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": 77237.52794641048,
-      "required_pages_per_second_upper_estimate": 0.005343471399088626
+      "required_token_positions_per_second": 77556.25235164243,
+      "required_pages_per_second_upper_estimate": 0.0053655214929850805
     },
     {
       "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": 9654.69099330131,
-      "required_pages_per_second_upper_estimate": 0.0040422175060211545
+      "required_token_positions_per_second": 9694.531543955303,
+      "required_pages_per_second_upper_estimate": 0.004058897912614678
     },
     {
       "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": 19309.38198660262,
-      "required_pages_per_second_upper_estimate": 0.004021635045162599
+      "required_token_positions_per_second": 19389.063087910607,
+      "required_pages_per_second_upper_estimate": 0.004038230517233103
     },
     {
       "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": 28964.072979903933,
-      "required_pages_per_second_upper_estimate": 0.0017363519495665444
+      "required_token_positions_per_second": 29083.59463186591,
+      "required_pages_per_second_upper_estimate": 0.0017435170901026697
     },
     {
       "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": 28964.072979903933,
-      "required_pages_per_second_upper_estimate": 0.002251829274694486
+      "required_token_positions_per_second": 29083.59463186591,
+      "required_pages_per_second_upper_estimate": 0.002261121557414342
     },
     {
       "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": 28964.072979903933,
-      "required_pages_per_second_upper_estimate": 0.37152002898762115
+      "required_token_positions_per_second": 29083.59463186591,
+      "required_pages_per_second_upper_estimate": 0.37305312440663807
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.060041337335945855,
-    "producer_p95_seconds": 0.13314628600005562,
-    "gpu_update_seconds": 0.9545618815137961,
-    "producer_batches_per_second": 16.65519197889876,
-    "gpu_batches_per_second": 1.0476010192384235,
-    "producer_headroom": 15.898411392351084,
+    "producer_batch_seconds": 0.06114402429689214,
+    "producer_p95_seconds": 0.13773508500025855,
+    "gpu_update_seconds": 0.9506390234756958,
+    "producer_batches_per_second": 16.354827990129994,
+    "gpu_batches_per_second": 1.0519239956548725,
+    "producer_headroom": 15.547537709650154,
     "sustainable": true,
-    "buffer_seconds": 60.13739853536915,
+    "buffer_seconds": 60.840897502444534,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 69,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350334.03834518243,
-    "upload_bytes_per_second": 5135795.84959118,
+    "generation_bytes_per_second": 368196.5733748329,
+    "upload_bytes_per_second": 4979562.453074225,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.17352926985363,
+    "estimated_full_shards_per_hour": 79.00641346868268,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 12411381,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 0.765915884000151,
+      "bytes": 1219645,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -15,3 +15,4 @@
 {"step": 558600, "loss": 0.5376635193824768, "lr": 0.0003, "grad_norm": 0.1672256278354582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81760870349, "tokens_per_second": 151810.27200362913, "data_wait_s": 1.2439000101949205e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 558700, "loss": 0.5934765338897705, "lr": 0.0003, "grad_norm": 0.15493456042378062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81775657961, "tokens_per_second": 153695.41540460917, "data_wait_s": 1.5023999822005862e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 558800, "loss": 0.5554922223091125, "lr": 0.0003, "grad_norm": 0.1730167379325697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81790350258, "tokens_per_second": 152810.90438993459, "data_wait_s": 1.1388000075385207e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 558900, "loss": 0.5704553127288818, "lr": 0.0003, "grad_norm": 0.14818424246611112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81805104411, "tokens_per_second": 153543.70651069615, "data_wait_s": 1.4848000319034327e-05, "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 sha256:936dc1aa6099daf2b39330f49bb42594e1765a20fc140c79723f467fddb32dd2
-size 36171
+oid sha256:97628490ef163fc4ead5564fa39d14d13d8e0e5d72427d6dde96849b12cf0477
+size 38289