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": 76987.85992451664,
-      "required_pages_per_second_upper_estimate": 0.005326198786024385
+      "required_token_positions_per_second": 77296.91881209308,
+      "required_pages_per_second_upper_estimate": 0.005347580196982342
     },
     {
       "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": 9623.48249056458,
-      "required_pages_per_second_upper_estimate": 0.004029151157632931
+      "required_token_positions_per_second": 9662.114851511635,
+      "required_pages_per_second_upper_estimate": 0.0040453257204260285
     },
     {
       "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": 19246.96498112916,
-      "required_pages_per_second_upper_estimate": 0.004008635228969604
+      "required_token_positions_per_second": 19324.22970302327,
+      "required_pages_per_second_upper_estimate": 0.0040247274329324075
     },
     {
       "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": 28870.447471693744,
-      "required_pages_per_second_upper_estimate": 0.0017307392433072173
+      "required_token_positions_per_second": 28986.344554534902,
+      "required_pages_per_second_upper_estimate": 0.0017376870964589458
     },
     {
       "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": 28870.447471693744,
-      "required_pages_per_second_upper_estimate": 0.002244550303246232
+      "required_token_positions_per_second": 28986.344554534902,
+      "required_pages_per_second_upper_estimate": 0.00225356079165974
     },
     {
       "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": 28870.447471693744,
-      "required_pages_per_second_upper_estimate": 0.37031910149553937
+      "required_token_positions_per_second": 28986.344554534902,
+      "required_pages_per_second_upper_estimate": 0.3718057048336335
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.052450782726587875,
-    "producer_p95_seconds": 0.12270393000017066,
-    "gpu_update_seconds": 0.9576574809624167,
-    "producer_batches_per_second": 19.065492410527728,
-    "gpu_batches_per_second": 1.0442146799657748,
-    "producer_headroom": 18.258211435174058,
+    "producer_batch_seconds": 0.04276236396876776,
+    "producer_p95_seconds": 0.11878730600005838,
+    "gpu_update_seconds": 0.9538284466322774,
+    "producer_batches_per_second": 23.38504954334067,
+    "gpu_batches_per_second": 1.0484065594088146,
+    "producer_headroom": 22.30532548034348,
     "sustainable": true,
-    "buffer_seconds": 60.33242130063225,
+    "buffer_seconds": 60.091192137833474,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 59,
+    "upload_pages_per_commit": 64,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -7,3 +7,4 @@
 {"step": 645700, "loss": 0.557036817073822, "lr": 0.0001, "grad_norm": 0.175969469020709, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94513673484, "tokens_per_second": 152917.05001293845, "data_wait_s": 1.1775999837482232e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 645800, "loss": 0.518687903881073, "lr": 0.0001, "grad_norm": 0.17502375398215597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94528455980, "tokens_per_second": 154864.7007960821, "data_wait_s": 1.1122000159957679e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 645900, "loss": 0.5465204119682312, "lr": 0.0001, "grad_norm": 0.19122648496474157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94543151532, "tokens_per_second": 154011.2369638154, "data_wait_s": 1.0971999927278375e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 646000, "loss": 0.6286115050315857, "lr": 0.0001, "grad_norm": 0.19377578634339912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94557920764, "tokens_per_second": 154680.50080561815, "data_wait_s": 1.4688999954159954e-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:d92bbcec6d9909b5fcc57df3ec322b8cf82af27b646b2c7927c5a85f5189e59f
-size 19227
+oid sha256:bdb8327d730f0f305abd5bb447d4471d52ed6b2ce739734018ee4e6bc24634ad
+size 21345