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": 77389.26735345914,
-      "required_pages_per_second_upper_estimate": 0.005353969083352184
+      "required_token_positions_per_second": 76815.56582923593,
+      "required_pages_per_second_upper_estimate": 0.005314279080735486
     },
     {
       "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": 9673.658419182393,
-      "required_pages_per_second_upper_estimate": 0.004050158771152671
+      "required_token_positions_per_second": 9601.945728654491,
+      "required_pages_per_second_upper_estimate": 0.0040201341651599064
     },
     {
       "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": 19347.316838364786,
-      "required_pages_per_second_upper_estimate": 0.004029535874375343
+      "required_token_positions_per_second": 19203.891457308982,
+      "required_pages_per_second_upper_estimate": 0.003999664149883071
     },
     {
       "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": 29020.97525754718,
-      "required_pages_per_second_upper_estimate": 0.00173976315422651
+      "required_token_positions_per_second": 28805.83718596347,
+      "required_pages_per_second_upper_estimate": 0.0017268659553318815
     },
     {
       "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": 29020.97525754718,
-      "required_pages_per_second_upper_estimate": 0.00225625317649458
+      "required_token_positions_per_second": 28805.83718596347,
+      "required_pages_per_second_upper_estimate": 0.0022395271377213164
     },
     {
       "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": 29020.97525754718,
-      "required_pages_per_second_upper_estimate": 0.3722499103083231
+      "required_token_positions_per_second": 28805.83718596347,
+      "required_pages_per_second_upper_estimate": 0.3694903501233113
     }
   ],
   "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": 0.1185351254924285,
-    "producer_p95_seconds": 0.26313206200211425,
-    "gpu_update_seconds": 0.9526902440265124,
-    "producer_batches_per_second": 8.436317891812378,
-    "gpu_batches_per_second": 1.0496591166647562,
-    "producer_headroom": 8.037197751035967,
+    "producer_batch_seconds": 0.17917462486721547,
+    "producer_p95_seconds": 0.24442220100172563,
+    "gpu_update_seconds": 0.9598054665626012,
+    "producer_batches_per_second": 5.581147446191614,
+    "gpu_batches_per_second": 1.0418777917376836,
+    "producer_headroom": 5.356815828546612,
     "sustainable": true,
-    "buffer_seconds": 60.01948537367028,
+    "buffer_seconds": 60.46774439344388,
     "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": 270119.6276159086,
+    "generation_bytes_per_second": 278494.52646345884,
     "upload_bytes_per_second": 5405825.050604455,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.9613840232653,
+    "estimated_full_shards_per_hour": 59.758442358282316,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -244,3 +244,4 @@
 {"step": 709400, "loss": 0.5547887682914734, "lr": 8e-05, "grad_norm": 0.21233899742040024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103836183866, "tokens_per_second": 152595.98621484012, "data_wait_s": 1.6240999684669077e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 709500, "loss": 0.5771985054016113, "lr": 8e-05, "grad_norm": 0.2334944533253198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103850896986, "tokens_per_second": 153313.8106532134, "data_wait_s": 1.3657001545652747e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 709600, "loss": 0.5296552181243896, "lr": 8e-05, "grad_norm": 0.2121789817855828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103865416611, "tokens_per_second": 151151.58401937014, "data_wait_s": 2.504800067981705e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 709700, "loss": 0.5234677791595459, "lr": 8e-05, "grad_norm": 0.1987354581438336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103879925495, "tokens_per_second": 151246.15992796325, "data_wait_s": 1.4503999409498647e-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:fc16a01c9120ced8626650845f23f4cdde44f7f21aaca9bff9dda9000bef89dc
-size 524325
+oid sha256:25d849ee6c2ff66ab7fa75a2a61c1fb7d664361b652eada5cbffc9046acf09fd
+size 526443