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": 76559.38956876173,
-      "required_pages_per_second_upper_estimate": 0.005296556212625063
+      "required_token_positions_per_second": 76268.78717966966,
+      "required_pages_per_second_upper_estimate": 0.005276451665057231
     },
     {
       "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": 9569.923696095217,
-      "required_pages_per_second_upper_estimate": 0.0040067272088234185
+      "required_token_positions_per_second": 9533.598397458707,
+      "required_pages_per_second_upper_estimate": 0.003991518564842806
     },
     {
       "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": 19139.847392190433,
-      "required_pages_per_second_upper_estimate": 0.003986325460074452
+      "required_token_positions_per_second": 19067.196794917414,
+      "required_pages_per_second_upper_estimate": 0.003971194256587573
     },
     {
       "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": 28709.77108828565,
-      "required_pages_per_second_upper_estimate": 0.001721106939460534
+      "required_token_positions_per_second": 28600.79519237612,
+      "required_pages_per_second_upper_estimate": 0.001714574000897316
     },
     {
       "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": 28709.77108828565,
-      "required_pages_per_second_upper_estimate": 0.002232058421176973
+      "required_token_positions_per_second": 28600.79519237612,
+      "required_pages_per_second_upper_estimate": 0.0022235860246043156
     },
     {
       "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": 28709.77108828565,
-      "required_pages_per_second_upper_estimate": 0.36825811737003955
+      "required_token_positions_per_second": 28600.79519237612,
+      "required_pages_per_second_upper_estimate": 0.36686029158651273
     }
   ],
   "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": 0.08250679651550286,
-    "producer_p95_seconds": 0.22898163399986515,
-    "gpu_update_seconds": 0.9630170827548367,
-    "producer_batches_per_second": 12.120213633698674,
-    "gpu_batches_per_second": 1.038403178829776,
-    "producer_headroom": 11.671972775889897,
+    "producer_batch_seconds": 0.08922082792959429,
+    "producer_p95_seconds": 0.21396848800031876,
+    "gpu_update_seconds": 0.9666864090327775,
+    "producer_batches_per_second": 11.208145263896423,
+    "gpu_batches_per_second": 1.0344616316686965,
+    "producer_headroom": 10.834761697073766,
     "sustainable": true,
-    "buffer_seconds": 60.67007621355471,
+    "buffer_seconds": 60.90124376906498,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309447.4574028314,
-    "upload_bytes_per_second": 4769630.996273566,
+    "generation_bytes_per_second": 310545.0696300857,
+    "upload_bytes_per_second": 4818087.501283067,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.40022079051691,
+    "estimated_full_shards_per_hour": 66.63574282338074,
     "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": 1.054843687000357,
-      "bytes": 1282245,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.0156906100000924,
       "bytes": 5808038,
@@ -562,6 +556,12 @@
       "bytes": 5168575,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.1543622389999655,
+      "bytes": 6027007,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -152,3 +152,4 @@
 {"step": 735200, "loss": 0.5171884894371033, "lr": 5e-05, "grad_norm": 0.2305606743511926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107613043877, "tokens_per_second": 145677.1785847433, "data_wait_s": 1.855300070019439e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 735300, "loss": 0.5019158124923706, "lr": 5e-05, "grad_norm": 0.2187215928319985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107627583056, "tokens_per_second": 150113.46539697645, "data_wait_s": 1.3545999536290765e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 735400, "loss": 0.5374253392219543, "lr": 5e-05, "grad_norm": 0.23699762739642305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107642136451, "tokens_per_second": 150306.35628847012, "data_wait_s": 9.063000106834807e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 735500, "loss": 0.5342980623245239, "lr": 5e-05, "grad_norm": 0.22605024981864008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107656763987, "tokens_per_second": 150737.127845136, "data_wait_s": 1.4728999303770252e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:745b24a7b9d231b233effff8ff0d33e0346592fd1f0a137dfb21df10a38882b3
-size 328686
+oid sha256:69cf652f8fd4be67fa44358e8318ccdd7571d2de21ad4d898e555829b20c95e9
+size 330804