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": 77201.10125740989,
-      "required_pages_per_second_upper_estimate": 0.005340951316222001
+      "required_token_positions_per_second": 76378.55489599323,
+      "required_pages_per_second_upper_estimate": 0.005284045650368689
     },
     {
       "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": 9650.137657176236,
-      "required_pages_per_second_upper_estimate": 0.004040311119270056
+      "required_token_positions_per_second": 9547.319361999154,
+      "required_pages_per_second_upper_estimate": 0.003997263246169569
     },
     {
       "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": 19300.27531435247,
-      "required_pages_per_second_upper_estimate": 0.00401973836549199
+      "required_token_positions_per_second": 19094.63872399831,
+      "required_pages_per_second_upper_estimate": 0.003976909686722787
     },
     {
       "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": 28950.41297152871,
-      "required_pages_per_second_upper_estimate": 0.001735533052922075
+      "required_token_positions_per_second": 28641.95808599746,
+      "required_pages_per_second_upper_estimate": 0.001717041653517816
     },
     {
       "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": 28950.41297152871,
-      "required_pages_per_second_upper_estimate": 0.0022507672691273388
+      "required_token_positions_per_second": 28641.95808599746,
+      "required_pages_per_second_upper_estimate": 0.00222678625852694
     },
     {
       "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": 28950.41297152871,
-      "required_pages_per_second_upper_estimate": 0.37134481306715805
+      "required_token_positions_per_second": 28641.95808599746,
+      "required_pages_per_second_upper_estimate": 0.3673882849886156
     }
   ],
   "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": 0.08921540196857336,
-    "producer_p95_seconds": 0.2516323130003002,
-    "gpu_update_seconds": 0.9550122834928273,
-    "producer_batches_per_second": 11.208826928250076,
-    "gpu_batches_per_second": 1.0471069506484632,
-    "producer_headroom": 10.704567400023999,
+    "producer_batch_seconds": 0.1025009051954271,
+    "producer_p95_seconds": 0.25530452900056844,
+    "gpu_update_seconds": 0.9652971321648784,
+    "producer_batches_per_second": 9.75601140393259,
+    "gpu_batches_per_second": 1.035950451605811,
+    "producer_headroom": 9.417449829583978,
     "sustainable": true,
-    "buffer_seconds": 60.16577386004812,
+    "buffer_seconds": 60.81371932638734,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363752.86500886036,
-    "upload_bytes_per_second": 4827037.79444969,
+    "generation_bytes_per_second": 430301.1185423427,
+    "upload_bytes_per_second": 4642072.979772448,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.05289709758146,
+    "estimated_full_shards_per_hour": 92.33260314181051,
     "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": 1.0075590680007736,
-      "bytes": 6216793,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8775070329993468,
       "bytes": 1128371,
@@ -562,6 +556,12 @@
       "bytes": 6205051,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 3.739576465002756,
+      "bytes": 2400719,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -233,3 +233,4 @@
 {"step": 463500, "loss": 0.5764064788818359, "lr": 0.0003, "grad_norm": 0.15915337962806428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67842717702, "tokens_per_second": 152862.0500351235, "data_wait_s": 1.4067001757211983e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 463600, "loss": 0.5628827810287476, "lr": 0.0003, "grad_norm": 0.16911696339108737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67857306867, "tokens_per_second": 151545.80803861198, "data_wait_s": 1.6538000636501238e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 463700, "loss": 0.5837123394012451, "lr": 0.0003, "grad_norm": 0.15454984864182733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67871826030, "tokens_per_second": 151179.7593776281, "data_wait_s": 1.3089000276522711e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 463800, "loss": 0.598275899887085, "lr": 0.0003, "grad_norm": 0.13880101897746747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67886414821, "tokens_per_second": 151588.37576327668, "data_wait_s": 8.752998837735504e-06, "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:d17cd104585dcbcaeccff2c740e3bb0c911ce958416da079e1c5ae0100858380
-size 501027
+oid sha256:760da586f5754948f673de2103f98fa61da376816c9b06259b2175d7adb3c30e
+size 503145