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": 76104.93254926016,
-      "required_pages_per_second_upper_estimate": 0.005265115821530379
+      "required_token_positions_per_second": 77244.17117457373,
+      "required_pages_per_second_upper_estimate": 0.0053439309930273715
     },
     {
       "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": 9513.11656865752,
-      "required_pages_per_second_upper_estimate": 0.003982943250832971
+      "required_token_positions_per_second": 9655.521396821716,
+      "required_pages_per_second_upper_estimate": 0.004042565178634351
     },
     {
       "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": 19026.23313731504,
-      "required_pages_per_second_upper_estimate": 0.0039626626069932955
+      "required_token_positions_per_second": 19311.04279364343,
+      "required_pages_per_second_upper_estimate": 0.004021980947470773
     },
     {
       "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": 28539.349705972563,
-      "required_pages_per_second_upper_estimate": 0.001710890437809248
+      "required_token_positions_per_second": 28966.564190465146,
+      "required_pages_per_second_upper_estimate": 0.001736501293835831
     },
     {
       "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": 28539.349705972563,
-      "required_pages_per_second_upper_estimate": 0.0022188089083065706
+      "required_token_positions_per_second": 28966.564190465146,
+      "required_pages_per_second_upper_estimate": 0.0022520229553579427
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28539.349705972563,
-      "required_pages_per_second_upper_estimate": 0.366072134861951
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28966.564190465146,
+      "required_pages_per_second_upper_estimate": 0.3715519835618469
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2383789302579089,
-    "producer_p95_seconds": 0.3850594230025308,
-    "gpu_update_seconds": 0.9687676938978738,
-    "producer_batches_per_second": 4.195001625848692,
-    "gpu_batches_per_second": 1.0322392110088456,
-    "producer_headroom": 4.063982050971269,
+    "producer_batch_seconds": 0.12210022925785324,
+    "producer_p95_seconds": 0.38375506599550135,
+    "gpu_update_seconds": 0.954479786356603,
+    "producer_batches_per_second": 8.189992812283618,
+    "gpu_batches_per_second": 1.0476911237870785,
+    "producer_headroom": 7.817182589730582,
     "sustainable": true,
-    "buffer_seconds": 60.06359702166817,
+    "buffer_seconds": 60.13222654046599,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 474090.1431688714,
-    "upload_bytes_per_second": 4601765.485301384,
+    "generation_bytes_per_second": 494987.90737296373,
+    "upload_bytes_per_second": 4517615.744659073,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.72870847034079,
+    "estimated_full_shards_per_hour": 106.21288219348607,
     "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": 1.4279748699991615,
-      "bytes": 13559996,
-      "pages": 51,
-      "shards": 4
-    },
     {
       "seconds": 1.198663317001774,
       "bytes": 9767212,
@@ -562,6 +556,12 @@
       "bytes": 5666402,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.5151443669965374,
+      "bytes": 5878935,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -436,3 +436,4 @@
 {"step": 803600, "loss": 0.5232123732566833, "lr": 3e-05, "grad_norm": 0.2649226164007285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117623837178, "tokens_per_second": 147991.02297028588, "data_wait_s": 1.6224999853875488e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 803700, "loss": 0.5388901233673096, "lr": 3e-05, "grad_norm": 0.2735688570848663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117638367567, "tokens_per_second": 150629.6751526873, "data_wait_s": 9.963005140889436e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 803800, "loss": 0.5099372267723083, "lr": 3e-05, "grad_norm": 0.2714258411883778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117652931556, "tokens_per_second": 151436.96065108583, "data_wait_s": 1.22519995784387e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 803900, "loss": 0.5207173824310303, "lr": 3e-05, "grad_norm": 0.2727557766442316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117667628222, "tokens_per_second": 152070.79059758803, "data_wait_s": 9.208997653331608e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c4fe8f262e43255962fca0f6cdc3fa5c3055600fd24affe749f478f943af1ff
-size 934113
+oid sha256:a4e8f5248edb9506c96f9dbf706a224ad6712934470ba18f9284065d7237060d
+size 936231