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": 76912.33336537452,
-      "required_pages_per_second_upper_estimate": 0.005320973683417172
+      "required_token_positions_per_second": 77390.67008774294,
+      "required_pages_per_second_upper_estimate": 0.005354066127764729
     },
     {
       "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": 9614.041670671815,
-      "required_pages_per_second_upper_estimate": 0.004025198483490568
+      "required_token_positions_per_second": 9673.833760967867,
+      "required_pages_per_second_upper_estimate": 0.004050232183096697
     },
     {
       "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": 19228.08334134363,
-      "required_pages_per_second_upper_estimate": 0.004004702681344605
+      "required_token_positions_per_second": 19347.667521935735,
+      "required_pages_per_second_upper_estimate": 0.004029608912515024
     },
     {
       "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": 28842.125012015447,
-      "required_pages_per_second_upper_estimate": 0.0017290413550953953
+      "required_token_positions_per_second": 29021.501282903606,
+      "required_pages_per_second_upper_estimate": 0.0017397946886434335
     },
     {
       "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": 28842.125012015447,
-      "required_pages_per_second_upper_estimate": 0.002242348356583579
+      "required_token_positions_per_second": 29021.501282903606,
+      "required_pages_per_second_upper_estimate": 0.002256294072652301
     },
     {
       "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": 28842.125012015447,
-      "required_pages_per_second_upper_estimate": 0.36995581139307404
+      "required_token_positions_per_second": 29021.501282903606,
+      "required_pages_per_second_upper_estimate": 0.37225665759679333
     }
   ],
   "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": 0.43612881853125884,
-    "producer_p95_seconds": 0.18114199600040592,
-    "gpu_update_seconds": 0.9585978837718101,
-    "producer_batches_per_second": 2.292900531929253,
-    "gpu_batches_per_second": 1.0431902854461605,
-    "producer_headroom": 2.1979695976066393,
+    "producer_batch_seconds": 0.07877160878904022,
+    "producer_p95_seconds": 0.17361144999995304,
+    "gpu_update_seconds": 0.9526729761663734,
+    "producer_batches_per_second": 12.694929243836057,
+    "gpu_batches_per_second": 1.0496781424661314,
+    "producer_headroom": 12.094116024946825,
     "sustainable": true,
-    "buffer_seconds": 60.39166667762404,
+    "buffer_seconds": 60.01839749848152,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321535.8727661735,
-    "upload_bytes_per_second": 3616827.686589382,
+    "generation_bytes_per_second": 320818.8324962328,
+    "upload_bytes_per_second": 3754908.962633108,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.99411332358268,
+    "estimated_full_shards_per_hour": 68.8402531734966,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 7296014,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.574737728999935,
+      "bytes": 10770405,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -64,3 +64,4 @@
 {"step": 931400, "loss": 0.5176578760147095, "lr": 3e-05, "grad_norm": 0.3709694984262481, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136330879043, "tokens_per_second": 152927.8336805708, "data_wait_s": 9.70699966273969e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 931500, "loss": 0.5382647514343262, "lr": 3e-05, "grad_norm": 0.3312104055239438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136345683501, "tokens_per_second": 154687.7342254308, "data_wait_s": 1.1297999662929215e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 931600, "loss": 0.5345258712768555, "lr": 3e-05, "grad_norm": 0.40260721277650025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136360299715, "tokens_per_second": 152914.35409240355, "data_wait_s": 9.089000741369091e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 931700, "loss": 0.5648506879806519, "lr": 3e-05, "grad_norm": 0.33845658496693376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136374738469, "tokens_per_second": 150816.21011358508, "data_wait_s": 1.3628000488097314e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:05cd84c0151a1dda448ec620569e422c97ac8fdb15adb1eb6ac34f184ad922b8
-size 140736
+oid sha256:08b95b6c26ca2548118642766aa13394b67e56247d0abbef931ab57acfe85a02
+size 142854