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": 77015.26364316173,
-      "required_pages_per_second_upper_estimate": 0.00532809463886564
+      "required_token_positions_per_second": 77268.418507879,
+      "required_pages_per_second_upper_estimate": 0.005345608479806995
     },
     {
       "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": 9626.907955395216,
-      "required_pages_per_second_upper_estimate": 0.004030585328225678
+      "required_token_positions_per_second": 9658.552313484875,
+      "required_pages_per_second_upper_estimate": 0.004043834160148478
     },
     {
       "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": 19253.815910790432,
-      "required_pages_per_second_upper_estimate": 0.004010062096946882
+      "required_token_positions_per_second": 19317.10462696975,
+      "required_pages_per_second_upper_estimate": 0.0040232434674913505
     },
     {
       "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": 28880.72386618565,
-      "required_pages_per_second_upper_estimate": 0.0017313552974658611
+      "required_token_positions_per_second": 28975.656940454628,
+      "required_pages_per_second_upper_estimate": 0.001737046390313875
     },
     {
       "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": 28880.72386618565,
-      "required_pages_per_second_upper_estimate": 0.0022453492477168954
+      "required_token_positions_per_second": 28975.656940454628,
+      "required_pages_per_second_upper_estimate": 0.0022527298766748445
     },
     {
       "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": 28880.72386618565,
-      "required_pages_per_second_upper_estimate": 0.3704509160501488
+      "required_token_positions_per_second": 28975.656940454628,
+      "required_pages_per_second_upper_estimate": 0.3716686155956777
     }
   ],
   "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": 0.07806048853120728,
-    "producer_p95_seconds": 0.22106678300042404,
-    "gpu_update_seconds": 0.9573167254429881,
-    "producer_batches_per_second": 12.810578294039457,
-    "gpu_batches_per_second": 1.044586366687849,
-    "producer_headroom": 12.263780863480873,
+    "producer_batch_seconds": 0.09460591092964421,
+    "producer_p95_seconds": 0.22190766699895903,
+    "gpu_update_seconds": 0.954180264379062,
+    "producer_batches_per_second": 10.570164064523118,
+    "gpu_batches_per_second": 1.0480199992930639,
+    "producer_headroom": 10.085841941616728,
     "sustainable": true,
-    "buffer_seconds": 60.31095370290825,
+    "buffer_seconds": 60.11335665588091,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 553468.3023232229,
-    "upload_bytes_per_second": 3427803.128800917,
+    "generation_bytes_per_second": 549530.5491955869,
+    "upload_bytes_per_second": 3399611.954917522,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.76141359589114,
+    "estimated_full_shards_per_hour": 117.91646344090181,
     "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": 1.271395830000074,
-      "bytes": 6214888,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4851975340006902,
       "bytes": 6638444,
@@ -562,6 +556,12 @@
       "bytes": 5891309,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.286470338000072,
+      "bytes": 2778002,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -150,3 +150,4 @@
 {"step": 857500, "loss": 0.48385173082351685, "lr": 3e-05, "grad_norm": 0.2918216930045562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125513973059, "tokens_per_second": 151250.70750019568, "data_wait_s": 8.109000191325322e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 857600, "loss": 0.4771803617477417, "lr": 3e-05, "grad_norm": 0.2943646493956219, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125528615300, "tokens_per_second": 152305.05674595098, "data_wait_s": 1.841400080593303e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 857700, "loss": 0.5434330105781555, "lr": 3e-05, "grad_norm": 0.3058737382274146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125543257099, "tokens_per_second": 152213.44510028054, "data_wait_s": 1.3621000107377768e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 857800, "loss": 0.5173053741455078, "lr": 3e-05, "grad_norm": 0.30637968074020516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125557762639, "tokens_per_second": 149577.61291883624, "data_wait_s": 1.5612000424880534e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b407a77e5c38390d71a02ea53e368834b0997fb9cc123dde542c86ee2070ae4b
-size 324450
+oid sha256:40e71145da8eab81b623f0dcb9c5815c0f3d1ad323fe51a3642db0d93836e875
+size 326568