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": 77288.03065043864,
-      "required_pages_per_second_upper_estimate": 0.005346965293335711
+      "required_token_positions_per_second": 77404.64191508145,
+      "required_pages_per_second_upper_estimate": 0.0053550327311991126
     },
     {
       "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": 9661.00383130483,
-      "required_pages_per_second_upper_estimate": 0.00404486055946618
+      "required_token_positions_per_second": 9675.58023938518,
+      "required_pages_per_second_upper_estimate": 0.0040509633971394085
     },
     {
       "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": 19322.00766260966,
-      "required_pages_per_second_upper_estimate": 0.004024264640513399
+      "required_token_positions_per_second": 19351.16047877036,
+      "required_pages_per_second_upper_estimate": 0.00403033640330821
     },
     {
       "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": 28983.011493914488,
-      "required_pages_per_second_upper_estimate": 0.0017374872845640407
+      "required_token_positions_per_second": 29026.740718155543,
+      "required_pages_per_second_upper_estimate": 0.0017401087847866328
     },
     {
       "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": 28983.011493914488,
-      "required_pages_per_second_upper_estimate": 0.0022533016608570866
+      "required_token_positions_per_second": 29026.740718155543,
+      "required_pages_per_second_upper_estimate": 0.002256701415697299
     },
     {
       "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": 28983.011493914488,
-      "required_pages_per_second_upper_estimate": 0.3717629519107565
+      "required_token_positions_per_second": 29026.740718155543,
+      "required_pages_per_second_upper_estimate": 0.3723238634465379
     }
   ],
   "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": 0.16258202030471125,
-    "producer_p95_seconds": 0.214224371000455,
-    "gpu_update_seconds": 0.9539381373742063,
-    "producer_batches_per_second": 6.150741626446761,
-    "gpu_batches_per_second": 1.048286006000958,
-    "producer_headroom": 5.867427010602619,
+    "producer_batch_seconds": 0.0737086499608921,
+    "producer_p95_seconds": 0.2116294959996594,
+    "gpu_update_seconds": 0.9525010151314312,
+    "producer_batches_per_second": 13.566928719092997,
+    "gpu_batches_per_second": 1.0498676475027322,
+    "producer_headroom": 12.922513377151846,
     "sustainable": true,
-    "buffer_seconds": 60.098102654574994,
+    "buffer_seconds": 60.00756395328016,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315188.36230490956,
+    "generation_bytes_per_second": 327788.42299367744,
     "upload_bytes_per_second": 5378013.190800452,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.63208534107652,
+    "estimated_full_shards_per_hour": 70.33576505048507,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -140,3 +140,4 @@
 {"step": 571100, "loss": 0.5819798111915588, "lr": 0.0003, "grad_norm": 0.167887702044046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83592114214, "tokens_per_second": 153179.62478965265, "data_wait_s": 7.991000529727899e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 571200, "loss": 0.6157252192497253, "lr": 0.0003, "grad_norm": 0.16089026064670048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83606793094, "tokens_per_second": 153577.0429783949, "data_wait_s": 1.6521000361535698e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 571300, "loss": 0.6090176105499268, "lr": 0.0003, "grad_norm": 0.1407998120763945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83621563952, "tokens_per_second": 154195.777095314, "data_wait_s": 9.938999937730841e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 571400, "loss": 0.6584348678588867, "lr": 0.0003, "grad_norm": 0.20700154343791624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83636324854, "tokens_per_second": 154613.4693512454, "data_wait_s": 1.1292999261058867e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1295c3cd3ee1ec890c02fd2c1b202540209f9dc9906ed5dbcdcbc4e78849e0d
-size 303270
+oid sha256:6bf7b48a4c93ba990bb1a16a523f2c5dc0ceeba31dd75a3e1ba593c283981c86
+size 305388