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": 75960.15035579941,
-      "required_pages_per_second_upper_estimate": 0.005255099453445792
+      "required_token_positions_per_second": 75900.36261827842,
+      "required_pages_per_second_upper_estimate": 0.005250963199037426
     },
     {
       "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": 9495.018794474927,
-      "required_pages_per_second_upper_estimate": 0.003975366090707219
+      "required_token_positions_per_second": 9487.545327284803,
+      "required_pages_per_second_upper_estimate": 0.003972237106058455
     },
     {
       "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": 18990.037588949854,
-      "required_pages_per_second_upper_estimate": 0.003955124028809627
+      "required_token_positions_per_second": 18975.090654569605,
+      "required_pages_per_second_upper_estimate": 0.003952010976555388
     },
     {
       "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": 28485.056383424777,
-      "required_pages_per_second_upper_estimate": 0.0017076356360236117
+      "required_token_positions_per_second": 28462.63598185441,
+      "required_pages_per_second_upper_estimate": 0.001706291567183439
     },
     {
       "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": 28485.056383424777,
-      "required_pages_per_second_upper_estimate": 0.0022145878412895676
+      "required_token_positions_per_second": 28462.63598185441,
+      "required_pages_per_second_upper_estimate": 0.002212844753684396
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28485.056383424777,
-      "required_pages_per_second_upper_estimate": 0.36537571841593586
+      "required_token_positions_per_second": 28462.63598185441,
+      "required_pages_per_second_upper_estimate": 0.3650881335777429
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.2688127188597491,
-    "producer_p95_seconds": 0.34204886400402756,
-    "gpu_update_seconds": 0.9706141925029906,
-    "producer_batches_per_second": 3.7200620723669777,
-    "gpu_batches_per_second": 1.0302754768310467,
-    "producer_headroom": 3.610745044431476,
+    "producer_batch_seconds": 0.11333866012506633,
+    "producer_p95_seconds": 0.3496470869940822,
+    "gpu_update_seconds": 0.9713787583703154,
+    "producer_batches_per_second": 8.823114715636532,
+    "gpu_batches_per_second": 1.0294645537418408,
+    "producer_headroom": 8.570586217433872,
     "sustainable": true,
-    "buffer_seconds": 60.17807993518542,
+    "buffer_seconds": 60.22548301895956,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 485371.94957725826,
+    "generation_bytes_per_second": 490817.03851570963,
     "upload_bytes_per_second": 3836117.136573954,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.14952149856863,
+    "estimated_full_shards_per_hour": 105.31791082957713,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -363,3 +363,4 @@
 {"step": 836300, "loss": 0.5561316013336182, "lr": 3e-05, "grad_norm": 0.2900443949403043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122410347896, "tokens_per_second": 149873.65379467962, "data_wait_s": 8.816001354716718e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 836400, "loss": 0.4622453451156616, "lr": 3e-05, "grad_norm": 0.295949444163284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122424908499, "tokens_per_second": 150424.64092202057, "data_wait_s": 8.505994628649205e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 836500, "loss": 0.5219582319259644, "lr": 3e-05, "grad_norm": 0.2973174038022154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122439594822, "tokens_per_second": 152322.13440774338, "data_wait_s": 8.138995326589793e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 836600, "loss": 0.5593646168708801, "lr": 3e-05, "grad_norm": 0.28687618863860886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122454269431, "tokens_per_second": 152319.09347932588, "data_wait_s": 1.5463992895092815e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d74ee6905f38c5eefdf6de86c0b5f59ddc97023130f1255f489bb894593d101
-size 778716
+oid sha256:2d73415cfaaacaefa55ed8529192c539483f9f7eeec774a320400a55a88dacf3
+size 780834