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": 75598.02752697378,
-      "required_pages_per_second_upper_estimate": 0.0052300469559069095
+      "required_token_positions_per_second": 76682.3503888323,
+      "required_pages_per_second_upper_estimate": 0.005305062927465956
     },
     {
       "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": 9449.753440871722,
-      "required_pages_per_second_upper_estimate": 0.003956414432401628
+      "required_token_positions_per_second": 9585.293798604038,
+      "required_pages_per_second_upper_estimate": 0.0040131623497796235
     },
     {
       "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": 18899.506881743444,
-      "required_pages_per_second_upper_estimate": 0.003936268869953831
+      "required_token_positions_per_second": 19170.587597208076,
+      "required_pages_per_second_upper_estimate": 0.003992727834105906
     },
     {
       "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": 28349.260322615166,
-      "required_pages_per_second_upper_estimate": 0.0016994948695266544
+      "required_token_positions_per_second": 28755.881395812114,
+      "required_pages_per_second_upper_estimate": 0.0017238711819903825
     },
     {
       "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": 28349.260322615166,
-      "required_pages_per_second_upper_estimate": 0.002204030294865367
+      "required_token_positions_per_second": 28755.881395812114,
+      "required_pages_per_second_upper_estimate": 0.002235643294769347
     },
     {
       "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": 28349.260322615166,
-      "required_pages_per_second_upper_estimate": 0.3636338723543203
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28755.881395812114,
+      "required_pages_per_second_upper_estimate": 0.3688495708855981
     }
   ],
   "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": 0.8275764796875507,
-    "producer_p95_seconds": 0.2345323079989612,
-    "gpu_update_seconds": 0.9752635407543333,
-    "producer_batches_per_second": 1.2083475359009084,
-    "gpu_batches_per_second": 1.0253638716223656,
-    "producer_headroom": 1.1784572963244937,
+    "producer_batch_seconds": 0.057418400093695254,
+    "producer_p95_seconds": 0.20218382500024745,
+    "gpu_update_seconds": 0.9614728764330812,
+    "producer_batches_per_second": 17.416019923373025,
+    "gpu_batches_per_second": 1.0400709416888063,
+    "producer_headroom": 16.745030771741312,
     "sustainable": true,
-    "buffer_seconds": 60.466339526768664,
+    "buffer_seconds": 60.57279121528412,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 258854.082937507,
+    "generation_bytes_per_second": 258923.8252490543,
     "upload_bytes_per_second": 5456007.611965329,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.54406038373859,
+    "estimated_full_shards_per_hour": 55.55902546027872,
     "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": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -100,3 +100,4 @@
 {"step": 770000, "loss": 0.5791223049163818, "lr": 3e-05, "grad_norm": 0.28168012182084734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112705486715, "tokens_per_second": 150574.57346352158, "data_wait_s": 1.4241999451769516e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 770100, "loss": 0.5264490842819214, "lr": 3e-05, "grad_norm": 0.24843234261534192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112720043214, "tokens_per_second": 102502.96052938142, "data_wait_s": 1.5718000213382766e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 770200, "loss": 0.541844367980957, "lr": 3e-05, "grad_norm": 0.2595089970675744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112734606935, "tokens_per_second": 151611.935820005, "data_wait_s": 1.6105001122923568e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 770300, "loss": 0.5459050536155701, "lr": 3e-05, "grad_norm": 0.24835216215139258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112749359352, "tokens_per_second": 153466.11905282698, "data_wait_s": 1.3890999980503693e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c1d4c553951d46b4094927d880bafa39b2169d5fe02c3d63a2fe79f8aa8066f5
-size 217767
+oid sha256:394d3aa9d525c6a93718d5d23561da4e9a81a81aac201a67dca23129acf6cf0f
+size 219885