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": 77636.33772862358,
-      "required_pages_per_second_upper_estimate": 0.005371061985188313
+      "required_token_positions_per_second": 76513.89309956823,
+      "required_pages_per_second_upper_estimate": 0.005293408661319904
     },
     {
       "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": 9704.542216077947,
-      "required_pages_per_second_upper_estimate": 0.004063089171985823
+      "required_token_positions_per_second": 9564.236637446029,
+      "required_pages_per_second_upper_estimate": 0.004004346156126292
     },
     {
       "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": 19409.084432155894,
-      "required_pages_per_second_upper_estimate": 0.004042400435241042
+      "required_token_positions_per_second": 19128.473274892058,
+      "required_pages_per_second_upper_estimate": 0.003983956531396845
     },
     {
       "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": 29113.626648233843,
-      "required_pages_per_second_upper_estimate": 0.0017453174636276932
+      "required_token_positions_per_second": 28692.709912338087,
+      "required_pages_per_second_upper_estimate": 0.0017200841480133847
     },
     {
       "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": 29113.626648233843,
-      "required_pages_per_second_upper_estimate": 0.0022634564145900694
+      "required_token_positions_per_second": 28692.709912338087,
+      "required_pages_per_second_upper_estimate": 0.0022307319897912317
     },
     {
       "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": 29113.626648233843,
-      "required_pages_per_second_upper_estimate": 0.37343834286673905
+      "required_token_positions_per_second": 28692.709912338087,
+      "required_pages_per_second_upper_estimate": 0.36803927492384764
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.14946511808625473,
-    "producer_p95_seconds": 0.33449703899532324,
-    "gpu_update_seconds": 0.9496583965322385,
-    "producer_batches_per_second": 6.690524269501534,
-    "gpu_batches_per_second": 1.0530102230987357,
-    "producer_headroom": 6.353712549734853,
+    "producer_batch_seconds": 0.47957559729690047,
+    "producer_p95_seconds": 0.32808332899730885,
+    "gpu_update_seconds": 0.9635897091794438,
+    "producer_batches_per_second": 2.0851769890637493,
+    "gpu_batches_per_second": 1.0377860934728764,
+    "producer_headroom": 2.0092550884796063,
     "sustainable": true,
-    "buffer_seconds": 60.77813737806326,
+    "buffer_seconds": 60.706151678304956,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 543379.7397463934,
-    "upload_bytes_per_second": 4989679.253709886,
+    "generation_bytes_per_second": 459784.8506387204,
+    "upload_bytes_per_second": 5015322.827406912,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.59664291662075,
+    "estimated_full_shards_per_hour": 98.65912570353707,
     "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": 0.7943182929993782,
-      "bytes": 1190422,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 1.1334048999997322,
       "bytes": 5665943,
@@ -562,6 +556,12 @@
       "bytes": 6235413,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.6395168350063614,
+      "bytes": 7620206,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -378,3 +378,4 @@
 {"step": 682800, "loss": 0.5299861431121826, "lr": 0.0001, "grad_norm": 0.19779118532524445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99944461786, "tokens_per_second": 153490.70288602065, "data_wait_s": 1.0031995770987123e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 682900, "loss": 0.5198776125907898, "lr": 0.0001, "grad_norm": 0.19194340221881184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99958976162, "tokens_per_second": 151873.5404881638, "data_wait_s": 1.4601995644625276e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 683000, "loss": 0.5247899889945984, "lr": 0.0001, "grad_norm": 0.20268397962436385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99973492914, "tokens_per_second": 151112.64794044953, "data_wait_s": 1.226499443873763e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 683100, "loss": 0.540785551071167, "lr": 0.0001, "grad_norm": 0.2015798868947121, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99988027993, "tokens_per_second": 151482.1991421006, "data_wait_s": 8.834998880047351e-06, "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:e4234fd2f238c404ded19b4cfc44c71cda9c83d137a19eaeb05911c0aa479c95
-size 810486
+oid sha256:178103844bf10f9be2bf621e272d19c832acabbcf948c04849dbdb1313107fbb
+size 812604