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": 77305.81581166395,
-      "required_pages_per_second_upper_estimate": 0.005348195712056538
+      "required_token_positions_per_second": 77602.28811995091,
+      "required_pages_per_second_upper_estimate": 0.005368706354254361
     },
     {
       "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": 9663.226976457994,
-      "required_pages_per_second_upper_estimate": 0.00404579134391726
+      "required_token_positions_per_second": 9700.286014993864,
+      "required_pages_per_second_upper_estimate": 0.004061307189471503
     },
     {
       "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": 19326.45395291599,
-      "required_pages_per_second_upper_estimate": 0.0040251906855276465
+      "required_token_positions_per_second": 19400.57202998773,
+      "required_pages_per_second_upper_estimate": 0.0040406275263565545
     },
     {
       "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": 28989.680929373986,
-      "required_pages_per_second_upper_estimate": 0.0017378871070361994
+      "required_token_positions_per_second": 29100.858044981593,
+      "required_pages_per_second_upper_estimate": 0.0017445520053592492
     },
     {
       "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": 28989.680929373986,
-      "required_pages_per_second_upper_estimate": 0.0022538201801283165
+      "required_token_positions_per_second": 29100.858044981593,
+      "required_pages_per_second_upper_estimate": 0.002262463711850357
     },
     {
       "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": 28989.680929373986,
-      "required_pages_per_second_upper_estimate": 0.37184850026774907
+      "required_token_positions_per_second": 29100.858044981593,
+      "required_pages_per_second_upper_estimate": 0.3732745609340772
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09668158811717831,
-    "producer_p95_seconds": 0.19336974900033965,
-    "gpu_update_seconds": 0.9537186720805017,
-    "producer_batches_per_second": 10.343231006797257,
-    "gpu_batches_per_second": 1.0485272326885846,
-    "producer_headroom": 9.864532540824552,
+    "producer_batch_seconds": 0.056567684632796045,
+    "producer_p95_seconds": 0.1767701959997794,
+    "gpu_update_seconds": 0.9500750787919762,
+    "producer_batches_per_second": 17.677937615644137,
+    "gpu_batches_per_second": 1.0525483957241606,
+    "producer_headroom": 16.795367973062742,
     "sustainable": true,
-    "buffer_seconds": 60.08427634107161,
+    "buffer_seconds": 60.804805042686475,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317768.88515306514,
-    "upload_bytes_per_second": 3238684.7304921937,
+    "generation_bytes_per_second": 350255.4186814308,
+    "upload_bytes_per_second": 3224443.4778573196,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.18580547279325,
+    "estimated_full_shards_per_hour": 75.15665932018464,
     "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": [
@@ -454,6 +454,12 @@
       "bytes": 6150655,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.157564569999522,
+      "bytes": 2379174,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -89,3 +89,4 @@
 {"step": 653900, "loss": 0.5059468746185303, "lr": 0.0001, "grad_norm": 0.17095308878730586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95713960871, "tokens_per_second": 152575.27706675188, "data_wait_s": 8.784998499322683e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 654000, "loss": 0.5774896740913391, "lr": 0.0001, "grad_norm": 0.19368160866251724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95728641982, "tokens_per_second": 153102.53756921992, "data_wait_s": 1.619199974811636e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 654100, "loss": 0.5593799948692322, "lr": 0.0001, "grad_norm": 0.19241948478255352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95743377997, "tokens_per_second": 153748.94903408128, "data_wait_s": 9.348001185571775e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 654200, "loss": 0.53191739320755, "lr": 0.0001, "grad_norm": 0.1870930580141497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95758071937, "tokens_per_second": 153250.9483127806, "data_wait_s": 1.4536000890075229e-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:ce0c158c5d59d26d4be4619acd617093f721ff54e82470cc8ffa4acb3da18ffb
-size 193686
+oid sha256:a1a308aa2a6b11c7d1e1507b16bc07c34234f4a2a8625db363af4c2d59db0ed2
+size 195804