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": 77344.65022357054,
-      "required_pages_per_second_upper_estimate": 0.005350882366780486
+      "required_token_positions_per_second": 76992.14477440978,
+      "required_pages_per_second_upper_estimate": 0.0053264952218822145
     },
     {
       "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": 9668.081277946318,
-      "required_pages_per_second_upper_estimate": 0.004047823738581079
+      "required_token_positions_per_second": 9624.018096801223,
+      "required_pages_per_second_upper_estimate": 0.0040293754047794825
     },
     {
       "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": 19336.162555892635,
-      "required_pages_per_second_upper_estimate": 0.004027212731494604
+      "required_token_positions_per_second": 19248.036193602446,
+      "required_pages_per_second_upper_estimate": 0.004008858334278016
     },
     {
       "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": 29004.24383383895,
-      "required_pages_per_second_upper_estimate": 0.0017387601309226065
+      "required_token_positions_per_second": 28872.054290403667,
+      "required_pages_per_second_upper_estimate": 0.001730835569635407
     },
     {
       "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": 29004.24383383895,
-      "required_pages_per_second_upper_estimate": 0.0022549523818950206
+      "required_token_positions_per_second": 28872.054290403667,
+      "required_pages_per_second_upper_estimate": 0.00224467522633328
     },
     {
       "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": 29004.24383383895,
-      "required_pages_per_second_upper_estimate": 0.3720352975697971
+      "required_token_positions_per_second": 28872.054290403667,
+      "required_pages_per_second_upper_estimate": 0.3703397120406827
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1106320823202509,
-    "producer_p95_seconds": 0.22376580000127433,
-    "gpu_update_seconds": 0.9532398140903561,
-    "producer_batches_per_second": 9.038969338977656,
-    "gpu_batches_per_second": 1.0490539581104945,
-    "producer_headroom": 8.61630545225549,
+    "producer_batch_seconds": 0.0670711567578337,
+    "producer_p95_seconds": 0.2072417340004904,
+    "gpu_update_seconds": 0.9576041843752521,
+    "producer_batches_per_second": 14.909538590643184,
+    "gpu_batches_per_second": 1.044272796961938,
+    "producer_headroom": 14.277436541504212,
     "sustainable": true,
-    "buffer_seconds": 60.05410828769243,
+    "buffer_seconds": 60.329063615640884,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 171,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 453688.19027739455,
+    "generation_bytes_per_second": 588983.2882178351,
     "upload_bytes_per_second": 5199576.169823156,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.35092431298615,
+    "estimated_full_shards_per_hour": 63.19105141115804,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -157,3 +157,4 @@
 {"step": 620100, "loss": 0.6086562871932983, "lr": 0.0001, "grad_norm": 0.1629300007082568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90765923478, "tokens_per_second": 141751.49508206395, "data_wait_s": 1.1692000043694861e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 620200, "loss": 0.542997419834137, "lr": 0.0001, "grad_norm": 0.15714926735442414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90780713195, "tokens_per_second": 154154.6075574205, "data_wait_s": 9.972998668672517e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 620300, "loss": 0.5748096108436584, "lr": 0.0001, "grad_norm": 0.1677302391892716, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90795426171, "tokens_per_second": 153378.42680677114, "data_wait_s": 1.5035000615171157e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 620400, "loss": 0.5376937389373779, "lr": 0.0001, "grad_norm": 0.15271596834778892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90810021604, "tokens_per_second": 151524.20968919215, "data_wait_s": 1.373899976897519e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a2738741e5dbea72da91522e5fed52c77eadb9694861a84887eb1925caf82eaf
-size 340059
+oid sha256:7fa80d013849a5ef340869a0c8514812e395059f13188dca16340fb0bccce8eb
+size 342177