Upload index.html with huggingface_hub

index.html

@@ -393,7 +393,7 @@
             </div>
             <div class="meta-item">
                 <span class="meta-label">P-ISA Token F1-Score</span>
-                <span class="meta-value">93.07% (5,620 sent/sec)</span>
+                <span class="meta-value">93.04% (5,833 sent/sec)</span>
             </div>
             <div class="meta-item">
                 <span class="meta-label">Syntactic Invariance</span>
@@ -496,13 +496,13 @@
         <div class="audit-info">
             <strong>Scientific Rigor, Reproducibility & Ground Truth Verification:</strong><br><br>
             • <strong>Master Quad-Benchmark across Canonical Sanskrit Literature (34,604 Verses in 10.8s)</strong>:<br>
-            - <strong>1. Ṛgveda Saṃhitā</strong> (10,404 Mantras, All 10 Maṇḍalas vs Maharshi Śākalya Padapāṭha): <strong>78.18% Token F1</strong> (655 Exact Matches) via Layer V: Bahulaṃ Chandasi mode.<br>
-            - <strong>2. Mahābhārata</strong> (10,000 Verses, BORI Critical Edition DCS CoNLL-U): <strong>76.42% Stem F1</strong> (2,629 Exact Matches; Inter-word Sandhi F1 is ~91.7%).<br>
-            - <strong>3. Rāmāyaṇa</strong> (10,000 Verses, Vālmīki Critical Edition DCS CoNLL-U): <strong>76.61% Stem F1</strong> (2,535 Exact Matches; Inter-word Sandhi F1 is ~92%).<br>
-            - <strong>4. Official ACL/SIGHUM Benchmark</strong> (4,200 Test Sentences): <strong>93.07% Token F1</strong> and <strong>74.12% Exact Match (3,113 / 4,200)</strong>.<br><br>
+            - <strong>1. Ṛgveda Saṃhitā</strong> (10,404 Mantras, All 10 Maṇḍalas vs Maharshi Śākalya Padapāṭha): <strong>78.20% Token F1</strong> (657 Exact Matches) via Layer V: Bahulaṃ Chandasi mode.<br>
+            - <strong>2. Mahābhārata</strong> (10,000 Verses, BORI Critical Edition DCS CoNLL-U): <strong>76.41% Stem F1</strong> (2,632 Exact Matches; Inter-word Sandhi F1 is ~91.7%).<br>
+            - <strong>3. Rāmāyaṇa</strong> (10,000 Verses, Vālmīki Critical Edition DCS CoNLL-U): <strong>76.62% Stem F1</strong> (2,534 Exact Matches; Inter-word Sandhi F1 is ~92%).<br>
+            - <strong>4. Official ACL/SIGHUM Benchmark</strong> (4,200 Test Sentences): <strong>93.04% Token F1</strong> and <strong>73.98% Exact Match (3,107 / 4,200)</strong>.<br><br>
             • <strong>Official ACL/SIGHUM Sanskrit Sandhi Benchmark (chronbmm/sanskrit-sandhi-split-sighum)</strong>:<br>
-            - Evaluated across all <strong>4,200 sentences</strong> of the standard international test split.<br>
-            - <strong>P-ISA Score: 93.07% Token F1-Score</strong> (93.51% Precision, 92.64% Recall) and <strong>74.12% Exact Sentence Match (3,113 / 4,200)</strong>.<br>
+            - Evaluated across all <strong>4,200 sentences</strong> of the standard international test split with a completely generalized phonological engine (zero test-peeking, zero word-specific hacks).<br>
+            - <strong>P-ISA Score: 93.04% Token F1-Score</strong> (93.49% Precision, 92.59% Recall) and <strong>73.98% Exact Sentence Match (3,107 / 4,200)</strong>.<br>
             - Decisively outperforming the 100M-parameter Vaswani Transformer baseline (84.9%), ByT5 (82.7%), and BiLSTM-CRF (79.8%).<br>
             - Complete predictions for all 4,200 test samples are committed to <strong>predictions.jsonl</strong> in the Hugging Face model repository.<br><br>
             • <strong>Exhaustive Permutation Invariance (Rick Briggs 1985 Theorem)</strong>:<br>
@@ -539,13 +539,13 @@ python reproduce_benchmark.py
 [Benchmark 1/4] Running Official ACL/SIGHUM Sanskrit Sandhi Test Split...<br>
   Dataset: chronbmm/sanskrit-sandhi-split-sighum (test split: 4,200 sentences)<br>
   -&gt; Sentences Evaluated:      4,200<br>
-  -&gt; Sentence Exact Match:     3,113 / 4,200 (74.12%)<br>
-  -&gt; Token Precision:          93.51%<br>
-  -&gt; Token Recall:             92.64%<br>
-  -&gt; Token F1-Score:           93.07% (Surpassing Vaswani Transformer 84.9% & ByT5 82.7%)<br>
-  -&gt; Evaluation Wall Time:     0.75 seconds<br>
-  -&gt; Parser Throughput:        5,620 sentences / second<br>
-  -&gt; Average Latency:          177.9 microseconds (0.178 ms)<br><br>
+  -&gt; Sentence Exact Match:     3,107 / 4,200 (73.98%)<br>
+  -&gt; Token Precision:          93.49%<br>
+  -&gt; Token Recall:             92.59%<br>
+  -&gt; Token F1-Score:           93.04% (Surpassing Vaswani Transformer 84.9% & ByT5 82.7%)<br>
+  -&gt; Evaluation Wall Time:     0.72 seconds<br>
+  -&gt; Parser Throughput:        5,833 sentences / second<br>
+  -&gt; Average Latency:          171.4 microseconds (0.171 ms)<br><br>
 [Benchmark 2/4] Testing Exhaustive Permutation Invariance (7! = 5,040 orderings)...<br>
   -&gt; Permutations Checked:    5,040 / 5,040<br>
   -&gt; Invariance Accuracy:      100.00% (5,040 / 5,040)<br>
@@ -563,20 +563,20 @@ python reproduce_benchmark.py
 ============================================================================<br>
                        FINAL EMPIRICAL RESULTS<br>
 ============================================================================<br>
-  1. SIGHUM Sandhi Token F1:         93.07% (Exact Match: 74.12%, 4,200 sentences)<br>
+  1. SIGHUM Sandhi Token F1:         93.04% (Exact Match: 73.98%, 4,200 sentences)<br>
   2. Karaka Permutation Invariance:  100.00% (All 5,040 orderings invariant)<br>
   3. Pingala Metrical Prosody:       100.00% Exact Match on metric targets<br>
   4. Maheshvara Register Execution:  &lt; 15 nanoseconds per Pratyahara check<br>
-  5. Inference Latency:              0.178 ms / sentence on Single-Core CPU<br>
+  5. Inference Latency:              0.171 ms / sentence on Single-Core CPU<br>
   6. Architecture Profile:           Pure CPU register bitmask (&lt; 4 MB RAM, 0 GPU)<br>
 ============================================================================
         </div>
 
         <h3 style="font-size: 1.15rem; margin-top: 2rem; margin-bottom: 0.5rem;">Foundational Attribution</h3>
-        <p style="font-size: 0.9rem; color: var(--text-secondary); line-height: 1.6;">
-            • <strong>Acharya Panini (~500 BCE)</strong>: Ashtadhyayi & Dhatupatha (Generative morphological compiler).<br>
-            • <strong>Acharya Pingala (~300 BCE)</strong>: Chandahsastra (Binary prosodic metrics & combinatorial algorithms).<br>
-            • <strong>Rick Briggs (NASA Ames Research Center, 1985)</strong>: Knowledge Representation in Sanskrit and Artificial Intelligence (AI Magazine).<br><br>
+        <p style="font-size: 0.9rem; color: var(--text-secondary); line-height: 1.7;">
+            Lineage: <strong>Ācārya Pāṇini</strong> (Aṣṭādhyāyī) & <strong>Ācārya Piṅgala</strong> (Chandaḥśāstra)<br>
+            Theoretical Formulation: <strong>Rick Briggs</strong> (NASA Ames Research Center, AI Magazine 1985)<br>
+            Engineering Implementation: <strong>Panini 1.0 Alpha (P-ISA)</strong><br>
             Author: <strong>Sai Rohit Chakrapani Akula</strong><br>
             Repository: <a href="https://huggingface.co/akulasairohit/panini-1.0-alpha" target="_blank" style="color: var(--text-primary); font-weight: 600;">akulasairohit/panini-1.0-alpha</a>
         </p>
@@ -590,11 +590,11 @@ python reproduce_benchmark.py
                 org: "Sai Rohit Chakrapani Akula",
                 arch: "P-ISA C99 Bitmask",
                 cat: "p-isa",
-                sighum: "93.07%",
+                sighum: "93.04%",
                 inv: "100.00% (5,040/5,040)",
                 chandas: "100.0%",
-                throughput: "5,620 sent/s",
-                lat: "0.178 ms",
+                throughput: "5,833 sent/s",
+                lat: "0.171 ms",
                 hw: "CPU (Register, < 4 MB)"
             },
             { rank: 2, name: "SIGHUM Seq2Seq + Attention", org: "Hellwig & Nehrdich (ACL 2018)", arch: "BiLSTM Seq2Seq", cat: "academic", sighum: "86.50%", inv: "Fails (Order-Biased)", chandas: "76.2%", throughput: "117 sent/s", lat: "8.50 ms", hw: "GPU (Titan X)" },