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": 76666.89628392222,
-      "required_pages_per_second_upper_estimate": 0.0053039937766298124
+      "required_token_positions_per_second": 77261.44503327008,
+      "required_pages_per_second_upper_estimate": 0.00534512603865286
     },
     {
       "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": 9583.362035490278,
-      "required_pages_per_second_upper_estimate": 0.004012353560903693
+      "required_token_positions_per_second": 9657.68062915876,
+      "required_pages_per_second_upper_estimate": 0.004043469204123972
     },
     {
       "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": 19166.724070980556,
-      "required_pages_per_second_upper_estimate": 0.003991923163480751
+      "required_token_positions_per_second": 19315.36125831752,
+      "required_pages_per_second_upper_estimate": 0.004022880369776814
     },
     {
       "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": 28750.086106470837,
-      "required_pages_per_second_upper_estimate": 0.0017235237632432674
+      "required_token_positions_per_second": 28973.041887476276,
+      "required_pages_per_second_upper_estimate": 0.0017368896218807777
     },
     {
       "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": 28750.086106470837,
-      "required_pages_per_second_upper_estimate": 0.0022351927365138464
+      "required_token_positions_per_second": 28973.041887476276,
+      "required_pages_per_second_upper_estimate": 0.00225252656780818
     },
     {
       "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": 28750.086106470837,
-      "required_pages_per_second_upper_estimate": 0.3687752351364251
+      "required_token_positions_per_second": 28973.041887476276,
+      "required_pages_per_second_upper_estimate": 0.3716350725038965
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.13340942717974258,
-    "producer_p95_seconds": 0.35133579099783674,
-    "gpu_update_seconds": 0.9616666850182829,
-    "producer_batches_per_second": 7.495722162517793,
-    "gpu_batches_per_second": 1.0398613319759418,
-    "producer_headroom": 7.20838628384656,
+    "producer_batch_seconds": 0.12061048737535884,
+    "producer_p95_seconds": 0.3548598529960145,
+    "gpu_update_seconds": 0.9542663869184881,
+    "producer_batches_per_second": 8.291152964898005,
+    "gpu_batches_per_second": 1.0479254154903168,
+    "producer_headroom": 7.91196858320173,
     "sustainable": true,
-    "buffer_seconds": 60.585001156151826,
+    "buffer_seconds": 60.11878237586475,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421591.23770546523,
-    "upload_bytes_per_second": 5302777.186454206,
+    "generation_bytes_per_second": 420290.59650202515,
+    "upload_bytes_per_second": 5302101.0399973355,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.46366546986549,
+    "estimated_full_shards_per_hour": 90.18457814498487,
     "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": 1.0296640300002764,
-      "bytes": 6697132,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.1917317950064898,
       "bytes": 7979357,
@@ -562,6 +556,12 @@
       "bytes": 11215432,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.9423328039993066,
+      "bytes": 6179115,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -429,3 +429,4 @@
 {"step": 600000, "loss": 0.5916668176651001, "lr": 0.0003, "grad_norm": 0.15561927337962328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87822268665, "tokens_per_second": 152571.70482406951, "data_wait_s": 1.7187994671985507e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600100, "loss": 0.6710374355316162, "lr": 0.0003, "grad_norm": 0.1682189764871132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87836984173, "tokens_per_second": 138292.41131993674, "data_wait_s": 1.3665005099028349e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600200, "loss": 0.5465240478515625, "lr": 0.0003, "grad_norm": 0.1641682083351951, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87851654044, "tokens_per_second": 153067.15591660165, "data_wait_s": 1.2889002391602844e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 600300, "loss": 0.5746079087257385, "lr": 0.0003, "grad_norm": 0.16822399315171796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87866176121, "tokens_per_second": 150920.4173547582, "data_wait_s": 9.325005521532148e-06, "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:c52252617866f595813d1f8e6b9b4dc114694e5e7af6deca990b920dbe8ad6d1
-size 920070
+oid sha256:9f277d77adf36de633555645436ee466166cd68e02c48213dafe314fa74c6f0c
+size 922188