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": 76732.84183893666,
-      "required_pages_per_second_upper_estimate": 0.00530855604314048
+      "required_token_positions_per_second": 76687.83797674823,
+      "required_pages_per_second_upper_estimate": 0.005305442571530938
     },
     {
       "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": 9591.605229867082,
-      "required_pages_per_second_upper_estimate": 0.004015804814251788
+      "required_token_positions_per_second": 9585.979747093528,
+      "required_pages_per_second_upper_estimate": 0.004013449542087939
     },
     {
       "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": 19183.210459734164,
-      "required_pages_per_second_upper_estimate": 0.0039953568434828194
+      "required_token_positions_per_second": 19171.959494187056,
+      "required_pages_per_second_upper_estimate": 0.0039930135640672704
     },
     {
       "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": 28774.815689601248,
-      "required_pages_per_second_upper_estimate": 0.0017250062639920483
+      "required_token_positions_per_second": 28757.939241280583,
+      "required_pages_per_second_upper_estimate": 0.001723994546684069
     },
     {
       "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": 28774.815689601248,
-      "required_pages_per_second_upper_estimate": 0.0022371153528282952
+      "required_token_positions_per_second": 28757.939241280583,
+      "required_pages_per_second_upper_estimate": 0.0022358032832030146
     },
     {
       "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": 28774.815689601248,
-      "required_pages_per_second_upper_estimate": 0.36909243967626437
+      "required_token_positions_per_second": 28757.939241280583,
+      "required_pages_per_second_upper_estimate": 0.3688759667177253
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05567494983593235,
-    "producer_p95_seconds": 0.16137791900018783,
-    "gpu_update_seconds": 0.9608402117408362,
-    "producer_batches_per_second": 17.961399210001705,
-    "gpu_batches_per_second": 1.0407557758102304,
-    "producer_headroom": 17.25803462009973,
+    "producer_batch_seconds": 0.04810483792187625,
+    "producer_p95_seconds": 0.13598946900037845,
+    "gpu_update_seconds": 0.961404075863429,
+    "producer_batches_per_second": 20.787929929709584,
+    "gpu_batches_per_second": 1.040145371863447,
+    "producer_headroom": 19.98560056318616,
     "sustainable": true,
-    "buffer_seconds": 60.53293333967268,
+    "buffer_seconds": 60.56845677939603,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -26,3 +26,4 @@
 {"step": 802600, "loss": 0.5096468329429626, "lr": 3e-05, "grad_norm": 0.279247773655295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117477393048, "tokens_per_second": 154675.89587132793, "data_wait_s": 1.648699981160462e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 802700, "loss": 0.5636854767799377, "lr": 3e-05, "grad_norm": 0.2700714123268359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117492060387, "tokens_per_second": 152633.89581679864, "data_wait_s": 1.2536999747680966e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 802800, "loss": 0.5458204746246338, "lr": 3e-05, "grad_norm": 0.2682161863187492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117506774323, "tokens_per_second": 153275.7391610355, "data_wait_s": 1.613099993846845e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 802900, "loss": 0.5392683744430542, "lr": 3e-05, "grad_norm": 0.28495113546919615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117521292308, "tokens_per_second": 150503.1734724028, "data_wait_s": 1.4216999261407182e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e1d76eacd8eabd5fc313de100707741fd55414874bc84db01539c4969cce131
-size 59469
+oid sha256:003722fec77d6532449d219aa9d5fb752edd76672e064aaa1adc41fe03859cae
+size 61587