Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76368.18148783939,
-      "required_pages_per_second_upper_estimate": 0.005283327993922983
+      "required_token_positions_per_second": 76522.33552442424,
+      "required_pages_per_second_upper_estimate": 0.005293992727860567
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9546.022685979924,
-      "required_pages_per_second_upper_estimate": 0.003996720354997992
+      "required_token_positions_per_second": 9565.29194055303,
+      "required_pages_per_second_upper_estimate": 0.004004787989499959
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19092.045371959848,
-      "required_pages_per_second_upper_estimate": 0.003976369559884468
+      "required_token_positions_per_second": 19130.58388110606,
+      "required_pages_per_second_upper_estimate": 0.003984396115010793
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28638.068057939774,
-      "required_pages_per_second_upper_estimate": 0.0017168084522754866
+      "required_token_positions_per_second": 28695.875821659087,
+      "required_pages_per_second_upper_estimate": 0.0017202739394431151
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28638.068057939774,
-      "required_pages_per_second_upper_estimate": 0.00222648382595588
+      "required_token_positions_per_second": 28695.875821659087,
+      "required_pages_per_second_upper_estimate": 0.0022309781253155763
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28638.068057939774,
-      "required_pages_per_second_upper_estimate": 0.3673383878854783
+      "required_token_positions_per_second": 28695.875821659087,
+      "required_pages_per_second_upper_estimate": 0.3680798838093289
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09895140872649222,
-    "producer_p95_seconds": 0.2897411169979023,
-    "gpu_update_seconds": 0.9654282524946622,
-    "producer_batches_per_second": 10.105970322909313,
-    "gpu_batches_per_second": 1.03580975325303,
-    "producer_headroom": 9.756589268609256,
+    "producer_batch_seconds": 0.07860808063270497,
+    "producer_p95_seconds": 0.27652679099992383,
+    "gpu_update_seconds": 0.963483399908353,
+    "producer_batches_per_second": 12.721338467383326,
+    "gpu_batches_per_second": 1.0379006011884797,
+    "producer_headroom": 12.256798437939404,
     "sustainable": true,
-    "buffer_seconds": 60.82197990716372,
+    "buffer_seconds": 60.699454194226234,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348429.82930139644,
-    "upload_bytes_per_second": 5807027.686056018,
+    "generation_bytes_per_second": 362104.6094353384,
+    "upload_bytes_per_second": 5864093.643543959,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.76493033677502,
+    "estimated_full_shards_per_hour": 77.69921982092966,
     "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": 0.9978380930006097,
-      "bytes": 912994,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 1.1661826020026638,
       "bytes": 7439707,
@@ -562,6 +556,12 @@
       "bytes": 8506674,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.141177229997993,
+      "bytes": 6001543,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -268,3 +268,4 @@
 {"step": 583900, "loss": 0.5762531161308289, "lr": 0.0003, "grad_norm": 0.17968149031744238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85465918528, "tokens_per_second": 150670.59452989485, "data_wait_s": 1.4969999028835446e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 584000, "loss": 0.591342568397522, "lr": 0.0003, "grad_norm": 0.1488606265249998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85480638056, "tokens_per_second": 152602.5453124756, "data_wait_s": 1.846899976953864e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 584100, "loss": 0.5973283648490906, "lr": 0.0003, "grad_norm": 0.1483917216884607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85495375524, "tokens_per_second": 152591.68297632172, "data_wait_s": 9.343999408883974e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 584200, "loss": 0.5949923992156982, "lr": 0.0003, "grad_norm": 0.17366341050922118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85510119447, "tokens_per_second": 153434.5968646849, "data_wait_s": 7.907998224254698e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f6854e9f6fa4e42166df56a49302f0ef7ee7cc7569b0a32eac6c125186d9956c
-size 575940
+oid sha256:944cb0d2f56d876eb543893a67c8dd2ce70e51cced3cfd264cf8e150d9572aac
+size 578058