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": 76442.29644663556,
-      "required_pages_per_second_upper_estimate": 0.0052884554387429
+      "required_token_positions_per_second": 77007.57001319193,
+      "required_pages_per_second_upper_estimate": 0.005327562375692648
     },
     {
       "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": 9555.287055829445,
-      "required_pages_per_second_upper_estimate": 0.004000599153191945
+      "required_token_positions_per_second": 9625.946251648991,
+      "required_pages_per_second_upper_estimate": 0.0040301826829497915
     },
     {
       "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": 19110.57411165889,
-      "required_pages_per_second_upper_estimate": 0.003980228607728053
+      "required_token_positions_per_second": 19251.892503297982,
+      "required_pages_per_second_upper_estimate": 0.004009661501889871
     },
     {
       "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": 28665.861167488336,
-      "required_pages_per_second_upper_estimate": 0.0017184746067553018
+      "required_token_positions_per_second": 28877.838754946973,
+      "required_pages_per_second_upper_estimate": 0.0017311823394524123
     },
     {
       "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": 28665.861167488336,
-      "required_pages_per_second_upper_estimate": 0.0022286446179743116
+      "required_token_positions_per_second": 28877.838754946973,
+      "required_pages_per_second_upper_estimate": 0.0022451249430083513
     },
     {
       "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": 28665.861167488336,
-      "required_pages_per_second_upper_estimate": 0.3676948880528512
+      "required_token_positions_per_second": 28877.838754946973,
+      "required_pages_per_second_upper_estimate": 0.37041390894097015
     }
   ],
   "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": 0.14646717580473023,
-    "producer_p95_seconds": 0.33533277999958955,
-    "gpu_update_seconds": 0.9644922173612299,
-    "producer_batches_per_second": 6.827468301383774,
-    "gpu_batches_per_second": 1.0368150017176048,
-    "producer_headroom": 6.585040040965145,
+    "producer_batch_seconds": 0.38302429506245517,
+    "producer_p95_seconds": 0.33408495200274047,
+    "gpu_update_seconds": 0.9574123685161067,
+    "producer_batches_per_second": 2.6108004450133953,
+    "gpu_batches_per_second": 1.0444820151528853,
+    "producer_headroom": 2.49961263778318,
     "sustainable": true,
-    "buffer_seconds": 60.76300969375748,
+    "buffer_seconds": 60.31697921651472,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 426541.5815549491,
+    "generation_bytes_per_second": 387102.8323510913,
     "upload_bytes_per_second": 4085834.5620456794,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.52589402185778,
+    "estimated_full_shards_per_hour": 83.0632565298038,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -382,3 +382,4 @@
 {"step": 838200, "loss": 0.5123892426490784, "lr": 3e-05, "grad_norm": 0.29468094144911877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122688655383, "tokens_per_second": 152371.31204812843, "data_wait_s": 8.53499659569934e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 838300, "loss": 0.5473647713661194, "lr": 3e-05, "grad_norm": 0.3092895005641432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122703193023, "tokens_per_second": 150404.36564645864, "data_wait_s": 9.864997991826385e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 838400, "loss": 0.5197517275810242, "lr": 3e-05, "grad_norm": 0.2894518852575219, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122717666672, "tokens_per_second": 150872.9618805877, "data_wait_s": 1.113000325858593e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 838500, "loss": 0.5905106067657471, "lr": 3e-05, "grad_norm": 0.35634493423609104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122732311458, "tokens_per_second": 132077.6518711381, "data_wait_s": 7.971997547429055e-06, "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:e3e1987ec556ebbe3dbeda2d19c05dab33bda3a70e68165575ed9c6f012fd1cf
-size 818958
+oid sha256:6358df3186d8e87aed05099380ba8c4ec0b8af09dda63e46c879ff3c53400edd
+size 821076