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": 74047.01375955714,
-      "required_pages_per_second_upper_estimate": 0.0064182074822378065
+      "required_token_positions_per_second": 74567.9593323174,
+      "required_pages_per_second_upper_estimate": 0.006463361724160182
     },
     {
       "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": 9255.876719944643,
-      "required_pages_per_second_upper_estimate": 0.0036923076112751886
+      "required_token_positions_per_second": 9320.994916539676,
+      "required_pages_per_second_upper_estimate": 0.0037182842335007483
     },
     {
       "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": 18511.753439889286,
-      "required_pages_per_second_upper_estimate": 0.00428236836258515
+      "required_token_positions_per_second": 18641.98983307935,
+      "required_pages_per_second_upper_estimate": 0.004312496260067436
     },
     {
       "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": 27767.630159833927,
-      "required_pages_per_second_upper_estimate": 0.0016909029912376059
+      "required_token_positions_per_second": 27962.984749619027,
+      "required_pages_per_second_upper_estimate": 0.0017027990608091978
     },
     {
       "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": 27767.630159833927,
-      "required_pages_per_second_upper_estimate": 0.002137629035423923
+      "required_token_positions_per_second": 27962.984749619027,
+      "required_pages_per_second_upper_estimate": 0.002152667972521696
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27767.630159833927,
-      "required_pages_per_second_upper_estimate": 0.3447810343051508
+      "required_token_positions_per_second": 27962.984749619027,
+      "required_pages_per_second_upper_estimate": 0.34720668450052805
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08674345926573324,
-    "producer_p95_seconds": 0.25631679099933535,
-    "gpu_update_seconds": 0.9956917403773631,
-    "producer_batches_per_second": 11.528246722747841,
-    "gpu_batches_per_second": 1.00432690103566,
-    "producer_headroom": 11.47858004287243,
+    "producer_batch_seconds": 0.1811480467968778,
+    "producer_p95_seconds": 0.2273682649993134,
+    "gpu_update_seconds": 0.98873565349195,
+    "producer_batches_per_second": 5.520346576639079,
+    "gpu_batches_per_second": 1.0113926775759197,
+    "producer_headroom": 5.458163479955289,
     "sustainable": true,
-    "buffer_seconds": 60.73719616301915,
+    "buffer_seconds": 60.31287486300895,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 494499.78269078064,
-    "upload_bytes_per_second": 5214310.416741437,
+    "generation_bytes_per_second": 497381.47944348684,
+    "upload_bytes_per_second": 5181282.510661123,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.10814199965061,
+    "estimated_full_shards_per_hour": 106.72648704031424,
     "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": 1.6214569329995356,
-      "bytes": 11854578,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 0.9589299640001627,
       "bytes": 2653813,
@@ -562,6 +556,12 @@
       "bytes": 2657482,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9979908650002471,
+      "bytes": 5962542,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -123,3 +123,4 @@
 {"step": 207300, "loss": 0.6095955148339272, "lr": 0.0003, "grad_norm": 0.1524404884898676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30341201833, "tokens_per_second": 147164.4404327976, "data_wait_s": 2.0792000213987194e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 207400, "loss": 0.6428727321326733, "lr": 0.0003, "grad_norm": 0.15677997656923812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30355729642, "tokens_per_second": 146366.67231165955, "data_wait_s": 1.9226999938837253e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 207500, "loss": 0.6107757650315762, "lr": 0.0003, "grad_norm": 0.1461868884554233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30370276350, "tokens_per_second": 147719.77099585536, "data_wait_s": 2.2218000594875775e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 207600, "loss": 0.569626696407795, "lr": 0.0003, "grad_norm": 0.14743573959776018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30384928391, "tokens_per_second": 147426.8888508114, "data_wait_s": 1.8900000213761814e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd0e39245ae3bc21cd78856df5721a616c7bb7d778282a6d1394ed7c2058ccad
-size 266481
+oid sha256:8df76ded869b7871c5ed485d55e642acfd3ee1331d6e9151c2f883296aa814a3
+size 268599