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": 77000.35479099842,
-      "required_pages_per_second_upper_estimate": 0.0053270632098796795
+      "required_token_positions_per_second": 77191.41736016082,
+      "required_pages_per_second_upper_estimate": 0.005340281361740573
     },
     {
       "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": 9625.044348874802,
-      "required_pages_per_second_upper_estimate": 0.0040298050750921105
+      "required_token_positions_per_second": 9648.927170020102,
+      "required_pages_per_second_upper_estimate": 0.004039804313575629
     },
     {
       "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": 19250.088697749605,
-      "required_pages_per_second_upper_estimate": 0.004009285816763695
+      "required_token_positions_per_second": 19297.854340040205,
+      "required_pages_per_second_upper_estimate": 0.004019234140388132
     },
     {
       "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": 28875.133046624407,
-      "required_pages_per_second_upper_estimate": 0.0017310201363698518
+      "required_token_positions_per_second": 28946.78151006031,
+      "required_pages_per_second_upper_estimate": 0.0017353153523519652
     },
     {
       "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": 28875.133046624407,
-      "required_pages_per_second_upper_estimate": 0.0022449145860874253
+      "required_token_positions_per_second": 28946.78151006031,
+      "required_pages_per_second_upper_estimate": 0.002250484938971282
     },
     {
       "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": 28875.133046624407,
-      "required_pages_per_second_upper_estimate": 0.3703792030197715
+      "required_token_positions_per_second": 28946.78151006031,
+      "required_pages_per_second_upper_estimate": 0.371298232578601
     }
   ],
   "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": 0.4125020222109512,
-    "producer_p95_seconds": 0.16129175200057944,
-    "gpu_update_seconds": 0.9575020816477983,
-    "producer_batches_per_second": 2.424230539865343,
-    "gpu_batches_per_second": 1.044384152438672,
-    "producer_headroom": 2.3212057883152317,
+    "producer_batch_seconds": 0.07801937187499419,
+    "producer_p95_seconds": 0.16368848699949012,
+    "gpu_update_seconds": 0.9551320926781126,
+    "producer_batches_per_second": 12.817329542235237,
+    "gpu_batches_per_second": 1.0469756043858618,
+    "producer_headroom": 12.242242788220137,
     "sustainable": true,
-    "buffer_seconds": 60.32263114381129,
+    "buffer_seconds": 60.173321838721094,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353132.63563590817,
-    "upload_bytes_per_second": 3526690.8116346407,
+    "generation_bytes_per_second": 370128.3663446542,
+    "upload_bytes_per_second": 3103901.075394054,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.77404310043272,
+    "estimated_full_shards_per_hour": 79.42093127016753,
     "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": [
@@ -250,6 +250,12 @@
       "bytes": 7490143,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 4.696426351999435,
+      "bytes": 5652066,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -54,3 +54,4 @@
 {"step": 930400, "loss": 0.5374108552932739, "lr": 3e-05, "grad_norm": 0.3393927115605374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136184392468, "tokens_per_second": 151769.74928183996, "data_wait_s": 1.6635000065434724e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 930500, "loss": 0.5489927530288696, "lr": 3e-05, "grad_norm": 0.34256797904167613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136198959130, "tokens_per_second": 151590.67135968982, "data_wait_s": 1.4008000107423868e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 930600, "loss": 0.48753622174263, "lr": 3e-05, "grad_norm": 0.32634271697262146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136213457104, "tokens_per_second": 151309.8299642593, "data_wait_s": 1.1969000297540333e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 930700, "loss": 0.5303984880447388, "lr": 3e-05, "grad_norm": 0.3506626336186566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136228192909, "tokens_per_second": 153538.9367800053, "data_wait_s": 1.3176999345887452e-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 sha256:88eb0b64943874809f07ac796d87638f8395a359a6f52db1d3b74b482b5ae0fe
-size 119556
+oid sha256:4aa347b2d314ba6b283ba5201cd86e8c4144d8fcbdea6cf5bdbb41dbd4d2ab4b
+size 121674