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": 76243.69188357251,
-      "required_pages_per_second_upper_estimate": 0.005274715514244121
+      "required_token_positions_per_second": 76556.67154490015,
+      "required_pages_per_second_upper_estimate": 0.005296368173427634
     },
     {
       "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": 9530.461485446564,
-      "required_pages_per_second_upper_estimate": 0.003990205205289493
+      "required_token_positions_per_second": 9569.583943112519,
+      "required_pages_per_second_upper_estimate": 0.004006584961344412
     },
     {
       "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": 19060.922970893127,
-      "required_pages_per_second_upper_estimate": 0.00396988758449518
+      "required_token_positions_per_second": 19139.167886225037,
+      "required_pages_per_second_upper_estimate": 0.003986183936901638
     },
     {
       "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": 28591.38445633969,
-      "required_pages_per_second_upper_estimate": 0.0017140098416412938
+      "required_token_positions_per_second": 28708.751829337554,
+      "required_pages_per_second_upper_estimate": 0.0017210458364429139
     },
     {
       "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": 28591.38445633969,
-      "required_pages_per_second_upper_estimate": 0.002222854381270936
+      "required_token_positions_per_second": 28708.751829337554,
+      "required_pages_per_second_upper_estimate": 0.0022319791782770045
     },
     {
       "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": 28591.38445633969,
-      "required_pages_per_second_upper_estimate": 0.3667395807690985
+      "required_token_positions_per_second": 28708.751829337554,
+      "required_pages_per_second_upper_estimate": 0.36824504341064834
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06752632381227386,
+    "producer_batch_seconds": 0.08236491332797868,
     "producer_p95_seconds": 0.21280864300206304,
-    "gpu_update_seconds": 0.9670045898693616,
-    "producer_batches_per_second": 14.80903954996934,
-    "gpu_batches_per_second": 1.0341212549312677,
-    "producer_headroom": 14.320409216377257,
+    "gpu_update_seconds": 0.9630512731572827,
+    "producer_batches_per_second": 12.141092117926242,
+    "gpu_batches_per_second": 1.0383663132717575,
+    "producer_headroom": 11.692494221688715,
     "sustainable": true,
-    "buffer_seconds": 60.92128916176978,
+    "buffer_seconds": 60.67223020890881,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340139.7682331972,
-    "upload_bytes_per_second": 4764400.786329702,
+    "generation_bytes_per_second": 363040.57472194306,
+    "upload_bytes_per_second": 4719917.870342578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.9860762142843,
+    "estimated_full_shards_per_hour": 77.90005618327827,
     "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": 1.144911826999305,
-      "bytes": 6031558,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1537458799994056,
       "bytes": 6763666,
@@ -562,6 +556,12 @@
       "bytes": 10636216,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 0.9501123180016293,
+      "bytes": 1078327,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -177,3 +177,4 @@
 {"step": 457900, "loss": 0.6169613003730774, "lr": 0.0003, "grad_norm": 0.1489924384099695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67022638589, "tokens_per_second": 154146.99362011062, "data_wait_s": 8.695998985785991e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 458000, "loss": 0.6119412779808044, "lr": 0.0003, "grad_norm": 0.16279702943387112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67037286837, "tokens_per_second": 151251.47856399947, "data_wait_s": 1.3622000551549718e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 458100, "loss": 0.6139224767684937, "lr": 0.0003, "grad_norm": 0.16996936809218305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67052086938, "tokens_per_second": 154054.12114908855, "data_wait_s": 1.1897998774657026e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 458200, "loss": 0.5873919725418091, "lr": 0.0003, "grad_norm": 0.15693635068716677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67066827930, "tokens_per_second": 153708.19222548825, "data_wait_s": 1.3896002201363444e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ae2f4a9d2486be87c99e5ef8ffc50c22bf501dde83abc2f267ea798e440898a
-size 381636
+oid sha256:b29446678fadbd0cb32b82a93b61ea392d8b5a50c0376ed3c78557fe130d8b11
+size 383754