Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74400.35879107245,
-      "required_pages_per_second_upper_estimate": 0.006448834534024757
+      "required_token_positions_per_second": 73774.50273072516,
+      "required_pages_per_second_upper_estimate": 0.0063945869169315375
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9300.044848884056,
-      "required_pages_per_second_upper_estimate": 0.0037099269382814964
+      "required_token_positions_per_second": 9221.812841340645,
+      "required_pages_per_second_upper_estimate": 0.0036787190207996827
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18600.08969776811,
-      "required_pages_per_second_upper_estimate": 0.004302803401180376
+      "required_token_positions_per_second": 18443.62568268129,
+      "required_pages_per_second_upper_estimate": 0.004266608204962656
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27900.134546652167,
-      "required_pages_per_second_upper_estimate": 0.0016989718131980467
+      "required_token_positions_per_second": 27665.438524021938,
+      "required_pages_per_second_upper_estimate": 0.0016846800567747325
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27900.134546652167,
-      "required_pages_per_second_upper_estimate": 0.00214782958991681
+      "required_token_positions_per_second": 27665.438524021938,
+      "required_pages_per_second_upper_estimate": 0.0021297620404145626
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27900.134546652167,
-      "required_pages_per_second_upper_estimate": 0.3464262953257778
+      "required_token_positions_per_second": 27665.438524021938,
+      "required_pages_per_second_upper_estimate": 0.34351215620176984
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10711294135941074,
-    "producer_p95_seconds": 0.19747575099972892,
-    "gpu_update_seconds": 0.9909629630555878,
-    "producer_batches_per_second": 9.335940058303159,
-    "gpu_batches_per_second": 1.009119449748704,
-    "producer_headroom": 9.251570823085457,
+    "producer_batch_seconds": 0.4828587816640564,
+    "producer_p95_seconds": 0.1952516729998024,
+    "gpu_update_seconds": 0.9993696639217631,
+    "producer_batches_per_second": 2.0709988882334107,
+    "gpu_batches_per_second": 1.0006307336524138,
+    "producer_headroom": 2.0696934629161685,
     "sustainable": true,
-    "buffer_seconds": 60.44874074639086,
+    "buffer_seconds": 60.961549499227544,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319778.97969109664,
+    "generation_bytes_per_second": 292768.5228058975,
     "upload_bytes_per_second": 4986153.199748853,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.6171249680488,
+    "estimated_full_shards_per_hour": 62.821309691740936,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -60,3 +60,4 @@
 {"step": 201000, "loss": 0.5869835671037436, "lr": 0.0003, "grad_norm": 0.1414334957624636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29418521976, "tokens_per_second": 145139.11236632828, "data_wait_s": 1.3099000170768704e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 201100, "loss": 0.6412553377449512, "lr": 0.0003, "grad_norm": 0.15963036641565348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29433097753, "tokens_per_second": 148173.20644179365, "data_wait_s": 1.4231000022846274e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 201200, "loss": 0.5985647439956665, "lr": 0.0003, "grad_norm": 0.1404202134853768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29447861160, "tokens_per_second": 148901.12254889208, "data_wait_s": 1.8498000827094074e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 201300, "loss": 0.5749208759516478, "lr": 0.0003, "grad_norm": 0.1506905842714099, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29462379900, "tokens_per_second": 147481.41415537766, "data_wait_s": 1.6646000403852668e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9718885e62497a3109d03b88590f5091f634c0c83a7e3d6d97c40db6c3152d1
-size 132264
+oid sha256:59f10b620eafecfcabb777d9622d2b9d941e60ef8eae5a1486e522d94157e55e
+size 134382