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": 75676.73453439961,
-      "required_pages_per_second_upper_estimate": 0.0065594675485116105
+      "required_token_positions_per_second": 76150.5155527581,
+      "required_pages_per_second_upper_estimate": 0.00660053368639591
     },
     {
       "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": 9459.591816799952,
-      "required_pages_per_second_upper_estimate": 0.0037735726092228946
+      "required_token_positions_per_second": 9518.814444094762,
+      "required_pages_per_second_upper_estimate": 0.003797197400707979
     },
     {
       "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": 18919.183633599903,
-      "required_pages_per_second_upper_estimate": 0.004376620167373596
+      "required_token_positions_per_second": 19037.628888189523,
+      "required_pages_per_second_upper_estimate": 0.004404020392457621
     },
     {
       "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": 28378.775450399855,
-      "required_pages_per_second_upper_estimate": 0.0017281185330015375
+      "required_token_positions_per_second": 28556.443332284285,
+      "required_pages_per_second_upper_estimate": 0.0017389375748569633
     },
     {
       "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": 28378.775450399855,
-      "required_pages_per_second_upper_estimate": 0.00218467669165013
+      "required_token_positions_per_second": 28556.443332284285,
+      "required_pages_per_second_upper_estimate": 0.002198354057013768
     },
     {
       "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": 28378.775450399855,
-      "required_pages_per_second_upper_estimate": 0.35236941344226697
+      "required_token_positions_per_second": 28556.443332284285,
+      "required_pages_per_second_upper_estimate": 0.3545754539191215
     }
   ],
   "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": 0.05846806322666964,
-    "producer_p95_seconds": 0.1913682489994244,
-    "gpu_update_seconds": 0.9742492253875754,
-    "producer_batches_per_second": 17.103354289728887,
-    "gpu_batches_per_second": 1.0264314037326336,
-    "producer_headroom": 16.662929668297632,
+    "producer_batch_seconds": 0.10921463221093575,
+    "producer_p95_seconds": 0.2578285590006999,
+    "gpu_update_seconds": 0.9681877983993458,
+    "producer_batches_per_second": 9.156282265077932,
+    "gpu_batches_per_second": 1.0328574700623656,
+    "producer_headroom": 8.865000767748777,
     "sustainable": true,
-    "buffer_seconds": 60.40345197402968,
+    "buffer_seconds": 60.02764350075944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 433746.1538721722,
-    "upload_bytes_per_second": 4680208.792313828,
+    "generation_bytes_per_second": 436399.59958035685,
+    "upload_bytes_per_second": 4762251.150245147,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.07182752727388,
+    "estimated_full_shards_per_hour": 93.64119520719557,
     "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": 1.007327716999498,
-      "bytes": 296016,
-      "pages": 9,
-      "shards": 1
-    },
     {
       "seconds": 1.1130705590003345,
       "bytes": 6042311,
@@ -562,6 +556,12 @@
       "bytes": 1102759,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0251637159999518,
+      "bytes": 7585581,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -154,3 +154,4 @@
 {"step": 252500, "loss": 0.6091282796114683, "lr": 0.0003, "grad_norm": 0.1357873190525464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36959122419, "tokens_per_second": 149281.72665783617, "data_wait_s": 1.4854000255581923e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 252600, "loss": 0.6744916886091232, "lr": 0.0003, "grad_norm": 0.19989628006280338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36973722873, "tokens_per_second": 149224.28293830017, "data_wait_s": 9.437999324291013e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 252700, "loss": 0.6244487520307302, "lr": 0.0003, "grad_norm": 0.15268375710365056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36988352592, "tokens_per_second": 149682.95299119884, "data_wait_s": 8.667000656714663e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 252800, "loss": 0.5966698415577412, "lr": 0.0003, "grad_norm": 0.12863293663952916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37003126292, "tokens_per_second": 151806.60624724728, "data_wait_s": 9.617999239708297e-06, "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 sha256:80c695bd777b47ea7d7ff164d6a9e572f5824a5a9a80c24cce5db06fc1c6c98c
-size 332922
+oid sha256:d843218bd7d48fcaeaa97075fb018de7c1684290ab815ccf479825e944e9d8f6
+size 335040