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": 76924.8172200969,
-      "required_pages_per_second_upper_estimate": 0.0053218373454534
+      "required_token_positions_per_second": 76818.91813542468,
+      "required_pages_per_second_upper_estimate": 0.005314511001056046
     },
     {
       "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": 9615.602152512112,
-      "required_pages_per_second_upper_estimate": 0.004025851824650534
+      "required_token_positions_per_second": 9602.364766928085,
+      "required_pages_per_second_upper_estimate": 0.004020309607734546
     },
     {
       "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": 19231.204305024225,
-      "required_pages_per_second_upper_estimate": 0.004005352695773925
+      "required_token_positions_per_second": 19204.72953385617,
+      "required_pages_per_second_upper_estimate": 0.003999838699126286
     },
     {
       "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": 28846.80645753634,
-      "required_pages_per_second_upper_estimate": 0.0017293220006062192
+      "required_token_positions_per_second": 28807.094300784254,
+      "required_pages_per_second_upper_estimate": 0.001726941317445884
     },
     {
       "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": 28846.80645753634,
-      "required_pages_per_second_upper_estimate": 0.0022427123183813256
+      "required_token_positions_per_second": 28807.094300784254,
+      "required_pages_per_second_upper_estimate": 0.0022396248728691684
     },
     {
       "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": 28846.80645753634,
-      "required_pages_per_second_upper_estimate": 0.3700158599496715
+      "required_token_positions_per_second": 28807.094300784254,
+      "required_pages_per_second_upper_estimate": 0.36950647504244755
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3044116976871578,
-    "producer_p95_seconds": 0.2697191000006569,
-    "gpu_update_seconds": 0.9584423163340098,
-    "producer_batches_per_second": 3.2850248778143025,
-    "gpu_batches_per_second": 1.0433596085625123,
-    "producer_headroom": 3.148506853107188,
+    "producer_batch_seconds": 0.11638977343724832,
+    "producer_p95_seconds": 0.27129352799966,
+    "gpu_update_seconds": 0.9597635815441233,
+    "producer_batches_per_second": 8.591820144225567,
+    "gpu_batches_per_second": 1.0419232603003565,
+    "producer_headroom": 8.246116073604878,
     "sustainable": true,
-    "buffer_seconds": 60.38186592904262,
+    "buffer_seconds": 60.46510563727977,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320565.27807511686,
+    "generation_bytes_per_second": 323082.81843111303,
     "upload_bytes_per_second": 3577646.262733048,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.78584629716997,
+    "estimated_full_shards_per_hour": 69.32605185222667,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -199,3 +199,4 @@
 {"step": 862400, "loss": 0.5745303630828857, "lr": 3e-05, "grad_norm": 0.32823499344767804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126230716829, "tokens_per_second": 152779.29174583525, "data_wait_s": 9.913997928379104e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 862500, "loss": 0.5842007994651794, "lr": 3e-05, "grad_norm": 0.33762071104045116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126245328482, "tokens_per_second": 150705.49796333996, "data_wait_s": 1.340000017080456e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 862600, "loss": 0.5194213390350342, "lr": 3e-05, "grad_norm": 0.3362587762873102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126260021225, "tokens_per_second": 153076.7618359109, "data_wait_s": 1.6547997802263126e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 862700, "loss": 0.5218643546104431, "lr": 3e-05, "grad_norm": 0.299700068937601, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126274764617, "tokens_per_second": 153213.28386212495, "data_wait_s": 1.0991996532538906e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:055fa54b98076bffaded6d2868ca2be3ee402a4e2fcdbfa06c704ac5545bbb2f
-size 429015
+oid sha256:d30644e546181584553bc83dcf1891bdd1096e67508d3a16a4afefd183d94bff
+size 431133