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": 77269.59157608017,
-      "required_pages_per_second_upper_estimate": 0.005345689635387046
+      "required_token_positions_per_second": 77210.35887805611,
+      "required_pages_per_second_upper_estimate": 0.005341591779898946
     },
     {
       "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": 9658.698947010022,
-      "required_pages_per_second_upper_estimate": 0.004043895552543401
+      "required_token_positions_per_second": 9651.294859757014,
+      "required_pages_per_second_upper_estimate": 0.00404079561582549
     },
     {
       "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": 19317.397894020043,
-      "required_pages_per_second_upper_estimate": 0.004023304547283954
+      "required_token_positions_per_second": 19302.589719514028,
+      "required_pages_per_second_upper_estimate": 0.004020220395052174
     },
     {
       "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": 28976.096841030067,
-      "required_pages_per_second_upper_estimate": 0.0017370727616816833
+      "required_token_positions_per_second": 28953.88457927104,
+      "required_pages_per_second_upper_estimate": 0.0017357411705053898
     },
     {
       "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": 28976.096841030067,
-      "required_pages_per_second_upper_estimate": 0.002252764077009667
+      "required_token_positions_per_second": 28953.88457927104,
+      "required_pages_per_second_upper_estimate": 0.0022510371713593165
     },
     {
       "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": 28976.096841030067,
-      "required_pages_per_second_upper_estimate": 0.371674258167931
+      "required_token_positions_per_second": 28953.88457927104,
+      "required_pages_per_second_upper_estimate": 0.3713893431237547
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07851436600779493,
-    "producer_p95_seconds": 0.1524199979999139,
-    "gpu_update_seconds": 0.9541657784926545,
-    "producer_batches_per_second": 12.73652263715305,
-    "gpu_batches_per_second": 1.0480359100488306,
-    "producer_headroom": 12.152754037368455,
+    "producer_batch_seconds": 0.06822611565627312,
+    "producer_p95_seconds": 0.1532281109998621,
+    "gpu_update_seconds": 0.9548977763002494,
+    "producer_batches_per_second": 14.657143974574991,
+    "gpu_batches_per_second": 1.0472325151646065,
+    "producer_headroom": 13.996074188234257,
     "sustainable": true,
-    "buffer_seconds": 60.112444045037236,
+    "buffer_seconds": 60.15855990691571,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 70,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364291.9669122501,
-    "upload_bytes_per_second": 5331001.0046273405,
+    "generation_bytes_per_second": 393878.7480437419,
+    "upload_bytes_per_second": 5452637.346098469,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.16857581639889,
+    "estimated_full_shards_per_hour": 84.51721030220216,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 2637170,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9096160310000414,
+      "bytes": 6003921,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -11,3 +11,4 @@
 {"step": 486100, "loss": 0.5828445553779602, "lr": 0.0003, "grad_norm": 0.18576490721476466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71148753587, "tokens_per_second": 152251.18830952255, "data_wait_s": 1.3674999991053483e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 486200, "loss": 0.597375214099884, "lr": 0.0003, "grad_norm": 0.1392577956125053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71163414535, "tokens_per_second": 152562.46730330552, "data_wait_s": 1.1869999980262946e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 486300, "loss": 0.5967402458190918, "lr": 0.0003, "grad_norm": 0.15727533062311258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71178073082, "tokens_per_second": 152497.0445122778, "data_wait_s": 8.466000053886091e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 486400, "loss": 0.6144828796386719, "lr": 0.0003, "grad_norm": 0.15597788152546366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71192908927, "tokens_per_second": 153212.1731888354, "data_wait_s": 9.916999260894954e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec2efcbba7b7727f83655e6d2af78b4576857c40cdf6d7cd08aba4e95c5b8473
-size 27699
+oid sha256:1bf4f88b74c37e22b178235cae4465239107141d15ad7c298784bed0fe3fd199
+size 29817