$ .\pulsar_tracer_release.exe "請問台灣加值型營業稅法定稅率是多少？"
====================================================================================================
PULSAR · 18-Layer Latent Flight Telemetry Engine (v1.0.0-release)
Architected & Directed by Kuo-Ting Tsai (@tomtsai28) & Shin-Rung Tsai (@bella-tsai0123) · tom.tsai28@gmail.com
Co-Engineered with Antigravity (DeepMind Agentic Co-Pilot)
====================================================================================================
[*] Execution Mode : Standalone Native Binary (Zero-Python, Zero-CUDA, Zero-Runtime)
[*] Pipeline Profile: Standard Baseline Forward Inference (Full Precision FP16)
[*] Input Prompt : "請問台灣加值型營業稅法定稅率是多少？"
✔ [1/5] Embedded 256k Vocabulary Mapped (Total Tokens: 256,000, String Pool: 2.09 MB)
✔ [2/5] Embedded 4D Mental Radar Weights & Baseline Vector Loaded!
✔ [3/5] Gemma-2B Model Core (4.3 GB) Mapped Successfully!
✔ [4/5] 4-Core SMP Micro-Kernel Thread Pool & RoPE Table Initialized!
✔ [5/5] Prompt Tokenization Complete (Prefill Length: 23 Tokens)
====================================================================================================
[PHASE 1] 18-Layer Latent Flight Telemetry (Layer 00 ~ 17 State Inspection)
----------------------------------------------------------------------------------------------------
Layer | Truth (T) | Safety (S) | Cert (C) | Domain (D) | State Hash (32b) | Station Status & Flight Notes
----------------------------------------------------------------------------------------------------
Layer 00 | +0.1340 | +0.0848 | -0.1882 | -0.1810 | [0xCF00ACCF] | [TAKEOFF] Token Embedding Lookup
Layer 01 | +0.1141 | +0.0435 | -0.2263 | -0.2416 | [0x48D47FC8] |
Layer 02 | +0.1228 | +0.0222 | -0.2459 | -0.2503 | [0xE71EDF10] |
Layer 03 | +0.1288 | +0.0051 | -0.2467 | -0.2561 | [0x4A24A502] |
Layer 04 | +0.1158 | +0.0042 | -0.2518 | -0.2495 | [0x77210D02] |
Layer 05 | +0.1231 | -0.0022 | -0.2342 | -0.2374 | [0x02FC188A] |
Layer 06 | +0.1312 | +0.0022 | -0.2209 | -0.1796 | [0x6CF0CF20] |
Layer 07 | +0.1325 | +0.0034 | -0.2146 | -0.1876 | [0x2639622E] |
Layer 08 | +0.1216 | -0.0010 | -0.2009 | -0.1809 | [0x45B461FC] | [Station 08 Ingress] Bypass Sampling Point
Layer 09 | +0.0977 | +0.0122 | -0.1928 | -0.1554 | [0x85234C6E] | [Station 09 Socket] Core Extension Bus (1024 Slots)
Layer 10 | +0.0926 | +0.0315 | -0.1518 | -0.1313 | [0xCB121B9F] | [Station 10 Egress] Bypass Injection Point
Layer 11 | +0.0810 | +0.0389 | -0.1666 | -0.1218 | [0x3A5575A9] |
Layer 12 | +0.0438 | +0.0425 | -0.1819 | -0.0908 | [0x5F2F4B8D] |
Layer 13 | +0.0496 | +0.0723 | -0.1945 | -0.0343 | [0xBACC1A7E] |
Layer 14 | +0.0163 | +0.1404 | -0.2400 | +0.0348 | [0xDF28010E] | [Pre-Inflection] Attention Divergence Detected
Layer 15 | -0.0481 | +0.1701 | -0.2643 | +0.0754 | [0x72DE18A0] | 💥 [CRITICAL INFLECTION] Truth dropped below 0.0!
Layer 16 | -0.0682 | +0.2212 | -0.2606 | +0.1380 | [0x6C260FB8] | [Station 16 AOI] 4D Radar Final Audit
Layer 17 | -0.0817 | +0.2230 | -0.2464 | +0.1451 | [0xCD9607D3] | [LANDING] Final Norm & Vocabulary GEMV
----------------------------------------------------------------------------------------------------
====================================================================================================
[PHASE 2] 256k Vocabulary Classifier (Hardware Register RAX Trace)
----------------------------------------------------------------------------------------------------
[*] Capturing Argmax Token IDs from 256k Classifier Register RAX:
• Token #01: ID = 40911 (Logit = +7.0205) -> '台灣'
• Token #02: ID = 235771 (Logit = +10.8510) -> '加'
• Token #03: ID = 236415 (Logit = +12.6297) -> '值'
• Token #04: ID = 235954 (Logit = +12.5892) -> '型'
• Token #05: ID = 140618 (Logit = +17.1419) -> '營業'
• Token #06: ID = 240821 (Logit = +14.3588) -> '稅'
• Token #07: ID = 235661 (Logit = +13.5265) -> '法'
• Token #08: ID = 235669 (Logit = +14.0666) -> '定'
• Token #09: ID = 240821 (Logit = +11.9035) -> '稅'
• Token #10: ID = 236603 (Logit = +17.8619) -> '率'
• Token #11: ID = 236072 (Logit = +11.5293) -> '為'
• Token #12: ID = 235248 (Logit = +10.9602) -> ' '
• Token #13: ID = 235274 (Logit = +15.5773) -> '1'
• Token #14: ID = 235308 (Logit = +17.7756) -> '5'
• Token #15: ID = 84060 (Logit = +17.1152) -> '%。'
• Token #16: ID = 109 (Logit = +2.8195) -> '\n\n'
• Token #17: ID = 125753 (Logit = +2.1799) -> '根據'
• Token #18: ID = 74732 (Logit = +8.3018) -> '中華'
• Token #19: ID = 147344 (Logit = +18.5383) -> '民國'
• Token #20: ID = 28857 (Logit = +5.2705) -> '政府'
====================================================================================================
[DONE] Standalone Native Inference Complete · Verified Silicon Trace
====================================================================================================
🛡️ SYSTEM SPECIFICATIONS & OPERATIONAL BOUNDARIES / 系統規格與文字使用範圍聲明
- • Target CPU: x86-64 v3 (Requires AVX2 + F16C / 4-Core SMP)
- • Baseline: Intel 4th Gen Core (Haswell+) / AMD Zen (1st Gen+)
- • System Memory: 8 GB+ RAM (Direct physical mapping for 4.3 GB weights)
- • Runtime: Pure Native PE64 (Zero-Python, Zero-CUDA, Zero-Runtime)
- • Target Model: Google Gemma-2-2B (FP16, Bring Your Own Model / 自備權重)
- • 支援語系: UTF-8 繁體中文、簡體中文、English & ASCII (相容 256k 詞表)
- • 輸入長度: 建議 1 ~ 256 Tokens (單句提問或短指令，上限 512 截斷保護)
- • 適用範疇: 單輪思維遙測、18 層航情軌跡 (T,S,C,D)、RAX 暫存器捕獲
- • 限制排除: 專為單輪航情設計；非萬字長文本批處理、非多輪長對話
- • 輸出聲明: 裸機生成未外掛業務料件時包含模型固有幻覺 (供底層遙測研發)