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": 77508.20808454201,
-      "required_pages_per_second_upper_estimate": 0.0053621976791090155
+      "required_token_positions_per_second": 76732.00919984178,
+      "required_pages_per_second_upper_estimate": 0.005308498439235907
     },
     {
       "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": 9688.526010567752,
-      "required_pages_per_second_upper_estimate": 0.004056383521195107
+      "required_token_positions_per_second": 9591.501149980222,
+      "required_pages_per_second_upper_estimate": 0.004015761238176594
     },
     {
       "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": 19377.052021135503,
-      "required_pages_per_second_upper_estimate": 0.00403572892877698
+      "required_token_positions_per_second": 19183.002299960444,
+      "required_pages_per_second_upper_estimate": 0.003995313489291496
     },
     {
       "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": 29065.57803170325,
-      "required_pages_per_second_upper_estimate": 0.0017424370224326737
+      "required_token_positions_per_second": 28774.503449940665,
+      "required_pages_per_second_upper_estimate": 0.0017249875457011588
     },
     {
       "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": 29065.57803170325,
-      "required_pages_per_second_upper_estimate": 0.0022597208459983445
+      "required_token_positions_per_second": 28774.503449940665,
+      "required_pages_per_second_upper_estimate": 0.0022370910775680822
     },
     {
       "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": 29065.57803170325,
-      "required_pages_per_second_upper_estimate": 0.3728220268044696
+      "required_token_positions_per_second": 28774.503449940665,
+      "required_pages_per_second_upper_estimate": 0.36908843460115526
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05976310076579239,
-    "producer_p95_seconds": 0.209997089998069,
-    "gpu_update_seconds": 0.9512282869393813,
-    "producer_batches_per_second": 16.73273286001229,
-    "gpu_batches_per_second": 1.05127235357723,
-    "producer_headroom": 15.916648814243786,
+    "producer_batch_seconds": 0.39928509835146997,
+    "producer_p95_seconds": 0.22588819999873522,
+    "gpu_update_seconds": 0.960850638069204,
+    "producer_batches_per_second": 2.5044761352945657,
+    "gpu_batches_per_second": 1.040744482419729,
+    "producer_headroom": 2.4064274926268774,
     "sustainable": true,
-    "buffer_seconds": 60.878610364120405,
+    "buffer_seconds": 60.53359019835985,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304544.9367569112,
-    "upload_bytes_per_second": 5427240.817982717,
+    "generation_bytes_per_second": 290518.2136550418,
+    "upload_bytes_per_second": 5417812.69215267,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.34825398474219,
+    "estimated_full_shards_per_hour": 62.3384457324833,
     "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": 0.9305929850015673,
-      "bytes": 1085853,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.4772485220000817,
       "bytes": 7565247,
@@ -562,6 +556,12 @@
       "bytes": 10806597,
       "pages": 46,
       "shards": 3
+    },
+    {
+      "seconds": 0.989914507001231,
+      "bytes": 672610,
+      "pages": 18,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -175,3 +175,4 @@
 {"step": 542500, "loss": 0.6364598870277405, "lr": 0.0003, "grad_norm": 0.16701580214921663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79404247980, "tokens_per_second": 152739.05460617493, "data_wait_s": 8.732997230254114e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 542600, "loss": 0.682414710521698, "lr": 0.0003, "grad_norm": 0.17781160082241837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79419011113, "tokens_per_second": 154304.11100270035, "data_wait_s": 1.7385998944519088e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 542700, "loss": 0.5699137449264526, "lr": 0.0003, "grad_norm": 0.1479214455784314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79433781242, "tokens_per_second": 150418.86932520557, "data_wait_s": 9.545001375954598e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 542800, "loss": 0.5423920154571533, "lr": 0.0003, "grad_norm": 0.16229336772674785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79448303344, "tokens_per_second": 151220.55167053622, "data_wait_s": 8.675000572111458e-06, "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 sha256:6f110be1eb23115ec7739f592e19032d1df75df2cad5f6d1b2b5eb69ab56667f
-size 377400
+oid sha256:df046b8bd88633e25bf3f99ae86d18c5c690045756847cd6b9d972569d997c5f
+size 379518