File size: 3,573 Bytes
8488b6b 70c5281 8488b6b be6b51c 70c5281 be6b51c 70c5281 8488b6b 08aeec5 70c5281 8488b6b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 | {
"schema_version": 1,
"title": "Reproduction: RealtimeTool: Parallel Decoding for Real-Time LLM Function Calling",
"emoji": "🎯",
"space_id": "Alogotron/repro-realtimetool-parallel-decoding-for-real-time-llm-function-calling",
"paper": {
"arxiv_id": "2603.00030"
},
"tags": [
"icml2026-repro",
"paper-f0IWycligk"
],
"updated_at": "2026-07-29T12:47:04+00:00",
"root": {
"slug": "index",
"title": "Reproduction: RealtimeTool: Parallel Decoding for Real-Time LLM Function Calling",
"file": "pages/index.md",
"children": [
{
"slug": "executive-summary",
"title": "Executive summary",
"file": "pages/executive-summary/page.md",
"children": []
},
{
"slug": "claim-1-realtimetool-achieves-3-6x-end-to-end-speedup-up-to-9-6x-with-only-8-2-parallelization-overhead-on-function-calling-tasks",
"title": "Claim 1: RealtimeTool achieves 3-6x end-to-end speedup (up to 9.6x) with only 8.2% parallelization overhead on function calling tasks",
"file": "pages/claim-1-realtimetool-achieves-3-6x-end-to-end-speedup-up-to-9-6x-with-only-8-2-parallelization-overhead-on-function-calling-tasks/page.md",
"children": []
},
{
"slug": "claim-2-on-mobile-actions-rt-qwen-0-5b-outperforms-google-s-functiongemma-in-both-accuracy-and-latency-consistency",
"title": "Claim 2: On Mobile Actions, RT-Qwen-0.5B outperforms Google's FunctionGemma in both accuracy and latency consistency",
"file": "pages/claim-2-on-mobile-actions-rt-qwen-0-5b-outperforms-google-s-functiongemma-in-both-accuracy-and-latency-consistency/page.md",
"children": []
},
{
"slug": "claim-2-on-the-rtx-4090-qwen2-5-0-5b-attains-5-35x-speedup-under-transformers-and-3-07x-under-vllm-while-qwen3-4b-attains-2-83x-transformers-and-3-88x-vllm-table-3",
"title": "Claim 2: On the RTX 4090, Qwen2.5-0.5B attains 5.35x speedup under Transformers and 3.07x under vLLM, while Qwen3-4B attains 2.83x (Transformers) and 3.88x (vLLM) (Table 3)",
"file": "pages/claim-2-on-the-rtx-4090-qwen2-5-0-5b-attains-5-35x-speedup-under-transformers-and-3-07x-under-vllm-while-qwen3-4b-attains-2-83x-transformers-and-3-88x-vllm-table-3/page.md",
"children": []
},
{
"slug": "claim-3-achieves-61-2ms-p50-latency-on-consumer-grade-gpu-with-quantization-enabling-16-hz-real-time-control-at-4b-model-scale",
"title": "Claim 3: Achieves 61.2ms P50 latency on consumer-grade GPU with quantization, enabling 16 Hz real-time control at 4B model scale",
"file": "pages/claim-3-achieves-61-2ms-p50-latency-on-consumer-grade-gpu-with-quantization-enabling-16-hz-real-time-control-at-4b-model-scale/page.md",
"children": []
},
{
"slug": "claim-5-with-8-head-parallel-decoding-the-framework-demonstrates-93-0-average-tpot-time-per-output-token-efficiency-section-2-1-3-appendix-g",
"title": "Claim 5: With 8-head parallel decoding, the framework demonstrates 93.0% average TPOT (time-per-output-token) efficiency (Section 2.1.3, Appendix G)",
"file": "pages/claim-5-with-8-head-parallel-decoding-the-framework-demonstrates-93-0-average-tpot-time-per-output-token-efficiency-section-2-1-3-appendix-g/page.md",
"children": []
},
{
"slug": "conclusion",
"title": "Conclusion",
"file": "pages/conclusion/page.md",
"children": []
}
]
},
"agent_view_tokens": 7200,
"revision": "1785329224177308255"
} |