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": 76733.90965571793,
-      "required_pages_per_second_upper_estimate": 0.005308629917183084
+      "required_token_positions_per_second": 77317.72116003113,
+      "required_pages_per_second_upper_estimate": 0.005349019351680793
     },
     {
       "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": 9591.738706964741,
-      "required_pages_per_second_upper_estimate": 0.004015860698325259
+      "required_token_positions_per_second": 9664.715145003891,
+      "required_pages_per_second_upper_estimate": 0.0040464144090109335
     },
     {
       "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": 19183.477413929482,
-      "required_pages_per_second_upper_estimate": 0.003995412443001649
+      "required_token_positions_per_second": 19329.430290007782,
+      "required_pages_per_second_upper_estimate": 0.004025810578052579
     },
     {
       "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": 28775.216120894227,
-      "required_pages_per_second_upper_estimate": 0.0017250302692366393
+      "required_token_positions_per_second": 28994.145435011673,
+      "required_pages_per_second_upper_estimate": 0.0017381547473322733
     },
     {
       "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": 28775.216120894227,
-      "required_pages_per_second_upper_estimate": 0.0022371464846000135
+      "required_token_positions_per_second": 28994.145435011673,
+      "required_pages_per_second_upper_estimate": 0.0022541672758043616
     },
     {
       "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": 28775.216120894227,
-      "required_pages_per_second_upper_estimate": 0.3690975759789411
+      "required_token_positions_per_second": 28994.145435011673,
+      "required_pages_per_second_upper_estimate": 0.37190576615245663
     }
   ],
   "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": 0.4754746826874907,
-    "producer_p95_seconds": 0.34991208000064944,
-    "gpu_update_seconds": 0.9608268408425356,
-    "producer_batches_per_second": 2.10316140146048,
-    "gpu_batches_per_second": 1.0407702590022507,
-    "producer_headroom": 2.0207739251472328,
+    "producer_batch_seconds": 0.4391798182344928,
+    "producer_p95_seconds": 0.2805520309993881,
+    "gpu_update_seconds": 0.9535718189029243,
+    "producer_batches_per_second": 2.2769716605376127,
+    "gpu_batches_per_second": 1.048688709310318,
+    "producer_headroom": 2.171256007929263,
     "sustainable": true,
-    "buffer_seconds": 60.53209097307974,
+    "buffer_seconds": 60.07502459088423,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289426.91016294615,
-    "upload_bytes_per_second": 5452592.90564726,
+    "generation_bytes_per_second": 270351.5603075062,
+    "upload_bytes_per_second": 5452135.465033714,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.10427740732468,
+    "estimated_full_shards_per_hour": 58.01115137976541,
     "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": 1.200595816000714,
-      "bytes": 10503607,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.0193097230003332,
       "bytes": 2594227,
@@ -562,6 +556,12 @@
       "bytes": 667279,
       "pages": 18,
       "shards": 1
+    },
+    {
+      "seconds": 1.4021912890020758,
+      "bytes": 11567150,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -326,3 +326,4 @@
 {"step": 677600, "loss": 0.5918698906898499, "lr": 0.0001, "grad_norm": 0.2128557232345933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99183354464, "tokens_per_second": 152053.96163570735, "data_wait_s": 8.975999662652612e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 677700, "loss": 0.6165102124214172, "lr": 0.0001, "grad_norm": 0.21980467872222695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99197824231, "tokens_per_second": 134294.2795112557, "data_wait_s": 8.979000995168462e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 677800, "loss": 0.5002837777137756, "lr": 0.0001, "grad_norm": 0.23840045262852252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99212380014, "tokens_per_second": 133157.99911329185, "data_wait_s": 1.3387998478719965e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 677900, "loss": 0.5715132355690002, "lr": 0.0001, "grad_norm": 0.2154137197133443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99227007310, "tokens_per_second": 152083.80053568436, "data_wait_s": 1.0281997674610466e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:86113ea03b2f595a6538616c6281a52cc67768ce8048ffaeebaf829e3641018e
-size 699567
+oid sha256:b8fa50c8940facf70272cd30ade71b38cf339c43f1d46e133545ae0c27b6a5cc
+size 701685