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": 76558.20814949888,
-      "required_pages_per_second_upper_estimate": 0.005296474479299189
+      "required_token_positions_per_second": 77170.13682853756,
+      "required_pages_per_second_upper_estimate": 0.0053388091251852406
     },
     {
       "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": 9569.77601868736,
-      "required_pages_per_second_upper_estimate": 0.0040066653793766034
+      "required_token_positions_per_second": 9646.267103567196,
+      "required_pages_per_second_upper_estimate": 0.004038690599300301
     },
     {
       "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": 19139.55203737472,
-      "required_pages_per_second_upper_estimate": 0.00398626394545537
+      "required_token_positions_per_second": 19292.53420713439,
+      "required_pages_per_second_upper_estimate": 0.00401812609700519
     },
     {
       "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": 28709.328056062077,
-      "required_pages_per_second_upper_estimate": 0.0017210803803552018
+      "required_token_positions_per_second": 28938.801310701583,
+      "required_pages_per_second_upper_estimate": 0.0017348369515854682
     },
     {
       "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": 28709.328056062077,
-      "required_pages_per_second_upper_estimate": 0.002232023977370283
+      "required_token_positions_per_second": 28938.801310701583,
+      "required_pages_per_second_upper_estimate": 0.002249864513572328
     },
     {
       "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": 28709.328056062077,
-      "required_pages_per_second_upper_estimate": 0.36825243462836643
+      "required_token_positions_per_second": 28938.801310701583,
+      "required_pages_per_second_upper_estimate": 0.3711958711496977
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.4123389078439743,
-    "producer_p95_seconds": 0.25340694400074426,
-    "gpu_update_seconds": 0.9630319436947611,
-    "producer_batches_per_second": 2.425189524871109,
-    "gpu_batches_per_second": 1.0383871548054862,
-    "producer_headroom": 2.335534981964798,
+    "producer_batch_seconds": 0.08906326558602018,
+    "producer_p95_seconds": 0.25963005800076644,
+    "gpu_update_seconds": 0.955395481075982,
+    "producer_batches_per_second": 11.227973659175653,
+    "gpu_batches_per_second": 1.0466869687030376,
+    "producer_headroom": 10.727155295616578,
     "sustainable": true,
-    "buffer_seconds": 60.67101245276995,
+    "buffer_seconds": 60.18991530778687,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344339.4475213339,
-    "upload_bytes_per_second": 3823941.8079875074,
+    "generation_bytes_per_second": 358072.56606147625,
+    "upload_bytes_per_second": 3846490.20823898,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.88722962598813,
+    "estimated_full_shards_per_hour": 76.83403717406479,
     "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": 1.0505489870010933,
-      "bytes": 7688939,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.4415530749975005,
       "bytes": 9558241,
@@ -562,6 +556,12 @@
       "bytes": 5548542,
       "pages": 23,
       "shards": 2
+    },
+    {
+      "seconds": 1.0629937340017932,
+      "bytes": 10277378,
+      "pages": 44,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -272,3 +272,4 @@
 {"step": 552200, "loss": 0.5662043690681458, "lr": 0.0003, "grad_norm": 0.16982648571386338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80824575209, "tokens_per_second": 153258.87326916007, "data_wait_s": 1.0771000233944505e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 552300, "loss": 0.5912648439407349, "lr": 0.0003, "grad_norm": 0.14267824863647247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80839119291, "tokens_per_second": 150291.8241835844, "data_wait_s": 1.692500154604204e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 552400, "loss": 0.5595518350601196, "lr": 0.0003, "grad_norm": 0.1525354226985604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80853665610, "tokens_per_second": 151328.334717717, "data_wait_s": 8.303999493364245e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 552500, "loss": 0.6008196473121643, "lr": 0.0003, "grad_norm": 0.16624135242016908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80868153498, "tokens_per_second": 150446.95472104294, "data_wait_s": 1.3756998669123277e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4d47432ce03bca1456baa04667511e484ec015aed6ceed70fd6a0f1b65054b94
-size 584412
+oid sha256:9dde805e1902adf0ac94613321a962e7d4fe7c12b3d338425183928e2536740c
+size 586530