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": 77601.60796538623,
-      "required_pages_per_second_upper_estimate": 0.005368659299583399
+      "required_token_positions_per_second": 77627.6672588882,
+      "required_pages_per_second_upper_estimate": 0.005370462141973759
     },
     {
       "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": 9700.200995673278,
-      "required_pages_per_second_upper_estimate": 0.004061271593657379
+      "required_token_positions_per_second": 9703.458407361026,
+      "required_pages_per_second_upper_estimate": 0.0040626354039085474
     },
     {
       "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": 19400.401991346556,
-      "required_pages_per_second_upper_estimate": 0.004040592111791819
+      "required_token_positions_per_second": 19406.91681472205,
+      "required_pages_per_second_upper_estimate": 0.004041948977693489
     },
     {
       "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": 29100.602987019833,
-      "required_pages_per_second_upper_estimate": 0.0017445367150238926
+      "required_token_positions_per_second": 29110.375222083076,
+      "required_pages_per_second_upper_estimate": 0.001745122545594595
     },
     {
       "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": 29100.602987019833,
-      "required_pages_per_second_upper_estimate": 0.0022624438822157118
+      "required_token_positions_per_second": 29110.375222083076,
+      "required_pages_per_second_upper_estimate": 0.0022632036305083596
     },
     {
       "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": 29100.602987019833,
-      "required_pages_per_second_upper_estimate": 0.3732712893244036
+      "required_token_positions_per_second": 29110.375222083076,
+      "required_pages_per_second_upper_estimate": 0.37339663706318643
     }
   ],
   "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": 0.04896854089063041,
-    "producer_p95_seconds": 0.13258478699981424,
-    "gpu_update_seconds": 0.9500834059119752,
-    "producer_batches_per_second": 20.42127418567497,
-    "gpu_batches_per_second": 1.052539170537465,
-    "producer_headroom": 19.401913731388372,
+    "producer_batch_seconds": 0.056710747101561765,
+    "producer_p95_seconds": 0.1607731250001052,
+    "gpu_update_seconds": 0.9497644667605067,
+    "producer_batches_per_second": 17.63334202261745,
+    "gpu_batches_per_second": 1.0528926223265,
+    "producer_headroom": 16.7475216833169,
     "sustainable": true,
-    "buffer_seconds": 60.80533797836641,
+    "buffer_seconds": 60.78492587267243,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "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

@@ -18,3 +18,4 @@
 {"step": 686800, "loss": 0.5589227080345154, "lr": 8e-05, "grad_norm": 0.19261947866849372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100529481032, "tokens_per_second": 153035.72695572837, "data_wait_s": 1.5040000107546803e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 686900, "loss": 0.5605591535568237, "lr": 8e-05, "grad_norm": 0.1983484123714947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100544126019, "tokens_per_second": 153805.36907094505, "data_wait_s": 1.56619998961105e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687000, "loss": 0.5661516189575195, "lr": 8e-05, "grad_norm": 0.2105792467816602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100558815263, "tokens_per_second": 154327.9849955852, "data_wait_s": 1.541499977975036e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 687100, "loss": 0.5114176273345947, "lr": 8e-05, "grad_norm": 0.2032273692054656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100573505138, "tokens_per_second": 154395.6483364498, "data_wait_s": 1.3889000001654495e-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:d9cf7c2c472662fcead53982510ec82803912c4dcc2586e807445528e76bc86e
-size 42525
+oid sha256:33b0e1022616ef2d8d08e10dc31b8afba0427c4be485419f1bfa08e8e0cf0136
+size 44643