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": 76985.55120453195,
-      "required_pages_per_second_upper_estimate": 0.005326039063418876
+      "required_token_positions_per_second": 76468.76119793548,
+      "required_pages_per_second_upper_estimate": 0.005290286331644511
     },
     {
       "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": 9623.193900566494,
-      "required_pages_per_second_upper_estimate": 0.0040290303310271764
+      "required_token_positions_per_second": 9558.595149741936,
+      "required_pages_per_second_upper_estimate": 0.004001984183032248
     },
     {
       "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": 19246.387801132987,
-      "required_pages_per_second_upper_estimate": 0.004008515017597661
+      "required_token_positions_per_second": 19117.19029948387,
+      "required_pages_per_second_upper_estimate": 0.003981606585171391
     },
     {
       "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": 28869.581701699477,
-      "required_pages_per_second_upper_estimate": 0.001730687341718015
+      "required_token_positions_per_second": 28675.78544922581,
+      "required_pages_per_second_upper_estimate": 0.0017190695522919625
     },
     {
       "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": 28869.581701699477,
-      "required_pages_per_second_upper_estimate": 0.0022444829934372976
+      "required_token_positions_per_second": 28675.78544922581,
+      "required_pages_per_second_upper_estimate": 0.002229416187227098
     },
     {
       "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": 28869.581701699477,
-      "required_pages_per_second_upper_estimate": 0.3703079963276443
+      "required_token_positions_per_second": 28675.78544922581,
+      "required_pages_per_second_upper_estimate": 0.367822186083116
     }
   ],
   "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": 0.0902693556717793,
-    "producer_p95_seconds": 0.3424191880003491,
-    "gpu_update_seconds": 0.9576862001562161,
-    "producer_batches_per_second": 11.077956550792438,
-    "gpu_batches_per_second": 1.044183365946885,
-    "producer_headroom": 10.609206114624072,
+    "producer_batch_seconds": 0.13901584924997223,
+    "producer_p95_seconds": 0.35109388600176317,
+    "gpu_update_seconds": 0.9641584203143926,
+    "producer_batches_per_second": 7.1934243857464315,
+    "gpu_batches_per_second": 1.0371739528799844,
+    "producer_headroom": 6.935600692412309,
     "sustainable": true,
-    "buffer_seconds": 60.33423060984161,
+    "buffer_seconds": 60.741980479806735,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276597.8815827699,
-    "upload_bytes_per_second": 5592785.726130952,
+    "generation_bytes_per_second": 277392.97438130545,
+    "upload_bytes_per_second": 5521855.13212207,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.35146651851962,
+    "estimated_full_shards_per_hour": 59.522074924272275,
     "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": 1.4130291219989886,
-      "bytes": 17178237,
-      "pages": 40,
-      "shards": 5
-    },
     {
       "seconds": 1.2973765460010327,
       "bytes": 5824332,
@@ -562,6 +556,12 @@
       "bytes": 5987324,
       "pages": 24,
       "shards": 2
+    },
+    {
+      "seconds": 1.2340771259987378,
+      "bytes": 10655993,
+      "pages": 48,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -331,3 +331,4 @@
 {"step": 678100, "loss": 0.5835674405097961, "lr": 0.0001, "grad_norm": 0.22842088470374425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99256445591, "tokens_per_second": 154111.1735606397, "data_wait_s": 1.3627999578602612e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 678200, "loss": 0.5442383885383606, "lr": 0.0001, "grad_norm": 0.21608107373354424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99271132121, "tokens_per_second": 143973.4307800373, "data_wait_s": 8.634000550955534e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 678300, "loss": 0.5567049384117126, "lr": 0.0001, "grad_norm": 0.19036875865297515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99285865302, "tokens_per_second": 152639.1467377215, "data_wait_s": 9.129998943535611e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 678400, "loss": 0.5780566334724426, "lr": 0.0001, "grad_norm": 0.21393389684835812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99300550481, "tokens_per_second": 153261.01173797937, "data_wait_s": 1.1306998203508556e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c2a1f3921f3bf4cca949a51e3a66b447107ad3805242da97004fbaeedc74ac51
-size 710157
+oid sha256:35294c660cb32101e415e9f95a8104e419f1188abe7b13d8b9ae68650b38dca5
+size 712275