Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75402.45433018998,
-      "required_pages_per_second_upper_estimate": 0.006535693635567542
+      "required_token_positions_per_second": 75898.72193351365,
+      "required_pages_per_second_upper_estimate": 0.006578708853644895
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9425.306791273748,
-      "required_pages_per_second_upper_estimate": 0.0037598957999336797
+      "required_token_positions_per_second": 9487.340241689206,
+      "required_pages_per_second_upper_estimate": 0.0037846418707871412
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18850.613582547496,
-      "required_pages_per_second_upper_estimate": 0.004360757692854303
+      "required_token_positions_per_second": 18974.68048337841,
+      "required_pages_per_second_upper_estimate": 0.004389458386858657
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28275.92037382124,
-      "required_pages_per_second_upper_estimate": 0.0017218551984767812
+      "required_token_positions_per_second": 28462.02072506762,
+      "required_pages_per_second_upper_estimate": 0.0017331877334746024
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28275.92037382124,
-      "required_pages_per_second_upper_estimate": 0.0021767586231340335
+      "required_token_positions_per_second": 28462.02072506762,
+      "required_pages_per_second_upper_estimate": 0.0021910851433317234
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28275.92037382124,
-      "required_pages_per_second_upper_estimate": 0.35109229762495797
+      "required_token_positions_per_second": 28462.02072506762,
+      "required_pages_per_second_upper_estimate": 0.3534030411496284
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.1862787745702903,
-    "producer_p95_seconds": 0.1637777689993527,
-    "gpu_update_seconds": 0.9777931057408624,
-    "producer_batches_per_second": 5.368298145115083,
-    "gpu_batches_per_second": 1.0227112403725855,
-    "producer_headroom": 5.249084915854987,
+    "producer_batch_seconds": 0.20269526366402602,
+    "producer_p95_seconds": 0.16424058900065575,
+    "gpu_update_seconds": 0.9713997564357516,
+    "producer_batches_per_second": 4.933514389648159,
+    "gpu_batches_per_second": 1.0294423005305129,
+    "producer_headroom": 4.792414676476498,
     "sustainable": true,
-    "buffer_seconds": 60.62317255593347,
+    "buffer_seconds": 60.2267848990166,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369141.59805309924,
-    "upload_bytes_per_second": 5031288.450927107,
+    "generation_bytes_per_second": 368846.0506655112,
+    "upload_bytes_per_second": 5270655.027290741,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.20919376559003,
+    "estimated_full_shards_per_hour": 79.1457761762047,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -244,6 +244,12 @@
       "bytes": 5758950,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.0318703240000104,
+      "bytes": 8301935,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -37,3 +37,4 @@
 {"step": 358700, "loss": 0.5217609908431768, "lr": 0.0003, "grad_norm": 0.17284499424465385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52503965450, "tokens_per_second": 148584.709432047, "data_wait_s": 1.101300040318165e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 358800, "loss": 0.5568342115730047, "lr": 0.0003, "grad_norm": 0.15941198218727748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52518633527, "tokens_per_second": 149875.30207616428, "data_wait_s": 1.5557999176962767e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 358900, "loss": 0.5688133426010609, "lr": 0.0003, "grad_norm": 0.15956965421208585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52533196374, "tokens_per_second": 147767.6676919024, "data_wait_s": 8.003999937500339e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 359000, "loss": 0.564917329698801, "lr": 0.0003, "grad_norm": 0.1610518875786081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52547888928, "tokens_per_second": 129261.42408392014, "data_wait_s": 1.4450999515247531e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c1e28ff20c951108abd3aa25b7466dfa2ccc4a569e12ef3666f197f79bdf6ab
-size 82767
+oid sha256:e890503b16b00f9a35eab7a97a4b4273d4dfffa9c72a652c2ee21d6a57ff4467
+size 84885