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": 75370.88600524074,
-      "required_pages_per_second_upper_estimate": 0.006532957373170129
+      "required_token_positions_per_second": 75066.76861649213,
+      "required_pages_per_second_upper_estimate": 0.0065065972486918755
     },
     {
       "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": 9421.360750655092,
-      "required_pages_per_second_upper_estimate": 0.003758321665332333
+      "required_token_positions_per_second": 9383.346077061517,
+      "required_pages_per_second_upper_estimate": 0.0037431570436658357
     },
     {
       "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": 18842.721501310185,
-      "required_pages_per_second_upper_estimate": 0.004358931998755937
+      "required_token_positions_per_second": 18766.692154123033,
+      "required_pages_per_second_upper_estimate": 0.004341343947354999
     },
     {
       "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": 28264.082251965276,
-      "required_pages_per_second_upper_estimate": 0.0017211343189655782
+      "required_token_positions_per_second": 28150.038231184553,
+      "required_pages_per_second_upper_estimate": 0.0017141896364427668
     },
     {
       "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": 28264.082251965276,
-      "required_pages_per_second_upper_estimate": 0.0021758472917435424
+      "required_token_positions_per_second": 28150.038231184553,
+      "required_pages_per_second_upper_estimate": 0.0021670678673297874
     },
     {
       "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": 28264.082251965276,
-      "required_pages_per_second_upper_estimate": 0.35094530777115207
+      "required_token_positions_per_second": 28150.038231184553,
+      "required_pages_per_second_upper_estimate": 0.34952926271384027
     }
   ],
   "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": 0.1257512091639228,
-    "producer_p95_seconds": 0.3557953109993832,
-    "gpu_update_seconds": 0.9782026443854395,
-    "producer_batches_per_second": 7.952209816896882,
-    "gpu_batches_per_second": 1.0222830675624015,
-    "producer_headroom": 7.778872671596381,
+    "producer_batch_seconds": 0.11325992539838126,
+    "producer_p95_seconds": 0.37669194000045536,
+    "gpu_update_seconds": 0.9821656287973209,
+    "producer_batches_per_second": 8.829248266609685,
+    "gpu_batches_per_second": 1.0181582114867098,
+    "producer_headroom": 8.671784175582358,
     "sustainable": true,
-    "buffer_seconds": 60.648563951897245,
+    "buffer_seconds": 60.8942689854339,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387927.91475817043,
-    "upload_bytes_per_second": 4221895.326348876,
+    "generation_bytes_per_second": 388414.49686122756,
+    "upload_bytes_per_second": 4165866.0444432665,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.24030000742755,
+    "estimated_full_shards_per_hour": 83.34470919969196,
     "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": 0.8321514729977935,
-      "bytes": 6044525,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4008962299994892,
       "bytes": 7557731,
@@ -562,6 +556,12 @@
       "bytes": 2695308,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.7943546639944543,
+      "bytes": 258770,
+      "pages": 8,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -348,3 +348,4 @@
 {"step": 34800, "loss": 0.6325910575687885, "lr": 0.0003, "grad_norm": 0.14124464919473864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5092382626, "tokens_per_second": 149926.7043873406, "data_wait_s": 1.4612000086344779e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 34900, "loss": 0.6349800880998373, "lr": 0.0003, "grad_norm": 0.138709396061437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5107014641, "tokens_per_second": 149005.7573006429, "data_wait_s": 1.1950003681704402e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 35000, "loss": 0.6313854027539492, "lr": 0.0003, "grad_norm": 0.12737554729872055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5121453079, "tokens_per_second": 147427.45620340458, "data_wait_s": 1.8901002476923168e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 35100, "loss": 0.6899380572140217, "lr": 0.0003, "grad_norm": 0.15498536011013583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5136085480, "tokens_per_second": 122253.15101073103, "data_wait_s": 1.4624005416408181e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04af29cbe0f0f099b49fbe191490434e9961f8daeba60c3f1f21168794792f36
-size 741172
+oid sha256:b6bf2e8a0eb6e37695a2aecf53d403167f00ea7f425870ac63033cc9bfa7d25a
+size 743290