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": 76673.10835950509,
-      "required_pages_per_second_upper_estimate": 0.005304423542432636
+      "required_token_positions_per_second": 77150.97845171661,
+      "required_pages_per_second_upper_estimate": 0.005337483704223208
     },
     {
       "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": 9584.138544938136,
-      "required_pages_per_second_upper_estimate": 0.00401267866919415
+      "required_token_positions_per_second": 9643.872306464576,
+      "required_pages_per_second_upper_estimate": 0.0040376879477624856
     },
     {
       "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": 19168.277089876272,
-      "required_pages_per_second_upper_estimate": 0.0039922466163608665
+      "required_token_positions_per_second": 19287.744612929153,
+      "required_pages_per_second_upper_estimate": 0.004017128550842332
     },
     {
       "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": 28752.415634814406,
-      "required_pages_per_second_upper_estimate": 0.0017236634149104806
+      "required_token_positions_per_second": 28931.61691939373,
+      "required_pages_per_second_upper_estimate": 0.0017344062582964929
     },
     {
       "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": 28752.415634814406,
-      "required_pages_per_second_upper_estimate": 0.0022353738470960467
+      "required_token_positions_per_second": 28931.61691939373,
+      "required_pages_per_second_upper_estimate": 0.0022493059587489405
     },
     {
       "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": 28752.415634814406,
-      "required_pages_per_second_upper_estimate": 0.3688051158247637
+      "required_token_positions_per_second": 28931.61691939373,
+      "required_pages_per_second_upper_estimate": 0.37110371749199894
     }
   ],
   "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": 0.0768606520313142,
-    "producer_p95_seconds": 0.19771702500111132,
-    "gpu_update_seconds": 0.961588770528305,
-    "producer_batches_per_second": 13.010558375078379,
-    "gpu_batches_per_second": 1.0399455886434608,
-    "producer_headroom": 12.51080683177836,
+    "producer_batch_seconds": 0.15616249915633773,
+    "producer_p95_seconds": 0.22872485700099787,
+    "gpu_update_seconds": 0.9556327279263371,
+    "producer_batches_per_second": 6.403586042759715,
+    "gpu_batches_per_second": 1.0464271165868682,
+    "producer_headroom": 6.119476398553484,
     "sustainable": true,
-    "buffer_seconds": 60.580092543283214,
+    "buffer_seconds": 60.20486185935924,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 253704.16431534512,
-    "upload_bytes_per_second": 4814753.355319817,
+    "generation_bytes_per_second": 270004.06490133156,
+    "upload_bytes_per_second": 4824294.805783559,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.43900773139253,
+    "estimated_full_shards_per_hour": 57.93658695487938,
     "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": 1.452950692000286,
-      "bytes": 7702767,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.831108772999869,
       "bytes": 11417123,
@@ -562,6 +556,12 @@
       "bytes": 1101377,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.815276637998977,
+      "bytes": 10307802,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -111,3 +111,4 @@
 {"step": 853600, "loss": 0.5497733354568481, "lr": 3e-05, "grad_norm": 0.2856565265019204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124943209301, "tokens_per_second": 153404.2340199271, "data_wait_s": 1.5001000065240078e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 853700, "loss": 0.5567955374717712, "lr": 3e-05, "grad_norm": 0.30432902070128004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124957907790, "tokens_per_second": 152826.08968260873, "data_wait_s": 1.6863999917404726e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 853800, "loss": 0.5145642757415771, "lr": 3e-05, "grad_norm": 0.2945418334246771, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124972555962, "tokens_per_second": 152459.7652867361, "data_wait_s": 1.4790999557590112e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 853900, "loss": 0.5904368758201599, "lr": 3e-05, "grad_norm": 0.3303986266848577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124987376595, "tokens_per_second": 154703.17878011853, "data_wait_s": 1.7057000150089152e-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 sha256:8bda1da4e1779f91be50b8352349d51813aab271a30d93800ae6e1f1b28d150b
-size 241065
+oid sha256:de59349249c385534011b213f929e82c3d1aec0cd53759eac8264e8c2c4232ad
+size 243183