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": 77003.31108215684,
-      "required_pages_per_second_upper_estimate": 0.00532726773296156
+      "required_token_positions_per_second": 76886.06691899558,
+      "required_pages_per_second_upper_estimate": 0.005319156509710125
     },
     {
       "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": 9625.413885269605,
-      "required_pages_per_second_upper_estimate": 0.004029959792263817
+      "required_token_positions_per_second": 9610.758364874448,
+      "required_pages_per_second_upper_estimate": 0.004023823831916432
     },
     {
       "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": 19250.82777053921,
-      "required_pages_per_second_upper_estimate": 0.00400943974613511
+      "required_token_positions_per_second": 19221.516729748895,
+      "required_pages_per_second_upper_estimate": 0.004003335029322614
     },
     {
       "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": 28876.241655808815,
-      "required_pages_per_second_upper_estimate": 0.0017310865957977089
+      "required_token_positions_per_second": 28832.275094623343,
+      "required_pages_per_second_upper_estimate": 0.0017284508675876917
     },
     {
       "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": 28876.241655808815,
-      "required_pages_per_second_upper_estimate": 0.0022450007755752564
+      "required_token_positions_per_second": 28832.275094623343,
+      "required_pages_per_second_upper_estimate": 0.0022415825688315506
     },
     {
       "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": 28876.241655808815,
-      "required_pages_per_second_upper_estimate": 0.3703934230680573
+      "required_token_positions_per_second": 28832.275094623343,
+      "required_pages_per_second_upper_estimate": 0.36982946722878546
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.44927313051576334,
-    "producer_p95_seconds": 0.2618710770002508,
-    "gpu_update_seconds": 0.9574653214760814,
-    "producer_batches_per_second": 2.225817508498683,
-    "gpu_batches_per_second": 1.044424249703733,
-    "producer_headroom": 2.131143076321782,
+    "producer_batch_seconds": 0.1243443915547573,
+    "producer_p95_seconds": 0.2668740259978222,
+    "gpu_update_seconds": 0.958925367813094,
+    "producer_batches_per_second": 8.042180169900401,
+    "gpu_batches_per_second": 1.0428340239664116,
+    "producer_headroom": 7.711850577440913,
     "sustainable": true,
-    "buffer_seconds": 60.32031525299313,
+    "buffer_seconds": 60.41229817222492,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307430.21234080393,
+    "generation_bytes_per_second": 318817.51138996455,
     "upload_bytes_per_second": 3949280.9851866667,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.96736695926751,
+    "estimated_full_shards_per_hour": 68.41081625246241,
     "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-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -282,3 +282,4 @@
 {"step": 553200, "loss": 0.5633944869041443, "lr": 0.0003, "grad_norm": 0.16626660762691406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80970770951, "tokens_per_second": 153594.97259423923, "data_wait_s": 1.8889997591031715e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 553300, "loss": 0.6275357604026794, "lr": 0.0003, "grad_norm": 0.17894805029477764, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80985417451, "tokens_per_second": 149258.7486075883, "data_wait_s": 1.2778997188434005e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 553400, "loss": 0.5868434906005859, "lr": 0.0003, "grad_norm": 0.18740991507098057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81000037393, "tokens_per_second": 151778.89971913135, "data_wait_s": 1.3491000572685152e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 553500, "loss": 0.5721019506454468, "lr": 0.0003, "grad_norm": 0.13858546248677567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81014783354, "tokens_per_second": 153291.6785604879, "data_wait_s": 1.3586002751253545e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:41e590a6381aef55f804679cc19bdd66b3363c28fd0e7ec22ae0881a4ac772d9
-size 605592
+oid sha256:b3e450accfcd0235590e68030829bea90db3549c09378c67b5ea357a359495a5
+size 607710