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": 76481.61050310897,
-      "required_pages_per_second_upper_estimate": 0.005291175276390908
+      "required_token_positions_per_second": 76811.81564125927,
+      "required_pages_per_second_upper_estimate": 0.005314019634029625
     },
     {
       "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": 9560.201312888621,
-      "required_pages_per_second_upper_estimate": 0.004002656650001265
+      "required_token_positions_per_second": 9601.476955157408,
+      "required_pages_per_second_upper_estimate": 0.004019937899485781
     },
     {
       "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": 19120.402625777242,
-      "required_pages_per_second_upper_estimate": 0.003982275628023559
+      "required_token_positions_per_second": 19202.953910314816,
+      "required_pages_per_second_upper_estimate": 0.003999468883568964
     },
     {
       "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": 28680.603938665863,
-      "required_pages_per_second_upper_estimate": 0.0017193584133764865
+      "required_token_positions_per_second": 28804.430865472223,
+      "required_pages_per_second_upper_estimate": 0.001726781648565758
     },
     {
       "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": 28680.603938665863,
-      "required_pages_per_second_upper_estimate": 0.002229790803586767
+      "required_token_positions_per_second": 28804.430865472223,
+      "required_pages_per_second_upper_estimate": 0.002239417802488872
     },
     {
       "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": 28680.603938665863,
-      "required_pages_per_second_upper_estimate": 0.3678839924919622
+      "required_token_positions_per_second": 28804.430865472223,
+      "required_pages_per_second_upper_estimate": 0.36947231135403885
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07352219521106917,
-    "producer_p95_seconds": 0.255219880000368,
-    "gpu_update_seconds": 0.9639964367251781,
-    "producer_batches_per_second": 13.601334904775047,
-    "gpu_batches_per_second": 1.0373482327353105,
-    "producer_headroom": 13.111638382908936,
+    "producer_batch_seconds": 0.089442131687548,
+    "producer_p95_seconds": 0.2647361590024957,
+    "gpu_update_seconds": 0.9598523272036444,
+    "producer_batches_per_second": 11.180413314536628,
+    "gpu_batches_per_second": 1.0418269265578786,
+    "producer_headroom": 10.731545739056594,
     "sustainable": true,
-    "buffer_seconds": 60.73177551368622,
+    "buffer_seconds": 60.4706966138296,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392318.91666500096,
-    "upload_bytes_per_second": 5364724.667990984,
+    "generation_bytes_per_second": 423002.34433009673,
+    "upload_bytes_per_second": 5292571.983536604,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.18250679934046,
+    "estimated_full_shards_per_hour": 90.76645610263039,
     "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": 1.3973069029998442,
-      "bytes": 13145960,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.3652075969985162,
       "bytes": 5700613,
@@ -562,6 +556,12 @@
       "bytes": 5879867,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1945516140003747,
+      "bytes": 6236357,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -248,3 +248,4 @@
 {"step": 744800, "loss": 0.5238848328590393, "lr": 5e-05, "grad_norm": 0.23703562577114629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109018512323, "tokens_per_second": 152851.77457409486, "data_wait_s": 1.4173001545714214e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 744900, "loss": 0.553131103515625, "lr": 5e-05, "grad_norm": 0.24299733871269344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109033203236, "tokens_per_second": 151043.68807401773, "data_wait_s": 1.4522000128636137e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 745000, "loss": 0.5192437767982483, "lr": 5e-05, "grad_norm": 0.23177553583662464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109047827948, "tokens_per_second": 151153.61984531832, "data_wait_s": 1.3546999980462715e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 745100, "loss": 0.5198097229003906, "lr": 5e-05, "grad_norm": 0.22867919331413317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109062409782, "tokens_per_second": 135812.38742473102, "data_wait_s": 8.88399881660007e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3575b5c7491965e1a8cf9c1cd1d93fb3bb6bcd7331bf0dca92f3ab3775e2886d
-size 533580
+oid sha256:852fe49bb16a240533a6388a1943793783fa370f52ae03d5ac4fdee85bd448bc
+size 535698