{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/empowering-1000-tokens-second-on-device-llm","title":"Fast On-device LLM Inference with NPUs","arxiv_id":"2407.05858","date":"2024-07-08","proceeding":null,"authors":["Daliang Xu","Hao Zhang","Liming Yang","Ruiqi Liu","Gang Huang","Mengwei Xu","Xuanzhe Liu"],"abstract":"On-device inference for Large Language Models (LLMs), driven by increasing privacy concerns and advancements of mobile-sized models, has gained significant interest. However, even mobile-sized LLMs (e.g., Gemma-2B) encounter unacceptably high inference latency, often bottlenecked by the prefill stage in tasks like screen UI understanding. We present llm.npu, the first LLM inference system utilizing on-device Neural Processing Unit (NPU) offloading to reduce prefill latency. llm.npu enhances NPU offloading efficiency by re-constructing the prompt and model in three levels: (1) At prompt level, it divides variable-length prompts into multiple fixed-sized chunks while maintaining data dependencies; (2) At tensor level, it identifies and extracts significant outliers to run on the CPU/GPU in parallel with minimal overhead; (3) At block level, it schedules Transformer blocks in an out-of-order manner to the CPU/GPU and NPU based on their hardware affinity and sensitivity to accuracy. Compared to competitive baselines, llm.npu achieves 22.4x faster prefill speed and 30.7$\\times$ energy savings on average, and up to 32.8x speedup in an end-to-end real-world application. For the first time, llm.npu achieves more than 1,000 tokens/sec prefilling for a billion-sized model.","url_abs":"https://arxiv.org/abs/2407.05858v2","url_pdf":"https://arxiv.org/pdf/2407.05858v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"empowering-1000-tokens-second-on-device-llm","repo_url":"https://github.com/ubiquitouslearning/mllm","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":null,"task_name":"CPU"},{"task_slug":null,"task_name":"GPU"}],"methods":[{"method_slug":"absolute-position-encodings","method_name":"Absolute Position Encodings"},{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"label-smoothing","method_name":"Label Smoothing"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"position-wise-feed-forward-layer","method_name":"Position-Wise Feed-Forward Layer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"speed","method_name":"SPEED"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"transformer","method_name":"Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/2407.05858","atlas_url":"https://app.syntology.ai/?focus=2407.05858","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2407.05858"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-25T09:33:49+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/ubiquitouslearning/mllm","reach":{"status":"ok","spdx":"MIT"}}],"summary":{"ran":7,"unverified":1},"by_repo_kind":{"official":{"samples":8,"ran":7,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"a28ecbe6e5033c8a","entry":"bench","repo":"ubiquitouslearning/mllm","repo_kind":"official","path":"mllm-kernel/benchmarks/bench_w4a16_vs_w8a8.py","file_url":"https://github.com/ubiquitouslearning/mllm/blob/HEAD/mllm-kernel/benchmarks/bench_w4a16_vs_w8a8.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"a28ecbe6e5033c8a"}},{"code_sha256_prefix":"7357a837ed37641c","entry":"bench_fn","repo":"ubiquitouslearning/mllm","repo_kind":"official","path":"mllm-kernel/benchmarks/bench_int8_scaled_mm.py","file_url":"https://github.com/ubiquitouslearning/mllm/blob/HEAD/mllm-kernel/benchmarks/bench_int8_scaled_mm.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"7357a837ed37641c"}},{"code_sha256_prefix":"3561fac4ee82308f","entry":"cache_once","repo":"ubiquitouslearning/mllm","repo_kind":"official","path":"mllm-kernel/mllm_kernel/jit_utils/compile.py","file_url":"https://github.com/ubiquitouslearning/mllm/blob/HEAD/mllm-kernel/mllm_kernel/jit_utils/compile.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"3561fac4ee82308f"}},{"code_sha256_prefix":"bbbbfb5c0137cbc7","entry":"define_ir","repo":"ubiquitouslearning/mllm","repo_kind":"official","path":"mllm/compile/ir/rtti_kind_gen.py","file_url":"https://github.com/ubiquitouslearning/mllm/blob/HEAD/mllm/compile/ir/rtti_kind_gen.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"bbbbfb5c0137cbc7"}},{"code_sha256_prefix":"bcfe58bee8070d9b","entry":"dump_cls_to_kinds","repo":"ubiquitouslearning/mllm","repo_kind":"official","path":"mllm/compile/ir/rtti_kind_gen.py","file_url":"https://github.com/ubiquitouslearning/mllm/blob/HEAD/mllm/compile/ir/rtti_kind_gen.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"bcfe58bee8070d9b"}},{"code_sha256_prefix":"0f9d587155d9c348","entry":"filter_files","repo":"ubiquitouslearning/mllm","repo_kind":"official","path":"task.py","file_url":"https://github.com/ubiquitouslearning/mllm/blob/HEAD/task.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"0f9d587155d9c348"}},{"code_sha256_prefix":"09b13e2db7924d88","entry":"wildcard_to_regex","repo":"ubiquitouslearning/mllm","repo_kind":"official","path":"task.py","file_url":"https://github.com/ubiquitouslearning/mllm/blob/HEAD/task.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"09b13e2db7924d88"}},{"code_sha256_prefix":"2e0d11defc022118","entry":"prepare_marlin_weights","repo":"ubiquitouslearning/mllm","repo_kind":"official","path":"mllm-kernel/benchmarks/bench_w4a16_vs_w8a8.py","file_url":"https://github.com/ubiquitouslearning/mllm/blob/HEAD/mllm-kernel/benchmarks/bench_w4a16_vs_w8a8.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"2e0d11defc022118"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}