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": 75477.93384762073,
-      "required_pages_per_second_upper_estimate": 0.006542236008837351
+      "required_token_positions_per_second": 76011.72263492744,
+      "required_pages_per_second_upper_estimate": 0.006588503468045784
     },
     {
       "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": 9434.741730952592,
-      "required_pages_per_second_upper_estimate": 0.0037636595384364893
+      "required_token_positions_per_second": 9501.46532936593,
+      "required_pages_per_second_upper_estimate": 0.0037902765794502673
     },
     {
       "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": 18869.483461905184,
-      "required_pages_per_second_upper_estimate": 0.00436512290734517
+      "required_token_positions_per_second": 19002.93065873186,
+      "required_pages_per_second_upper_estimate": 0.004395993567740572
     },
     {
       "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": 28304.22519285778,
-      "required_pages_per_second_upper_estimate": 0.0017235788134521979
+      "required_token_positions_per_second": 28504.39598809779,
+      "required_pages_per_second_upper_estimate": 0.0017357681646673166
     },
     {
       "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": 28304.22519285778,
-      "required_pages_per_second_upper_estimate": 0.00217893760645622
+      "required_token_positions_per_second": 28504.39598809779,
+      "required_pages_per_second_upper_estimate": 0.0021943473083820214
     },
     {
       "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": 28304.22519285778,
-      "required_pages_per_second_upper_estimate": 0.35144374874725626
+      "required_token_positions_per_second": 28504.39598809779,
+      "required_pages_per_second_upper_estimate": 0.35392920009558076
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.30017357530469724,
-    "producer_p95_seconds": 0.25362771000072826,
-    "gpu_update_seconds": 0.9768152921202956,
-    "producer_batches_per_second": 3.331405834057611,
-    "gpu_batches_per_second": 1.0237349968481546,
-    "producer_headroom": 3.2541681629662422,
+    "producer_batch_seconds": 0.08249286178906345,
+    "producer_p95_seconds": 0.22215260600023612,
+    "gpu_update_seconds": 0.9699556521578151,
+    "producer_batches_per_second": 12.122260984920464,
+    "gpu_batches_per_second": 1.030974970634324,
+    "producer_headroom": 11.758055559255766,
     "sustainable": true,
-    "buffer_seconds": 60.56254811145833,
+    "buffer_seconds": 60.13725043378454,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421212.13056651806,
-    "upload_bytes_per_second": 4755147.224722894,
+    "generation_bytes_per_second": 420075.11336269835,
+    "upload_bytes_per_second": 4694451.599786369,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.38231790300995,
+    "estimated_full_shards_per_hour": 90.1383404794761,
     "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": 1.176693084999897,
-      "bytes": 12183576,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.0236794259999442,
       "bytes": 7487764,
@@ -562,6 +556,12 @@
       "bytes": 11828419,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1863015779999841,
+      "bytes": 6891470,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -136,3 +136,4 @@
 {"step": 250700, "loss": 0.5758683197200298, "lr": 0.0003, "grad_norm": 0.14111654771792698, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36695727555, "tokens_per_second": 147699.5653436398, "data_wait_s": 1.1117001122329384e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250800, "loss": 0.6128380689769983, "lr": 0.0003, "grad_norm": 0.1576773410874419, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36710345453, "tokens_per_second": 149630.99916733598, "data_wait_s": 1.034299930324778e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250900, "loss": 0.6041381638497114, "lr": 0.0003, "grad_norm": 0.1502225315899803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36725101635, "tokens_per_second": 150323.68269575783, "data_wait_s": 1.3262999345897697e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 251000, "loss": 0.5999299883842468, "lr": 0.0003, "grad_norm": 0.15063485009886873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36739706017, "tokens_per_second": 149506.29546070055, "data_wait_s": 1.5731999155832455e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7b751083b286ea17293a168fb6497d5bbaae18bbf39ce73818ce63228bee35a
-size 294798
+oid sha256:e3c0c17ac26222da592fff477ea615f6e0c9d0d33cc3deeca5fd0c1a1610e139
+size 296916