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": 75929.53717953284,
-      "required_pages_per_second_upper_estimate": 0.006581379841069312
+      "required_token_positions_per_second": 75758.10769105548,
+      "required_pages_per_second_upper_estimate": 0.006566520767492161
     },
     {
       "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": 9491.192147441605,
-      "required_pages_per_second_upper_estimate": 0.0037861784535828964
+      "required_token_positions_per_second": 9469.763461381935,
+      "required_pages_per_second_upper_estimate": 0.003777630230326287
     },
     {
       "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": 18982.38429488321,
-      "required_pages_per_second_upper_estimate": 0.00439124052806784
+      "required_token_positions_per_second": 18939.52692276387,
+      "required_pages_per_second_upper_estimate": 0.004381326229292019
     },
     {
       "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": 28473.57644232481,
-      "required_pages_per_second_upper_estimate": 0.0017338914160274024
+      "required_token_positions_per_second": 28409.290384145803,
+      "required_pages_per_second_upper_estimate": 0.0017299767323671815
     },
     {
       "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": 28473.57644232481,
-      "required_pages_per_second_upper_estimate": 0.002191974734434455
+      "required_token_positions_per_second": 28409.290384145803,
+      "required_pages_per_second_upper_estimate": 0.002187025815720636
     },
     {
       "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": 28473.57644232481,
-      "required_pages_per_second_upper_estimate": 0.3535465244834649
+      "required_token_positions_per_second": 28409.290384145803,
+      "required_pages_per_second_upper_estimate": 0.35274830679247804
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10431687081273822,
-    "producer_p95_seconds": 0.3548281300027156,
-    "gpu_update_seconds": 0.971005523524694,
-    "producer_batches_per_second": 9.58617711793833,
-    "gpu_batches_per_second": 1.0298602590539936,
-    "producer_headroom": 9.308230931004152,
+    "producer_batch_seconds": 0.11943379561733991,
+    "producer_p95_seconds": 0.325602491000609,
+    "gpu_update_seconds": 0.9732027666354295,
+    "producer_batches_per_second": 8.372839486772667,
+    "gpu_batches_per_second": 1.0275350978061994,
+    "producer_headroom": 8.14847055312153,
     "sustainable": true,
-    "buffer_seconds": 60.20234245853103,
+    "buffer_seconds": 60.338571531396624,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 264274.94132433506,
-    "upload_bytes_per_second": 3687463.6772893798,
+    "generation_bytes_per_second": 265012.69704755576,
+    "upload_bytes_per_second": 3651736.483341772,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.70725040242709,
+    "estimated_full_shards_per_hour": 56.86555560655598,
     "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": 1.205836907000048,
-      "bytes": 7335325,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.085818534000282,
       "bytes": 10112362,
@@ -562,6 +556,12 @@
       "bytes": 10345973,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 0.963193997999042,
+      "bytes": 2354048,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 70900, "loss": 0.6655091941356659, "lr": 0.0003, "grad_norm": 0.12564991853884083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10376557970, "tokens_per_second": 148811.52897164496, "data_wait_s": 9.781000699149445e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 71000, "loss": 0.61689068749547, "lr": 0.0003, "grad_norm": 0.13976619797272147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10391028946, "tokens_per_second": 147444.15458343926, "data_wait_s": 1.0232000931864604e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 71100, "loss": 0.6081930510699749, "lr": 0.0003, "grad_norm": 0.13376156518994553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10405604142, "tokens_per_second": 149261.6912249404, "data_wait_s": 1.1981999705312774e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 71200, "loss": 0.6313820295035839, "lr": 0.0003, "grad_norm": 0.14232023127594137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10420259971, "tokens_per_second": 150352.63387042785, "data_wait_s": 1.6206999134737998e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:afba210db8d75ac578ac1a2422eac4c388a7dd45d38326b24fb8cbd5a2fa2e4f
-size 663561
+oid sha256:36be4e780d46e46a9dc60a2272311d03418de468402227c293810cfda70f89d8
+size 665679