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": 75041.30909793291,
-      "required_pages_per_second_upper_estimate": 0.006504390482149722
+      "required_token_positions_per_second": 75840.3346904527,
+      "required_pages_per_second_upper_estimate": 0.006573647995397487
     },
     {
       "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": 9380.163637241614,
-      "required_pages_per_second_upper_estimate": 0.003741887520839961
+      "required_token_positions_per_second": 9480.041836306587,
+      "required_pages_per_second_upper_estimate": 0.0037817304277591303
     },
     {
       "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": 18760.32727448323,
-      "required_pages_per_second_upper_estimate": 0.0043398715444683865
+      "required_token_positions_per_second": 18960.083672613175,
+      "required_pages_per_second_upper_estimate": 0.004386081671583214
     },
     {
       "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": 28140.490911724843,
-      "required_pages_per_second_upper_estimate": 0.0017136082547785844
+      "required_token_positions_per_second": 28440.125508919762,
+      "required_pages_per_second_upper_estimate": 0.001731854429686519
     },
     {
       "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": 28140.490911724843,
-      "required_pages_per_second_upper_estimate": 0.002166332888249107
+      "required_token_positions_per_second": 28440.125508919762,
+      "required_pages_per_second_upper_estimate": 0.0021893995889828237
     },
     {
       "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": 28140.490911724843,
-      "required_pages_per_second_upper_estimate": 0.3494107169589734
+      "required_token_positions_per_second": 28440.125508919762,
+      "required_pages_per_second_upper_estimate": 0.35313117584364656
     }
   ],
   "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": 0.23005064625010618,
+    "producer_batch_seconds": 0.17195995538278908,
     "producer_p95_seconds": 0.31466025700137834,
-    "gpu_update_seconds": 0.9824988514496865,
-    "producer_batches_per_second": 4.346868901697504,
-    "gpu_batches_per_second": 1.0178128946659737,
-    "producer_headroom": 4.270793703320158,
+    "gpu_update_seconds": 0.9721476085374052,
+    "producer_batches_per_second": 5.8153073939450834,
+    "gpu_batches_per_second": 1.028650372863128,
+    "producer_headroom": 5.653337175933603,
     "sustainable": true,
-    "buffer_seconds": 60.91492878988056,
+    "buffer_seconds": 60.273151729319125,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 157,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 723831.9004196002,
-    "upload_bytes_per_second": 4877485.287874559,
+    "generation_bytes_per_second": 692873.2188284385,
+    "upload_bytes_per_second": 4962461.077485358,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.65873794289114,
+    "estimated_full_shards_per_hour": 74.33723174877103,
     "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": 0.7857203480016324,
-      "bytes": 1089066,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.0898243610026839,
       "bytes": 13191576,
@@ -562,6 +556,12 @@
       "bytes": 6530808,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.5235254969993548,
+      "bytes": 11921244,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -269,3 +269,4 @@
 {"step": 381900, "loss": 0.6026112791150808, "lr": 0.0003, "grad_norm": 0.1552691653762093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55898769496, "tokens_per_second": 150092.91202039964, "data_wait_s": 1.5804002032382414e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 382000, "loss": 0.6758288592100143, "lr": 0.0003, "grad_norm": 0.1708545792673141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55913492207, "tokens_per_second": 148949.50972857204, "data_wait_s": 1.4430999726755545e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 382100, "loss": 0.6003127582371235, "lr": 0.0003, "grad_norm": 0.1837233994322293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55928187499, "tokens_per_second": 150168.98178720166, "data_wait_s": 1.3950000720797107e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 382200, "loss": 0.6282160775735974, "lr": 0.0003, "grad_norm": 0.1665576246673821, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55942868859, "tokens_per_second": 148819.82729323383, "data_wait_s": 1.0771000233944505e-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 sha256:172eb1449c84469c89c666d9d0082db160fba8cc2eee3305a4e204e66179bdfb
-size 578058
+oid sha256:6e50f32ef5a551f08e554b19549d3af3acedd6581eb99e8c0ba70e527329159c
+size 580176