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": 76149.76867906471,
-      "required_pages_per_second_upper_estimate": 0.006600468949277102
+      "required_token_positions_per_second": 75763.18033475344,
+      "required_pages_per_second_upper_estimate": 0.006566960451391403
     },
     {
       "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": 9518.72108488309,
-      "required_pages_per_second_upper_estimate": 0.003797160158322598
+      "required_token_positions_per_second": 9470.39754184418,
+      "required_pages_per_second_upper_estimate": 0.003777883174503024
     },
     {
       "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": 19037.44216976618,
-      "required_pages_per_second_upper_estimate": 0.004403977198436512
+      "required_token_positions_per_second": 18940.79508368836,
+      "required_pages_per_second_upper_estimate": 0.00438161959600303
     },
     {
       "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": 28556.163254649266,
-      "required_pages_per_second_upper_estimate": 0.0017389205195984445
+      "required_token_positions_per_second": 28411.19262553254,
+      "required_pages_per_second_upper_estimate": 0.0017300925688873445
     },
     {
       "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": 28556.163254649266,
-      "required_pages_per_second_upper_estimate": 0.0021983324958621195
+      "required_token_positions_per_second": 28411.19262553254,
+      "required_pages_per_second_upper_estimate": 0.0021871722555283283
     },
     {
       "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": 28556.163254649266,
-      "required_pages_per_second_upper_estimate": 0.35457197629225407
+      "required_token_positions_per_second": 28411.19262553254,
+      "required_pages_per_second_upper_estimate": 0.3527719262641089
     }
   ],
   "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": 0.09805698821864439,
-    "producer_p95_seconds": 0.33653807299924665,
-    "gpu_update_seconds": 0.9681972943441061,
-    "producer_batches_per_second": 10.198151280867728,
-    "gpu_batches_per_second": 1.0328473399395712,
-    "producer_headroom": 9.873822477448014,
+    "producer_batch_seconds": 0.24897754762500313,
+    "producer_p95_seconds": 0.3361078819980321,
+    "gpu_update_seconds": 0.9731376068723466,
+    "producer_batches_per_second": 4.016426418924117,
+    "gpu_batches_per_second": 1.0276038999396897,
+    "producer_headroom": 3.908535593490684,
     "sustainable": true,
-    "buffer_seconds": 60.02823224933458,
+    "buffer_seconds": 60.33453162608549,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350478.15094479884,
-    "upload_bytes_per_second": 5605026.84560679,
+    "generation_bytes_per_second": 356333.929902709,
+    "upload_bytes_per_second": 5541624.298394555,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.2044524789617,
+    "estimated_full_shards_per_hour": 76.46096632777169,
     "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": 0.9992884630009939,
-      "bytes": 9868032,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.1347405200012872,
       "bytes": 8295093,
@@ -562,6 +556,12 @@
       "bytes": 11887929,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.0596165809984086,
+      "bytes": 5348616,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 177900, "loss": 0.5778142791241407, "lr": 0.0003, "grad_norm": 0.15374156595544317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26036536685, "tokens_per_second": 149747.60026334377, "data_wait_s": 1.2282001989660785e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 178000, "loss": 0.6353771220892668, "lr": 0.0003, "grad_norm": 0.18181807817259837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26051312650, "tokens_per_second": 151056.3295164474, "data_wait_s": 1.0265001037623733e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 178100, "loss": 0.5576391108334064, "lr": 0.0003, "grad_norm": 0.14733169542950486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26065738654, "tokens_per_second": 146713.24230501478, "data_wait_s": 1.0224997822660953e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 178200, "loss": 0.6205142363905907, "lr": 0.0003, "grad_norm": 0.17995584088244035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26080409144, "tokens_per_second": 148965.27526433783, "data_wait_s": 1.1293999705230817e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd347418a5b324b5715d63e80c204d4f0977827c630d30b4b13bfa1a267919fe
-size 492555
+oid sha256:b5eacfc3c85938b1af4faba5c732d381f7d8ce6e96113bd5bbb3fb779d442bf9
+size 494673