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": 77322.99613092658,
-      "required_pages_per_second_upper_estimate": 0.005349384286406959
+      "required_token_positions_per_second": 77210.52630347686,
+      "required_pages_per_second_upper_estimate": 0.005341603362778034
     },
     {
       "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": 9665.374516365822,
-      "required_pages_per_second_upper_estimate": 0.004046690474030935
+      "required_token_positions_per_second": 9651.315787934607,
+      "required_pages_per_second_upper_estimate": 0.004040804378016419
     },
     {
       "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": 19330.749032731645,
-      "required_pages_per_second_upper_estimate": 0.004026085237384376
+      "required_token_positions_per_second": 19302.631575869214,
+      "required_pages_per_second_upper_estimate": 0.0040202291126271335
     },
     {
       "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": 28996.12354909747,
-      "required_pages_per_second_upper_estimate": 0.0017382733322513116
+      "required_token_positions_per_second": 28953.94736380382,
+      "required_pages_per_second_upper_estimate": 0.0017357449343422635
     },
     {
       "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": 28996.12354909747,
-      "required_pages_per_second_upper_estimate": 0.002254321065473728
+      "required_token_positions_per_second": 28953.94736380382,
+      "required_pages_per_second_upper_estimate": 0.0022510420525805805
     },
     {
       "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": 28996.12354909747,
-      "required_pages_per_second_upper_estimate": 0.3719311392760158
+      "required_token_positions_per_second": 28953.94736380382,
+      "required_pages_per_second_upper_estimate": 0.37139014845632845
     }
   ],
   "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": 0.10271904008544652,
-    "producer_p95_seconds": 0.3049558559941943,
-    "gpu_update_seconds": 0.9535067662815422,
-    "producer_batches_per_second": 9.735293468164743,
-    "gpu_batches_per_second": 1.0487602556820554,
-    "producer_headroom": 9.282668193631585,
+    "producer_batch_seconds": 0.12239163589083546,
+    "producer_p95_seconds": 0.3230817730000126,
+    "gpu_update_seconds": 0.9548957056737478,
+    "producer_batches_per_second": 8.170492964829133,
+    "gpu_batches_per_second": 1.04723478601721,
+    "producer_headroom": 7.801968645352908,
     "sustainable": true,
-    "buffer_seconds": 60.07092627573716,
+    "buffer_seconds": 60.15842945744611,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 41760.58536463547,
-    "upload_bytes_per_second": 5204090.962073306,
+    "generation_bytes_per_second": 38887.15004791474,
+    "upload_bytes_per_second": 5158491.84100879,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 8.960849482577306,
+    "estimated_full_shards_per_hour": 8.344277153759782,
     "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": 1.428008876999229,
-      "bytes": 7759017,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.8940000050024537,
       "bytes": 393500,
@@ -562,6 +556,12 @@
       "bytes": 5993179,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.1118833660002565,
+      "bytes": 2568385,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -327,3 +327,4 @@
 {"step": 717700, "loss": 0.5750818848609924, "lr": 8e-05, "grad_norm": 0.21498823118512472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105051626347, "tokens_per_second": 152844.6147458412, "data_wait_s": 9.174000297207385e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 717800, "loss": 0.6104263663291931, "lr": 8e-05, "grad_norm": 0.27834772034823047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105066320787, "tokens_per_second": 152397.20697045236, "data_wait_s": 1.7195001419167966e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 717900, "loss": 0.551626443862915, "lr": 8e-05, "grad_norm": 0.2137054670401659, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105080967918, "tokens_per_second": 151987.31471702774, "data_wait_s": 1.6012003470677882e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 718000, "loss": 0.5673214793205261, "lr": 8e-05, "grad_norm": 0.2272528752780872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105095720485, "tokens_per_second": 152871.38338677492, "data_wait_s": 1.561100361868739e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b81b15375ecc09cfa67baf95f4881d9519e55ba1150759f298c49f8ab51f5a4
-size 701685
+oid sha256:b26c552d4c0049298c09046f7f8b5b20f6938b0404bd35414640f1b95e052439
+size 703803