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": 75762.02234429102,
-      "required_pages_per_second_upper_estimate": 0.00656686007971302
+      "required_token_positions_per_second": 75832.88862808807,
+      "required_pages_per_second_upper_estimate": 0.006573002589583593
     },
     {
       "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": 9470.252793036378,
-      "required_pages_per_second_upper_estimate": 0.0037778254320394037
+      "required_token_positions_per_second": 9479.111078511009,
+      "required_pages_per_second_upper_estimate": 0.0037813591345584045
     },
     {
       "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": 18940.505586072755,
-      "required_pages_per_second_upper_estimate": 0.004381552625824643
+      "required_token_positions_per_second": 18958.222157022017,
+      "required_pages_per_second_upper_estimate": 0.004385651042712756
     },
     {
       "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": 28410.75837910913,
-      "required_pages_per_second_upper_estimate": 0.0017300661255584737
+      "required_token_positions_per_second": 28437.333235533024,
+      "required_pages_per_second_upper_estimate": 0.0017316843948871927
     },
     {
       "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": 28410.75837910913,
-      "required_pages_per_second_upper_estimate": 0.0021871388260365832
+      "required_token_positions_per_second": 28437.333235533024,
+      "required_pages_per_second_upper_estimate": 0.0021891846320480053
     },
     {
       "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": 28410.75837910913,
-      "required_pages_per_second_upper_estimate": 0.35276653437685945
+      "required_token_positions_per_second": 28437.333235533024,
+      "required_pages_per_second_upper_estimate": 0.35309650515332114
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.13053972392947344,
-    "producer_p95_seconds": 0.325967329001287,
-    "gpu_update_seconds": 0.9731524808690077,
-    "producer_batches_per_second": 7.6605034076850735,
-    "gpu_batches_per_second": 1.027588193688843,
-    "producer_headroom": 7.454837895894217,
+    "producer_batch_seconds": 0.08194227548437993,
+    "producer_p95_seconds": 0.3199367960005475,
+    "gpu_update_seconds": 0.9722430641088828,
+    "producer_batches_per_second": 12.203712846497933,
+    "gpu_batches_per_second": 1.028549379178712,
+    "producer_headroom": 11.864975171384087,
     "sustainable": true,
-    "buffer_seconds": 60.33545381387848,
+    "buffer_seconds": 60.27906997475073,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 163,
+    "upload_pages_per_commit": 172,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 693526.087508216,
-    "upload_bytes_per_second": 4883413.537551884,
+    "generation_bytes_per_second": 685715.9656808543,
+    "upload_bytes_per_second": 4847960.428993928,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.40727695910864,
+    "estimated_full_shards_per_hour": 73.56934179219829,
     "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": [
-    {
-      "seconds": 0.8497731899988139,
-      "bytes": 5387577,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.2923507409977901,
       "bytes": 7469632,
@@ -562,6 +556,12 @@
       "bytes": 6027863,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9187128839985235,
+      "bytes": 2707013,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -272,3 +272,4 @@
 {"step": 382200, "loss": 0.6282160775735974, "lr": 0.0003, "grad_norm": 0.1665576246673821, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55942868859, "tokens_per_second": 148819.82729323383, "data_wait_s": 1.0771000233944505e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 382300, "loss": 0.6177261415868998, "lr": 0.0003, "grad_norm": 0.15430029791062644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55957440697, "tokens_per_second": 148790.28635864134, "data_wait_s": 1.2992997653782368e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 382400, "loss": 0.5885235872119665, "lr": 0.0003, "grad_norm": 0.1462610040250813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55971996264, "tokens_per_second": 147671.78293035118, "data_wait_s": 1.0176001524087042e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 382500, "loss": 0.587795801460743, "lr": 0.0003, "grad_norm": 0.17074154889065904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55986664145, "tokens_per_second": 148913.49731703813, "data_wait_s": 1.3286000466905534e-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 sha256:664fa35f53c9b732a1f887495358b647e287174539f2f3a5fa9ad527a350095a
-size 584412
+oid sha256:578f68e04fc2526754f53a422af15ef45c8bf312dcd87957278c60221f747449
+size 586530