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": 76995.07005274466,
-      "required_pages_per_second_upper_estimate": 0.005326697599425002
+      "required_token_positions_per_second": 76796.44058226769,
+      "required_pages_per_second_upper_estimate": 0.005312955951773537
     },
     {
       "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": 9624.383756593083,
-      "required_pages_per_second_upper_estimate": 0.0040295284988984895
+      "required_token_positions_per_second": 9599.555072783462,
+      "required_pages_per_second_upper_estimate": 0.004019133247469278
     },
     {
       "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": 19248.767513186165,
-      "required_pages_per_second_upper_estimate": 0.004009010648861111
+      "required_token_positions_per_second": 19199.110145566923,
+      "required_pages_per_second_upper_estimate": 0.003998668328738871
     },
     {
       "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": 28873.151269779246,
-      "required_pages_per_second_upper_estimate": 0.0017309013318739825
+      "required_token_positions_per_second": 28798.665218350383,
+      "required_pages_per_second_upper_estimate": 0.0017264360068244373
     },
     {
       "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": 28873.151269779246,
-      "required_pages_per_second_upper_estimate": 0.002244760511654629
+      "required_token_positions_per_second": 28798.665218350383,
+      "required_pages_per_second_upper_estimate": 0.0022389695487855504
     },
     {
       "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": 28873.151269779246,
-      "required_pages_per_second_upper_estimate": 0.37035378291426796
+      "required_token_positions_per_second": 28798.665218350383,
+      "required_pages_per_second_upper_estimate": 0.36939835582342945
     }
   ],
   "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": 0.10281502670309806,
-    "producer_p95_seconds": 0.19412416799968923,
-    "gpu_update_seconds": 0.9575678020617867,
-    "producer_batches_per_second": 9.726204739388233,
-    "gpu_batches_per_second": 1.044312473588659,
-    "producer_headroom": 9.313500494698923,
+    "producer_batch_seconds": 0.0720036923984253,
+    "producer_p95_seconds": 0.18279787300161843,
+    "gpu_update_seconds": 0.9600444947838351,
+    "producer_batches_per_second": 13.88817665719973,
+    "gpu_batches_per_second": 1.0416183889739,
+    "producer_headroom": 13.333267542329969,
     "sustainable": true,
-    "buffer_seconds": 60.32677152989256,
+    "buffer_seconds": 60.48280317138161,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 423527.54896745057,
-    "upload_bytes_per_second": 3586096.548601037,
+    "generation_bytes_per_second": 440784.5434382343,
+    "upload_bytes_per_second": 3542790.187009241,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.8791527916683,
+    "estimated_full_shards_per_hour": 94.5821020828273,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 11592714,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.1851188379987434,
+      "bytes": 1113462,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -88,3 +88,4 @@
 {"step": 693800, "loss": 0.5116367340087891, "lr": 8e-05, "grad_norm": 0.2069830763113695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101553667688, "tokens_per_second": 153496.04986118193, "data_wait_s": 1.867600076366216e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 693900, "loss": 0.5128055810928345, "lr": 8e-05, "grad_norm": 0.19948939129563428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101568375394, "tokens_per_second": 152925.7557987483, "data_wait_s": 1.66709996847203e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 694000, "loss": 0.5408977270126343, "lr": 8e-05, "grad_norm": 0.20214827091926568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101583083857, "tokens_per_second": 153069.48275726812, "data_wait_s": 1.4594999811379239e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 694100, "loss": 0.5392832159996033, "lr": 8e-05, "grad_norm": 0.19794096891125976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101597757348, "tokens_per_second": 152439.76018710792, "data_wait_s": 1.3160999515093863e-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 sha256:83c561bf042bb83899a79bbde9d88d339937b1737aed59d4aec8ab6c269716b2
-size 191568
+oid sha256:82b3af2d903da41bf59db7adad6230a236d865fb007f6385411ba43847216085
+size 193686