_mem_fraction 0.85 --enable_kv_cache_reuse -enable_chunked_context

Expected behavior

No error message.

actual behavior

sh run.sh
[TensorRT-LLM][ERROR] Encountered an error in forward function: slice 712 exceeds buffer size 471
[TensorRT-LLM][WARNING] Step function failed, continuing.
[TensorRT-LLM][ERROR] Encountered an error in forward function: slice 712 exceeds buffer size 471
[TensorRT-LLM][WARNING] Step function failed, continuing.
[TensorRT-LLM][ERROR] Encountered an error in forward function: slice 1553 exceeds buffer size 927
[TensorRT-LLM][WARNING] Step function failed, continuing.
[TensorRT-LLM][ERROR] Encountered an error in forward function: slice 1553 exceeds buffer size 927
[TensorRT-LLM][WARNING] Step function failed, continuing.
[TensorRT-LLM][ERROR] Encountered an error in forward function: slice 884 exceeds buffer size 642
[TensorRT-LLM][WARNING] Step function failed, continuing.
[TensorRT-LLM][ERROR] Encountered an error in forward function: slice 884 exceeds buffer size 642
[TensorRT-LLM][WARNING] Step function failed, continuing.
[TensorRT-LLM][ERROR] Encountered an error in forward function: slice 1192 exceeds buffer size 951
[TensorRT-LLM][WARNING] Step function failed, continuing.
[TensorRT-LLM][ERROR] Encountered an error in forward function: slice 1192 exceeds buffer size 951
[TensorRT-LLM][WARNING] Step function failed, continuing.
[TensorRT-LLM][ERROR] Encountered an error in forward function: slice 1253 exceeds buffer size 1012
[TensorRT-LLM][ERROR] Encountered an error in forward function: slice 1253 exceeds buffer size 1012
[TensorRT-LLM][WARNING] Step function failed, continuing.
[TensorRT-LLM][WARNING] Step function failed, continuing.
[BENCHMARK] num_samples 200
[BENCHMARK] total_latency(ms) 71149.43
[BENCHMARK] seq_throughput(seq/sec) 2.81
[BENCHMARK] token_throughput(token/sec) 531.37
[BENCHMARK] avg_sequence_latency(ms) 22587.76
[BENCHMARK] p99_sequence_latency(ms) 50983.86
[BENCHMARK] p90_sequence_latency(ms) 45602.29
[BENCHMARK] p50_sequence_latency(ms) 14514.95
[TensorRT-LLM][INFO] Terminate signal received, worker thread exiting.
[TensorRT-LLM][INFO] Terminate signal received, worker thread exiting.

additional notes

no

Tushar-ml commented 2 months ago

I am getting the same issue when trying speculative decoding (medusa) with vicuna, after some inference, it is getting buffer size exceeds 2560

skyCreateXian commented 2 months ago

Encountered an issue while using speculative decoding: '[TensorRT LM] [ERROR] Encountered an error in forward function: slice 501760 excesses buffer size 250880', 0.9.0 dev20240222000 is normal

pcastonguay commented 1 month ago

Hi, thanks for reporting this issue. I haven't been able to reproduce on latest main on 2xA100. What --max_batch_size value did you use (it's not specified in the build cmd you shared)? Thanks.

pcastonguay commented 1 month ago

I also just tested on 2xA30 and cannot reproduce using latest main following the instructions shared above.

mpirun -n 2 --allow-run-as-root ./gptManagerBenchmark --engine_dir ../../../examples/llama/tmp/llama/13B/trt_engines/fp16/2-gpu/ --dataset ../../../benchmarks/cpp/token-norm-dist.json --kv_cache_free_gpu_mem_fraction 0.85 --enable_kv_cache_reuse
[BENCHMARK] num_samples 100
[BENCHMARK] num_error_samples 0

[BENCHMARK] num_samples 100
[BENCHMARK] total_latency(ms) 1506.20
[BENCHMARK] seq_throughput(seq/sec) 66.39
[BENCHMARK] token_throughput(token/sec) 995.88

[BENCHMARK] avg_sequence_latency(ms) 1116.72
[BENCHMARK] max_sequence_latency(ms) 1501.60
[BENCHMARK] min_sequence_latency(ms) 872.77
[BENCHMARK] p99_sequence_latency(ms) 1501.60
[BENCHMARK] p90_sequence_latency(ms) 1501.58
[BENCHMARK] p50_sequence_latency(ms) 900.98
sleepwalker2017 commented 1 month ago 
mpirun -n 2 --allow-run-as-root ./gptManagerBenchmark --engine_dir ../../../examples/llama/tmp/llama/13B/trt_engines/fp16/2-gpu/ --dataset ../../../benchmarks/cpp/token-norm-dist.json --kv_cache_free_gpu_mem_fraction 0.85 --enable_kv_cache_reuse -enable_chunked_context
mpirun -n 2 --allow-run-as-root ./gptManagerBenchmark --engine_dir ../../../examples/llama/tmp/llama/13B/trt_engines/fp16/2-gpu/ --dataset ../../../benchmarks/cpp/token-norm-dist.json --kv_cache_free_gpu_mem_fraction 0.85 --enable_kv_cache_reuse -enable_chunked_context

hi, this issue is reproduced by using --enable_kv_cache_reuse and -enable_chunked_context together.

I built it using max_batch = 24.