Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75786.90373130872,
-      "required_pages_per_second_upper_estimate": 0.006569016735278415
+      "required_token_positions_per_second": 75838.71113147883,
+      "required_pages_per_second_upper_estimate": 0.006573507269420505
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9473.36296641359,
-      "required_pages_per_second_upper_estimate": 0.0037790661267008098
+      "required_token_positions_per_second": 9479.838891434854,
+      "required_pages_per_second_upper_estimate": 0.0037816494700154994
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18946.72593282718,
-      "required_pages_per_second_upper_estimate": 0.004382991593571916
+      "required_token_positions_per_second": 18959.67778286971,
+      "required_pages_per_second_upper_estimate": 0.00438598777613448
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28420.088899240767,
-      "required_pages_per_second_upper_estimate": 0.0017306343052810346
+      "required_token_positions_per_second": 28439.516674304563,
+      "required_pages_per_second_upper_estimate": 0.0017318173548527616
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28420.088899240767,
-      "required_pages_per_second_upper_estimate": 0.002187857115304849
+      "required_token_positions_per_second": 28439.516674304563,
+      "required_pages_per_second_upper_estimate": 0.0021893527192087836
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28420.088899240767,
-      "required_pages_per_second_upper_estimate": 0.3528823882096523
+      "required_token_positions_per_second": 28439.516674304563,
+      "required_pages_per_second_upper_estimate": 0.35312361615536414
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07892648374215128,
-    "producer_p95_seconds": 0.2091615819990693,
-    "gpu_update_seconds": 0.9728329878918361,
-    "producer_batches_per_second": 12.670018384031247,
-    "gpu_batches_per_second": 1.0279256690986966,
-    "producer_headroom": 12.325811841181611,
+    "producer_batch_seconds": 0.2724060910624644,
+    "producer_p95_seconds": 0.21644839299915475,
+    "gpu_update_seconds": 0.9721684203227087,
+    "producer_batches_per_second": 3.670989867002254,
+    "gpu_batches_per_second": 1.0286283519352055,
+    "producer_headroom": 3.5688204200242515,
     "sustainable": true,
-    "buffer_seconds": 60.315645249293844,
+    "buffer_seconds": 60.274442060007935,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 247079.8643692135,
+    "generation_bytes_per_second": 237571.42323431477,
     "upload_bytes_per_second": 4800890.542730465,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.01758716876319,
+    "estimated_full_shards_per_hour": 50.97729704639514,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 7500, "loss": 0.7781140841543674, "lr": 0.0003, "grad_norm": 0.17605280490808378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1097209673, "tokens_per_second": 149652.25085815013, "data_wait_s": 9.50599951465847e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 7600, "loss": 0.7432269118726254, "lr": 0.0003, "grad_norm": 0.16820907072428087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1111896849, "tokens_per_second": 148647.4093221275, "data_wait_s": 1.3171000318834558e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 7700, "loss": 0.7400143817067146, "lr": 0.0003, "grad_norm": 0.15589289663933553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1126432553, "tokens_per_second": 148234.65333558523, "data_wait_s": 9.495999620412476e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 7800, "loss": 0.7411314342170954, "lr": 0.0003, "grad_norm": 0.17838141708518138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1140947358, "tokens_per_second": 148041.28679959124, "data_wait_s": 1.5087000065250322e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09a44a8886afa100cda0cf2ade7192bfc85a29ed6d1cb7663f35b0617c89c9b2
-size 161290
+oid sha256:1dbab8204f09729dff5a1da46641ca2f16c113ff3e3941d0b5441a1293a1f899
+size 163373