Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75579.84380801786,
-      "required_pages_per_second_upper_estimate": 0.006551069305916141
+      "required_token_positions_per_second": 75617.16638653328,
+      "required_pages_per_second_upper_estimate": 0.0065543043324286475
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9447.480476002233,
-      "required_pages_per_second_upper_estimate": 0.0037687412142980025
+      "required_token_positions_per_second": 9452.14579831666,
+      "required_pages_per_second_upper_estimate": 0.003770602281122012
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18894.960952004465,
-      "required_pages_per_second_upper_estimate": 0.004371016676291127
+      "required_token_positions_per_second": 18904.29159663332,
+      "required_pages_per_second_upper_estimate": 0.004373175156712276
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28342.441428006696,
-      "required_pages_per_second_upper_estimate": 0.0017259059816676769
+      "required_token_positions_per_second": 28356.437394949982,
+      "required_pages_per_second_upper_estimate": 0.0017267582626233596
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28342.441428006696,
-      "required_pages_per_second_upper_estimate": 0.0021818795980272933
+      "required_token_positions_per_second": 28356.437394949982,
+      "required_pages_per_second_upper_estimate": 0.0021829570463059037
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28342.441428006696,
-      "required_pages_per_second_upper_estimate": 0.35191826648629443
+      "required_token_positions_per_second": 28356.437394949982,
+      "required_pages_per_second_upper_estimate": 0.35209204955424195
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07856492539061577,
-    "producer_p95_seconds": 0.17956991299979563,
-    "gpu_update_seconds": 0.9754981789493801,
-    "producer_batches_per_second": 12.728326222269226,
-    "gpu_batches_per_second": 1.0251172391495478,
-    "producer_headroom": 12.416459050897272,
+    "producer_batch_seconds": 0.07343232874220007,
+    "producer_p95_seconds": 0.1926166089997423,
+    "gpu_update_seconds": 0.9750166995563361,
+    "producer_batches_per_second": 13.617980215644724,
+    "gpu_batches_per_second": 1.025623459018735,
+    "producer_headroom": 13.277758124481402,
     "sustainable": true,
-    "buffer_seconds": 60.480887094861565,
+    "buffer_seconds": 60.45103537249284,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412061.67496601917,
+    "generation_bytes_per_second": 437201.71978761896,
     "upload_bytes_per_second": 5935143.882253128,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.41884314284736,
+    "estimated_full_shards_per_hour": 93.81331153127124,
     "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": [

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -63,3 +63,4 @@
 {"step": 88200, "loss": 0.6165347024798393, "lr": 0.0003, "grad_norm": 0.13900520417190065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12907718727, "tokens_per_second": 147939.0365733943, "data_wait_s": 9.487999705015682e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 88300, "loss": 0.6625667437911034, "lr": 0.0003, "grad_norm": 0.1337257325070388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12922422282, "tokens_per_second": 149414.3213065964, "data_wait_s": 7.421999725920614e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 88400, "loss": 0.6176491789519787, "lr": 0.0003, "grad_norm": 0.12655935267375543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12937137368, "tokens_per_second": 149196.44004689562, "data_wait_s": 1.203600004373584e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 88500, "loss": 0.6558070499449968, "lr": 0.0003, "grad_norm": 0.1680909194797863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12951962743, "tokens_per_second": 150960.81700967898, "data_wait_s": 7.649000508536119e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2216f6d1c78f2af079d007b78d86e7419037be27ec032b003d7309518a927bd5
-size 138618
+oid sha256:95a334cb85fc2c5ce3774349cdb1423f2a5a54d873699fa420843cc9f58cf589
+size 140736