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": 75164.91597165882,
-      "required_pages_per_second_upper_estimate": 0.006515104412685526
+      "required_token_positions_per_second": 75217.95158068182,
+      "required_pages_per_second_upper_estimate": 0.006519701404858118
     },
     {
       "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": 9395.614496457352,
-      "required_pages_per_second_upper_estimate": 0.0037480510995920504
+      "required_token_positions_per_second": 9402.243947585228,
+      "required_pages_per_second_upper_estimate": 0.003750695686766087
     },
     {
       "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": 18791.228992914705,
-      "required_pages_per_second_upper_estimate": 0.0043470201131758385
+      "required_token_positions_per_second": 18804.487895170456,
+      "required_pages_per_second_upper_estimate": 0.004350087326863997
     },
     {
       "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": 28186.84348937206,
-      "required_pages_per_second_upper_estimate": 0.00171643088356412
+      "required_token_positions_per_second": 28206.731842755682,
+      "required_pages_per_second_upper_estimate": 0.0017176419799390566
     },
     {
       "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": 28186.84348937206,
-      "required_pages_per_second_upper_estimate": 0.002169901237988537
+      "required_token_positions_per_second": 28206.731842755682,
+      "required_pages_per_second_upper_estimate": 0.0021714322984864943
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28186.84348937206,
-      "required_pages_per_second_upper_estimate": 0.34998626084125384
+      "required_token_positions_per_second": 28206.731842755682,
+      "required_pages_per_second_upper_estimate": 0.35023320762824145
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.39414949141402644,
-    "producer_p95_seconds": 0.19430762699994375,
-    "gpu_update_seconds": 0.9808831560165567,
-    "producer_batches_per_second": 2.537108436731611,
-    "gpu_batches_per_second": 1.019489420188515,
-    "producer_headroom": 2.488606930577535,
+    "producer_batch_seconds": 0.07465409509376286,
+    "producer_p95_seconds": 0.19084727699919313,
+    "gpu_update_seconds": 0.9801915427185804,
+    "producer_batches_per_second": 13.395112468298436,
+    "gpu_batches_per_second": 1.0202087616737443,
+    "producer_headroom": 13.129775955190336,
     "sustainable": true,
-    "buffer_seconds": 60.81475567302652,
+    "buffer_seconds": 60.77187564855199,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304884.42196117964,
-    "upload_bytes_per_second": 5834564.47957412,
+    "generation_bytes_per_second": 316990.6801276196,
+    "upload_bytes_per_second": 5737993.573265884,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.4210996067671,
+    "estimated_full_shards_per_hour": 68.01882079001847,
     "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": 0.8917956289997164,
-      "bytes": 6042604,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.600639215999763,
       "bytes": 2577832,
@@ -562,6 +556,12 @@
       "bytes": 10722524,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.196992418001173,
+      "bytes": 774765,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -108,3 +108,4 @@
 {"step": 365800, "loss": 0.6179000269621611, "lr": 0.0003, "grad_norm": 0.1617581522578019, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53542720019, "tokens_per_second": 149295.59500863997, "data_wait_s": 8.774999514571391e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 365900, "loss": 0.5644729752093554, "lr": 0.0003, "grad_norm": 0.1614524000850837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53557545575, "tokens_per_second": 151592.21904008932, "data_wait_s": 1.0149999070563354e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 366000, "loss": 0.6050304397940636, "lr": 0.0003, "grad_norm": 0.15895203433216606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53572241822, "tokens_per_second": 150119.314526114, "data_wait_s": 1.3007000234210864e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 366100, "loss": 0.6828296761959791, "lr": 0.0003, "grad_norm": 0.2705137944203427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53586887872, "tokens_per_second": 149607.94873409357, "data_wait_s": 1.0300000212737359e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4c0263cffdb14229a2775e46cf3e989eff89092619ae5046f977ef93e838a77
-size 234711
+oid sha256:ef0da326b2607ed0d9bbe1fe827eade3c9d2c5d022be70bdd0233a96f2634f35
+size 236829