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": 77497.63938296225,
-      "required_pages_per_second_upper_estimate": 0.005361466511810961
+      "required_token_positions_per_second": 77535.96991890963,
+      "required_pages_per_second_upper_estimate": 0.005364118307226898
     },
     {
       "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": 9687.204922870282,
-      "required_pages_per_second_upper_estimate": 0.004055830409363625
+      "required_token_positions_per_second": 9691.996239863704,
+      "required_pages_per_second_upper_estimate": 0.004057836433734694
     },
     {
       "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": 19374.409845740563,
-      "required_pages_per_second_upper_estimate": 0.004035178633321065
+      "required_token_positions_per_second": 19383.992479727407,
+      "required_pages_per_second_upper_estimate": 0.0040371744432694725
     },
     {
       "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": 29061.614768610845,
-      "required_pages_per_second_upper_estimate": 0.0017421994308618363
+      "required_token_positions_per_second": 29075.98871959111,
+      "required_pages_per_second_upper_estimate": 0.001743061127275351
     },
     {
       "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": 29061.614768610845,
-      "required_pages_per_second_upper_estimate": 0.002259412719725462
+      "required_token_positions_per_second": 29075.98871959111,
+      "required_pages_per_second_upper_estimate": 0.002260530231189849
     },
     {
       "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": 29061.614768610845,
-      "required_pages_per_second_upper_estimate": 0.3727711903209405
+      "required_token_positions_per_second": 29075.98871959111,
+      "required_pages_per_second_upper_estimate": 0.3729555639305693
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05072662044528542,
-    "producer_p95_seconds": 0.14308032999997522,
-    "gpu_update_seconds": 0.9513580102184247,
-    "producer_batches_per_second": 19.71351513705938,
-    "gpu_batches_per_second": 1.0511290063878345,
-    "producer_headroom": 18.754610535203607,
+    "producer_batch_seconds": 0.058225461914073406,
+    "producer_p95_seconds": 0.14809365600012825,
+    "gpu_update_seconds": 0.9508876986656366,
+    "producer_batches_per_second": 17.174616862220113,
+    "gpu_batches_per_second": 1.0516488975546552,
+    "producer_headroom": 16.33113190358052,
     "sustainable": true,
-    "buffer_seconds": 60.88691265397918,
+    "buffer_seconds": 60.85681271460074,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -31,3 +31,4 @@
 {"step": 688100, "loss": 0.5432969927787781, "lr": 8e-05, "grad_norm": 0.19581578127697455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100719252292, "tokens_per_second": 152816.92337104827, "data_wait_s": 1.467199990656809e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688200, "loss": 0.510589599609375, "lr": 8e-05, "grad_norm": 0.20504630637480717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100733890275, "tokens_per_second": 153425.50161575255, "data_wait_s": 1.6199999663513154e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688300, "loss": 0.5151051878929138, "lr": 8e-05, "grad_norm": 0.18574555369965068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100748617566, "tokens_per_second": 154502.830518826, "data_wait_s": 1.603400005478761e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 688400, "loss": 0.5832341909408569, "lr": 8e-05, "grad_norm": 0.21190185625883837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100763397324, "tokens_per_second": 155012.97129702306, "data_wait_s": 1.5720999726909213e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bece2a62dfcf8242136e40d1ac5fb49978e25c86253026f6fb97f7857e8a0280
-size 70059
+oid sha256:cd89215821ff151331d662e0728cc45d484d973685637640de4f41081fbe8122
+size 72177