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": 75510.52270869169,
-      "required_pages_per_second_upper_estimate": 0.006545060728719268
+      "required_token_positions_per_second": 75388.70894284581,
+      "required_pages_per_second_upper_estimate": 0.006534502220229915
     },
     {
       "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": 9438.81533858646,
-      "required_pages_per_second_upper_estimate": 0.0037652845614275015
+      "required_token_positions_per_second": 9423.588617855727,
+      "required_pages_per_second_upper_estimate": 0.003759210394868249
     },
     {
       "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": 18877.63067717292,
-      "required_pages_per_second_upper_estimate": 0.0043670076222112695
+      "required_token_positions_per_second": 18847.177235711453,
+      "required_pages_per_second_upper_estimate": 0.0043599627544914235
     },
     {
       "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": 28316.446015759386,
-      "required_pages_per_second_upper_estimate": 0.0017243229974492032
+      "required_token_positions_per_second": 28270.765853567184,
+      "required_pages_per_second_upper_estimate": 0.0017215413152369867
     },
     {
       "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": 28316.446015759386,
-      "required_pages_per_second_upper_estimate": 0.0021798783992325885
+      "required_token_positions_per_second": 28270.765853567184,
+      "required_pages_per_second_upper_estimate": 0.0021763618138962507
     },
     {
       "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": 28316.446015759386,
-      "required_pages_per_second_upper_estimate": 0.35159549046723104
+      "required_token_positions_per_second": 28270.765853567184,
+      "required_pages_per_second_upper_estimate": 0.35102829573447214
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.21555432824243326,
-    "producer_p95_seconds": 0.43756548300007125,
-    "gpu_update_seconds": 0.9763937177925731,
-    "producer_batches_per_second": 4.6392016720504134,
-    "gpu_batches_per_second": 1.0241770115653712,
-    "producer_headroom": 4.529687368162825,
+    "producer_batch_seconds": 0.12732536664026384,
+    "producer_p95_seconds": 0.41907933500624495,
+    "gpu_update_seconds": 0.97797138369746,
+    "producer_batches_per_second": 7.853894525395948,
+    "gpu_batches_per_second": 1.0225248066249704,
+    "producer_headroom": 7.68088409641538,
     "sustainable": true,
-    "buffer_seconds": 60.536410503139535,
+    "buffer_seconds": 60.634225789242514,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 462631.7599567949,
-    "upload_bytes_per_second": 4725540.998060371,
+    "generation_bytes_per_second": 498548.53900297923,
+    "upload_bytes_per_second": 4676279.879999559,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.27000617053876,
+    "estimated_full_shards_per_hour": 106.9769108540252,
     "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": 1.0273188179999124,
-      "bytes": 9379394,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.4798621720037772,
       "bytes": 13649971,
@@ -562,6 +556,12 @@
       "bytes": 7233527,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.148407889995724,
+      "bytes": 5487074,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -410,3 +410,4 @@
 {"step": 196000, "loss": 0.5848307311534882, "lr": 0.0003, "grad_norm": 0.17119020365577028, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28686768471, "tokens_per_second": 148656.61338342412, "data_wait_s": 1.133199839387089e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 196100, "loss": 0.5884687602519989, "lr": 0.0003, "grad_norm": 0.13519273574053875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28701200508, "tokens_per_second": 146497.09774357386, "data_wait_s": 9.642004442866892e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 196200, "loss": 0.5501316413283348, "lr": 0.0003, "grad_norm": 0.12685191393783454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28715614316, "tokens_per_second": 147621.3208553125, "data_wait_s": 1.270500069949776e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 196300, "loss": 0.6403248496353626, "lr": 0.0003, "grad_norm": 0.17848715324458939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28730226039, "tokens_per_second": 148628.79923139638, "data_wait_s": 1.3712997315451503e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dfe1b4b412700dbef7d21b83d4b6c3c96a7ed819a38a0db1ea18b4a254b5d1a7
-size 879045
+oid sha256:9844c0d41c9f61f4a4d4a8db4c80949f01e4b5447420f5b5a75018b2ce44ca1a
+size 881163