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": 76537.65259061333,
-      "required_pages_per_second_upper_estimate": 0.005295052398040018
+      "required_token_positions_per_second": 75977.85097013613,
+      "required_pages_per_second_upper_estimate": 0.005256324023016691
     },
     {
       "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": 9567.206573826666,
-      "required_pages_per_second_upper_estimate": 0.004005589606469542
+      "required_token_positions_per_second": 9497.231371267017,
+      "required_pages_per_second_upper_estimate": 0.003976292450406209
     },
     {
       "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": 19134.413147653333,
-      "required_pages_per_second_upper_estimate": 0.00398519365024804
+      "required_token_positions_per_second": 18994.462742534033,
+      "required_pages_per_second_upper_estimate": 0.003956045671602064
     },
     {
       "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": 28701.619721479998,
-      "required_pages_per_second_upper_estimate": 0.0017206182774669515
+      "required_token_positions_per_second": 28491.69411380105,
+      "required_pages_per_second_upper_estimate": 0.0017080335578244419
     },
     {
       "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": 28701.619721479998,
-      "required_pages_per_second_upper_estimate": 0.0022314246882619067
+      "required_token_positions_per_second": 28491.69411380105,
+      "required_pages_per_second_upper_estimate": 0.0022151038956300328
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28701.619721479998,
-      "required_pages_per_second_upper_estimate": 0.3681535603889124
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28491.69411380105,
+      "required_pages_per_second_upper_estimate": 0.3654608600941631
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05992200722656449,
-    "producer_p95_seconds": 0.13660705500001313,
-    "gpu_update_seconds": 0.9632905831899278,
-    "producer_batches_per_second": 16.688359524056168,
-    "gpu_batches_per_second": 1.0381083521947336,
-    "producer_headroom": 16.075739578411252,
+    "producer_batch_seconds": 0.4955136434921723,
+    "producer_p95_seconds": 0.14089707499988435,
+    "gpu_update_seconds": 0.9703880678196536,
+    "producer_batches_per_second": 2.0181079030486817,
+    "gpu_batches_per_second": 1.0305155567781052,
+    "producer_headroom": 1.9583478286909832,
     "sustainable": true,
-    "buffer_seconds": 60.68730674096545,
+    "buffer_seconds": 60.164060204818526,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 64,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 465348.27414554404,
-    "upload_bytes_per_second": 3122923.41202882,
+    "generation_bytes_per_second": 379001.686412579,
+    "upload_bytes_per_second": 3225218.942903833,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.85290687823048,
+    "estimated_full_shards_per_hour": 81.32493919642474,
     "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": [
@@ -244,6 +244,12 @@
       "bytes": 6051595,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.509009006000042,
+      "bytes": 6599366,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -10,3 +10,4 @@
 {"step": 843500, "loss": 0.5259416103363037, "lr": 3e-05, "grad_norm": 0.2947109431887774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123464715616, "tokens_per_second": 151901.25280813174, "data_wait_s": 1.4360000022861641e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 843600, "loss": 0.5174587368965149, "lr": 3e-05, "grad_norm": 0.29567694041875386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123479457093, "tokens_per_second": 152771.79248379884, "data_wait_s": 1.0048000149254221e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 843700, "loss": 0.5038239359855652, "lr": 3e-05, "grad_norm": 0.2985374037328828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123494154041, "tokens_per_second": 130223.20923862039, "data_wait_s": 1.32210000174382e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 843800, "loss": 0.5686857104301453, "lr": 3e-05, "grad_norm": 0.3051568909555626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123508825676, "tokens_per_second": 151480.6816333096, "data_wait_s": 1.8688999944060924e-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:b4c28b15671aa669191e6dfa87bd78dc18cb1a9665e6ed172bcc0be6ec236292
-size 25581
+oid sha256:850c3936f7bd70cb71581ed2544a1c2e73a9395694c8121a41a9d6bf4be08067
+size 27699