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": 75567.00649204037,
-      "required_pages_per_second_upper_estimate": 0.006549956599903092
+      "required_token_positions_per_second": 75411.63119095686,
+      "required_pages_per_second_upper_estimate": 0.006536489062600273
     },
     {
       "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": 9445.875811505046,
-      "required_pages_per_second_upper_estimate": 0.0037681010896382025
+      "required_token_positions_per_second": 9426.453898869608,
+      "required_pages_per_second_upper_estimate": 0.003760353398304455
     },
     {
       "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": 18891.751623010092,
-      "required_pages_per_second_upper_estimate": 0.004370274254510541
+      "required_token_positions_per_second": 18852.907797739215,
+      "required_pages_per_second_upper_estimate": 0.004361288419162099
     },
     {
       "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": 28337.627434515136,
-      "required_pages_per_second_upper_estimate": 0.0017256128347211128
+      "required_token_positions_per_second": 28279.36169660882,
+      "required_pages_per_second_upper_estimate": 0.0017220647569793194
     },
     {
       "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": 28337.627434515136,
-      "required_pages_per_second_upper_estimate": 0.002181509003482326
+      "required_token_positions_per_second": 28279.36169660882,
+      "required_pages_per_second_upper_estimate": 0.002177023545688412
     },
     {
       "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": 28337.627434515136,
-      "required_pages_per_second_upper_estimate": 0.3518584927985291
+      "required_token_positions_per_second": 28279.36169660882,
+      "required_pages_per_second_upper_estimate": 0.3511350273366132
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.12316156856283555,
-    "producer_p95_seconds": 0.3393745300018054,
-    "gpu_update_seconds": 0.9756638964885546,
-    "producer_batches_per_second": 8.119415915767686,
-    "gpu_batches_per_second": 1.0249431219080996,
-    "producer_headroom": 7.921820969589085,
+    "producer_batch_seconds": 0.30460129902351696,
+    "producer_p95_seconds": 0.30531534100009594,
+    "gpu_update_seconds": 0.9776741178467075,
+    "producer_batches_per_second": 3.282980089729671,
+    "gpu_batches_per_second": 1.0228357095127611,
+    "producer_headroom": 3.2096846631347606,
     "sustainable": true,
-    "buffer_seconds": 60.49116158229038,
+    "buffer_seconds": 60.61579530649587,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 503469.10088892095,
-    "upload_bytes_per_second": 5051464.40603483,
+    "generation_bytes_per_second": 455913.48611340876,
+    "upload_bytes_per_second": 4972521.479172338,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.03274888992998,
+    "estimated_full_shards_per_hour": 97.82842099715897,
     "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": 1.0170475559971237,
-      "bytes": 6061997,
-      "pages": 35,
-      "shards": 2
-    },
-    {
-      "seconds": 0.9019806840005913,
-      "bytes": 1136025,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.5857687079987954,
       "bytes": 7492427,
@@ -562,6 +550,18 @@
       "bytes": 1093266,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.209129931001371,
+      "bytes": 7450282,
+      "pages": 31,
+      "shards": 3
+    },
+    {
+      "seconds": 2.1295198230000096,
+      "bytes": 324019,
+      "pages": 10,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -298,3 +298,4 @@
 {"step": 384800, "loss": 0.6171527206897736, "lr": 0.0003, "grad_norm": 0.15566373443762968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56324230617, "tokens_per_second": 149848.2530638584, "data_wait_s": 1.133399928221479e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 384900, "loss": 0.5901133706793189, "lr": 0.0003, "grad_norm": 0.1846310036791288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56338883655, "tokens_per_second": 149288.21830866195, "data_wait_s": 1.2441996659617871e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 385000, "loss": 0.570076484233141, "lr": 0.0003, "grad_norm": 0.14780607451161634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56353426047, "tokens_per_second": 148358.1146194847, "data_wait_s": 1.2352000339888036e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 385100, "loss": 0.6046611033380032, "lr": 0.0003, "grad_norm": 0.1473527156192229, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56367925588, "tokens_per_second": 129382.63438104208, "data_wait_s": 1.2508000509114936e-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 sha256:197a3303a3ffe1c646338fdd285aae6222220ad4a4e1918fdc4b0e77bebf1fa1
-size 640263
+oid sha256:03f7b032b161f3758cadc70cddeb04ae4324d159b327efe58d697d124f87c07c
+size 642381