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": 77227.10696224961,
-      "required_pages_per_second_upper_estimate": 0.005342750451224362
+      "required_token_positions_per_second": 76036.78666844648,
+      "required_pages_per_second_upper_estimate": 0.0052604013313754605
     },
     {
       "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": 9653.388370281202,
-      "required_pages_per_second_upper_estimate": 0.004041672124964497
+      "required_token_positions_per_second": 9504.59833355581,
+      "required_pages_per_second_upper_estimate": 0.003979376843676861
     },
     {
       "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": 19306.776740562404,
-      "required_pages_per_second_upper_estimate": 0.00402109244111738
+      "required_token_positions_per_second": 19009.19666711162,
+      "required_pages_per_second_upper_estimate": 0.003959114359531855
     },
     {
       "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": 28960.165110843605,
-      "required_pages_per_second_upper_estimate": 0.0017361176787833485
+      "required_token_positions_per_second": 28513.79500066743,
+      "required_pages_per_second_upper_estimate": 0.0017093584722459808
     },
     {
       "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": 28960.165110843605,
-      "required_pages_per_second_upper_estimate": 0.0022515254550639447
+      "required_token_positions_per_second": 28513.79500066743,
+      "required_pages_per_second_upper_estimate": 0.002216822142372366
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28960.165110843605,
-      "required_pages_per_second_upper_estimate": 0.37146990303925814
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28513.79500066743,
+      "required_pages_per_second_upper_estimate": 0.36574434654080157
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09282537171850436,
-    "producer_p95_seconds": 0.27284220900037326,
-    "gpu_update_seconds": 0.9546906895792424,
-    "producer_batches_per_second": 10.772916730487534,
-    "gpu_batches_per_second": 1.0474596755947485,
-    "producer_headroom": 10.284803302208902,
+    "producer_batch_seconds": 0.35503970422641373,
+    "producer_p95_seconds": 0.2906274579981982,
+    "gpu_update_seconds": 0.9696359253250167,
+    "producer_batches_per_second": 2.816586393284865,
+    "gpu_batches_per_second": 1.0313149233458996,
+    "producer_headroom": 2.731063353710622,
     "sustainable": true,
-    "buffer_seconds": 60.14551344349227,
+    "buffer_seconds": 60.11742737015104,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385525.34591926774,
+    "generation_bytes_per_second": 347549.10530162806,
     "upload_bytes_per_second": 4923662.6315234015,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.72476466353916,
+    "estimated_full_shards_per_hour": 74.57594746863013,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -253,3 +253,4 @@
 {"step": 910300, "loss": 0.5121055245399475, "lr": 3e-05, "grad_norm": 0.34960591957401166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133241637469, "tokens_per_second": 152278.4671584382, "data_wait_s": 2.0385999960126355e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 910400, "loss": 0.5518227815628052, "lr": 3e-05, "grad_norm": 0.33290578631262724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133256320466, "tokens_per_second": 153583.74441538518, "data_wait_s": 1.0552001185715199e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 910500, "loss": 0.5678766369819641, "lr": 3e-05, "grad_norm": 0.3356754157103991, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133271017271, "tokens_per_second": 152920.34405369387, "data_wait_s": 1.5575998986605555e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 910600, "loss": 0.6760541796684265, "lr": 3e-05, "grad_norm": 0.41506338403640874, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133285707646, "tokens_per_second": 153324.11839466242, "data_wait_s": 1.7293998098466545e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:10e5c3c1d099d34a2e6771b443abf8705d81fde72f82820b1e529b0063114551
-size 544170
+oid sha256:a51217c477a1e6bca81a74bcb5ccaae8df27d5285dd7b28cc711e43b37ac97d7
+size 546288