Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76434.50520467217,
-      "required_pages_per_second_upper_estimate": 0.0052879164225457
+      "required_token_positions_per_second": 76280.74427451317,
+      "required_pages_per_second_upper_estimate": 0.005277278884622788
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9554.313150584021,
-      "required_pages_per_second_upper_estimate": 0.00400019139940314
+      "required_token_positions_per_second": 9535.093034314146,
+      "required_pages_per_second_upper_estimate": 0.003992144338082611
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19108.626301168042,
-      "required_pages_per_second_upper_estimate": 0.003979822930170027
+      "required_token_positions_per_second": 19070.186068628293,
+      "required_pages_per_second_upper_estimate": 0.0039718168434691
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28662.939451752063,
-      "required_pages_per_second_upper_estimate": 0.0017182994543581138
+      "required_token_positions_per_second": 28605.27910294244,
+      "required_pages_per_second_upper_estimate": 0.0017148428044892318
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28662.939451752063,
-      "required_pages_per_second_upper_estimate": 0.0022284174674270857
+      "required_token_positions_per_second": 28605.27910294244,
+      "required_pages_per_second_upper_estimate": 0.0022239346289282085
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28662.939451752063,
-      "required_pages_per_second_upper_estimate": 0.3676574114204803
+      "required_token_positions_per_second": 28605.27910294244,
+      "required_pages_per_second_upper_estimate": 0.36691780637680943
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0784928742732518,
-    "producer_p95_seconds": 0.2847691209972254,
-    "gpu_update_seconds": 0.9645905314958887,
-    "producer_batches_per_second": 12.740009959614541,
-    "gpu_batches_per_second": 1.0367093262352454,
-    "producer_headroom": 12.288892978207507,
+    "producer_batch_seconds": 0.10023753603101682,
+    "producer_p95_seconds": 0.2932103279999865,
+    "gpu_update_seconds": 0.9665348798206982,
+    "producer_batches_per_second": 9.97630268655613,
+    "gpu_batches_per_second": 1.0346238101469343,
+    "producer_headroom": 9.642444518205439,
     "sustainable": true,
-    "buffer_seconds": 60.76920348424099,
+    "buffer_seconds": 60.891697428703985,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326941.97966876015,
-    "upload_bytes_per_second": 5254330.761793856,
+    "generation_bytes_per_second": 372435.8955552954,
+    "upload_bytes_per_second": 5181464.859854018,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.15413801714996,
+    "estimated_full_shards_per_hour": 79.91607332224032,
     "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": 1.3350182510002924,
-      "bytes": 7769901,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 2.9742520910003805,
       "bytes": 5474226,
@@ -562,6 +556,12 @@
       "bytes": 7624505,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.1959701189989573,
+      "bytes": 1136148,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -310,3 +310,4 @@
 {"step": 635400, "loss": 0.49283096194267273, "lr": 0.0001, "grad_norm": 0.18659582375256248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93005764105, "tokens_per_second": 149815.13955222492, "data_wait_s": 1.4457997167482972e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 635500, "loss": 0.5993461012840271, "lr": 0.0001, "grad_norm": 0.16563936719817438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93020257613, "tokens_per_second": 150486.24634771576, "data_wait_s": 1.4549001207342371e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 635600, "loss": 0.538130521774292, "lr": 0.0001, "grad_norm": 0.16519310237239834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93034955828, "tokens_per_second": 152800.79926206343, "data_wait_s": 1.4249002560973167e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 635700, "loss": 0.578747570514679, "lr": 0.0001, "grad_norm": 0.17352032128901876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93049641333, "tokens_per_second": 152073.30066789777, "data_wait_s": 7.675000233575702e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:510f9c3cf50c098bfa99eb6ef8f2609a894c79aa5e131334012b9608dc6df4a3
-size 666462
+oid sha256:1c0818dccb25d1c278395352d0a279b0e93efc75f37a87b7289f5925642edc0f
+size 668580