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": 75568.40123422566,
-      "required_pages_per_second_upper_estimate": 0.006550077492620771
+      "required_token_positions_per_second": 75493.3191841464,
+      "required_pages_per_second_upper_estimate": 0.006543569570813617
     },
     {
       "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": 9446.050154278208,
-      "required_pages_per_second_upper_estimate": 0.0037681706375770735
+      "required_token_positions_per_second": 9436.6648980183,
+      "required_pages_per_second_upper_estimate": 0.0037644267185329107
     },
     {
       "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": 18892.100308556415,
-      "required_pages_per_second_upper_estimate": 0.0043703549167750265
+      "required_token_positions_per_second": 18873.3297960366,
+      "required_pages_per_second_upper_estimate": 0.004366012689053305
     },
     {
       "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": 28338.150462834623,
-      "required_pages_per_second_upper_estimate": 0.001725644684401652
+      "required_token_positions_per_second": 28309.9946940549,
+      "required_pages_per_second_upper_estimate": 0.0017239301458048693
     },
     {
       "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": 28338.150462834623,
-      "required_pages_per_second_upper_estimate": 0.002181549267650195
+      "required_token_positions_per_second": 28309.9946940549,
+      "required_pages_per_second_upper_estimate": 0.0021793817586293757
     },
     {
       "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": 28338.150462834623,
-      "required_pages_per_second_upper_estimate": 0.3518649870597939
+      "required_token_positions_per_second": 28309.9946940549,
+      "required_pages_per_second_upper_estimate": 0.35151538664284615
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1405299670782938,
-    "producer_p95_seconds": 0.4313378119986737,
-    "gpu_update_seconds": 0.9756458889672509,
-    "producer_batches_per_second": 7.115919976291375,
-    "gpu_batches_per_second": 1.0249620393097014,
-    "producer_headroom": 6.942618071088617,
+    "producer_batch_seconds": 0.16415525443778733,
+    "producer_p95_seconds": 0.43558402399503393,
+    "gpu_update_seconds": 0.9766162197764763,
+    "producer_batches_per_second": 6.09179403623042,
+    "gpu_batches_per_second": 1.0239436738301109,
+    "producer_headroom": 5.949344863320235,
     "sustainable": true,
-    "buffer_seconds": 60.49004511596956,
+    "buffer_seconds": 60.550205626141526,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300287.3471039793,
-    "upload_bytes_per_second": 5631322.788170599,
+    "generation_bytes_per_second": 298945.43300630536,
+    "upload_bytes_per_second": 5682623.188487608,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.4346743568376,
+    "estimated_full_shards_per_hour": 64.1467308296382,
     "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": 1.064266913002939,
-      "bytes": 6180437,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.110571836004965,
       "bytes": 7330551,
@@ -562,6 +556,12 @@
       "bytes": 854296,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.2072945199979586,
+      "bytes": 10797475,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -409,3 +409,4 @@
 {"step": 235900, "loss": 0.601128863170743, "lr": 0.0003, "grad_norm": 0.13912574180998558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34528941017, "tokens_per_second": 146718.09281255124, "data_wait_s": 7.586997526232153e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 236000, "loss": 0.5956513024866581, "lr": 0.0003, "grad_norm": 0.1497077140744594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34543587690, "tokens_per_second": 148646.84811544805, "data_wait_s": 1.3183998817112297e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 236100, "loss": 0.5857977233827114, "lr": 0.0003, "grad_norm": 0.15777404977066692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34558248187, "tokens_per_second": 147882.62711027206, "data_wait_s": 1.2192002031952143e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 236200, "loss": 0.5758872497826815, "lr": 0.0003, "grad_norm": 0.16433870787146654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34573002617, "tokens_per_second": 148960.42532426154, "data_wait_s": 1.4053002814762294e-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:cd9461e128144f709ba44d1d8be44fecb7ada4d76cbcb4a281d83d454f79348a
-size 876927
+oid sha256:61c15ccde00ef59e2f13cff2c0bcf151c994703dc38c1536591d354e18018e6c
+size 879045