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": 76174.72147637223,
-      "required_pages_per_second_upper_estimate": 0.005269943981440613
+      "required_token_positions_per_second": 76742.70439241717,
+      "required_pages_per_second_upper_estimate": 0.0053092383574745395
     },
     {
       "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": 9521.840184546529,
-      "required_pages_per_second_upper_estimate": 0.003986595646635884
+      "required_token_positions_per_second": 9592.838049052147,
+      "required_pages_per_second_upper_estimate": 0.0040163209699003825
     },
     {
       "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": 19043.680369093057,
-      "required_pages_per_second_upper_estimate": 0.003966296405258213
+      "required_token_positions_per_second": 19185.676098104293,
+      "required_pages_per_second_upper_estimate": 0.0039958703709320606
     },
     {
       "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": 28565.520553639584,
-      "required_pages_per_second_upper_estimate": 0.0017124593401663152
+      "required_token_positions_per_second": 28778.514147156442,
+      "required_pages_per_second_upper_estimate": 0.0017252279808752647
     },
     {
       "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": 28565.520553639584,
-      "required_pages_per_second_upper_estimate": 0.002220843576599288
+      "required_token_positions_per_second": 28778.514147156442,
+      "required_pages_per_second_upper_estimate": 0.002237402891635416
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28565.520553639584,
-      "required_pages_per_second_upper_estimate": 0.36640782639575664
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28778.514147156442,
+      "required_pages_per_second_upper_estimate": 0.3691398795186881
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0929625415781743,
-    "producer_p95_seconds": 0.19732235099945683,
-    "gpu_update_seconds": 0.9678801388577293,
-    "producer_batches_per_second": 10.757020871240675,
-    "gpu_batches_per_second": 1.0331857839134688,
-    "producer_headroom": 10.411506854551916,
+    "producer_batch_seconds": 0.1648581179296542,
+    "producer_p95_seconds": 0.19809514199914702,
+    "gpu_update_seconds": 0.9607167298014186,
+    "producer_batches_per_second": 6.065822008393333,
+    "gpu_batches_per_second": 1.0408895452530542,
+    "producer_headroom": 5.8275366834611155,
     "sustainable": true,
-    "buffer_seconds": 60.008568609179214,
+    "buffer_seconds": 60.525153977489374,
     "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": 337816.5946361142,
-    "upload_bytes_per_second": 4067706.800341664,
+    "generation_bytes_per_second": 375296.30858700874,
+    "upload_bytes_per_second": 4130574.7776954207,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.48757724106378,
+    "estimated_full_shards_per_hour": 80.52985137183853,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 7557748,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.7842456250000396,
+      "bytes": 12131753,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -97,3 +97,4 @@
 {"step": 809700, "loss": 0.5788338780403137, "lr": 3e-05, "grad_norm": 0.3255125200327241, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118516943629, "tokens_per_second": 152159.41712562114, "data_wait_s": 1.7636999473324977e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 809800, "loss": 0.5648759603500366, "lr": 3e-05, "grad_norm": 0.34168568017784234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118531461221, "tokens_per_second": 150707.84311820107, "data_wait_s": 7.571001333417371e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 809900, "loss": 0.5182192325592041, "lr": 3e-05, "grad_norm": 0.26849332831787315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118546013215, "tokens_per_second": 150804.6009138714, "data_wait_s": 1.5571999028907157e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 810000, "loss": 0.5237100124359131, "lr": 3e-05, "grad_norm": 0.28498257341339567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118560875242, "tokens_per_second": 133640.02870263596, "data_wait_s": 1.4471999747911468e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd60f3e7ea7ab18f6511952e856a025bf89fa4c9aa954298829214207aea561d
-size 210630
+oid sha256:1dd5f35197d52e96eedc196895568978d24e029395fd338c5bc99f359bca08bb
+size 212748