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": 76419.10590740042,
-      "required_pages_per_second_upper_estimate": 0.005286851063429147
+      "required_token_positions_per_second": 76652.56833051989,
+      "required_pages_per_second_upper_estimate": 0.005303002535568007
     },
     {
       "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": 9552.388238425052,
-      "required_pages_per_second_upper_estimate": 0.003999385478878916
+      "required_token_positions_per_second": 9581.571041314986,
+      "required_pages_per_second_upper_estimate": 0.00401160370904271
     },
     {
       "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": 19104.776476850104,
-      "required_pages_per_second_upper_estimate": 0.003979021113291295
+      "required_token_positions_per_second": 19163.142082629973,
+      "required_pages_per_second_upper_estimate": 0.00399117712977071
     },
     {
       "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": 28657.164715275158,
-      "required_pages_per_second_upper_estimate": 0.0017179532677238354
+      "required_token_positions_per_second": 28744.71312394496,
+      "required_pages_per_second_upper_estimate": 0.0017232016611449102
     },
     {
       "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": 28657.164715275158,
-      "required_pages_per_second_upper_estimate": 0.002227968507066388
+      "required_token_positions_per_second": 28744.71312394496,
+      "required_pages_per_second_upper_estimate": 0.002234775010755712
     },
     {
       "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": 28657.164715275158,
-      "required_pages_per_second_upper_estimate": 0.3675833393013835
+      "required_token_positions_per_second": 28744.71312394496,
+      "required_pages_per_second_upper_estimate": 0.3687063162856423
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07639543868759802,
-    "producer_p95_seconds": 0.2603119070008688,
-    "gpu_update_seconds": 0.9647849071845813,
-    "producer_batches_per_second": 13.089786735688179,
-    "gpu_batches_per_second": 1.0365004598985517,
-    "producer_headroom": 12.628828680856882,
+    "producer_batch_seconds": 0.508844152523352,
+    "producer_p95_seconds": 0.2626986079994822,
+    "gpu_update_seconds": 0.9618464404492048,
+    "producer_batches_per_second": 1.9652382660604668,
+    "gpu_batches_per_second": 1.039666996670463,
+    "producer_headroom": 1.8902574308448272,
     "sustainable": true,
-    "buffer_seconds": 60.781449152628625,
+    "buffer_seconds": 60.5963257482999,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266731.09191876714,
-    "upload_bytes_per_second": 4685187.166482214,
+    "generation_bytes_per_second": 250996.09345919397,
+    "upload_bytes_per_second": 4752041.034790239,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.23428314373265,
+    "estimated_full_shards_per_hour": 53.85791876632561,
     "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": 1.5225848519985448,
-      "bytes": 6009508,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.197222058999614,
       "bytes": 6186205,
@@ -562,6 +556,12 @@
       "bytes": 10700049,
       "pages": 47,
       "shards": 3
+    },
+    {
+      "seconds": 1.4434581130008155,
+      "bytes": 11762868,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -283,3 +283,4 @@
 {"step": 913300, "loss": 0.5021415948867798, "lr": 3e-05, "grad_norm": 0.3269248681995199, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133680802070, "tokens_per_second": 123859.29355803684, "data_wait_s": 1.0933003068203107e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 913400, "loss": 0.5328166484832764, "lr": 3e-05, "grad_norm": 0.38551726637670986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133695394689, "tokens_per_second": 152045.93260453452, "data_wait_s": 1.412599885952659e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 913500, "loss": 0.5162062644958496, "lr": 3e-05, "grad_norm": 0.327755932234263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133710077592, "tokens_per_second": 146901.19571039188, "data_wait_s": 1.468299888074398e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 913600, "loss": 0.5379030704498291, "lr": 3e-05, "grad_norm": 0.3380548433952475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133724540984, "tokens_per_second": 150815.59801819993, "data_wait_s": 1.060399881680496e-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 sha256:070908054b8f93a5268ceaf396fff00f861462a66b55c5b63f8a6f62d0c8ea49
-size 607710
+oid sha256:42efc67ff76b031af5aaf2474d52cdd278b1fdfa4a4238b1c9c81962954e1153
+size 609828