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": 75763.61025782782,
-      "required_pages_per_second_upper_estimate": 0.005241502353432261
+      "required_token_positions_per_second": 75790.38877363384,
+      "required_pages_per_second_upper_estimate": 0.005243354953290438
     },
     {
       "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": 9470.451282228478,
-      "required_pages_per_second_upper_estimate": 0.003965080186357625
+      "required_token_positions_per_second": 9473.79859670423,
+      "required_pages_per_second_upper_estimate": 0.0039664816370287475
     },
     {
       "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": 18940.902564456956,
-      "required_pages_per_second_upper_estimate": 0.00394489049898549
+      "required_token_positions_per_second": 18947.59719340846,
+      "required_pages_per_second_upper_estimate": 0.00394628481364685
     },
     {
       "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": 28411.353846685437,
-      "required_pages_per_second_upper_estimate": 0.0017032172814833457
+      "required_token_positions_per_second": 28421.395790112692,
+      "required_pages_per_second_upper_estimate": 0.0017038192806586498
     },
     {
       "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": 28411.353846685437,
-      "required_pages_per_second_upper_estimate": 0.0022088577932412824
+      "required_token_positions_per_second": 28421.395790112692,
+      "required_pages_per_second_upper_estimate": 0.0022096385101729103
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28411.353846685437,
-      "required_pages_per_second_upper_estimate": 0.36443034141026204
+      "required_token_positions_per_second": 28421.395790112692,
+      "required_pages_per_second_upper_estimate": 0.36455914867834804
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5758578192968287,
-    "producer_p95_seconds": 0.17394959800003562,
-    "gpu_update_seconds": 0.9731320847712968,
-    "producer_batches_per_second": 1.7365397611880742,
-    "gpu_batches_per_second": 1.027609731144583,
-    "producer_headroom": 1.6898825580932004,
+    "producer_batch_seconds": 0.15217620494523487,
+    "producer_p95_seconds": 0.1721363199994812,
+    "gpu_update_seconds": 0.9727882544606855,
+    "producer_batches_per_second": 6.571329600182103,
+    "gpu_batches_per_second": 1.0279729380104417,
+    "producer_headroom": 6.392512251246982,
     "sustainable": true,
-    "buffer_seconds": 60.3341892558204,
+    "buffer_seconds": 60.3128717765625,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370833.81657867675,
+    "generation_bytes_per_second": 373719.53067025304,
     "upload_bytes_per_second": 4658638.935595296,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.57230446834781,
+    "estimated_full_shards_per_hour": 80.19151153641408,
     "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

@@ -47,3 +47,4 @@
 {"step": 689700, "loss": 0.5680307745933533, "lr": 8e-05, "grad_norm": 0.19993649434375324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100953492215, "tokens_per_second": 154183.4022057356, "data_wait_s": 1.3666999620909337e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 689800, "loss": 0.527131974697113, "lr": 8e-05, "grad_norm": 0.20848241318445657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100968277752, "tokens_per_second": 152955.31559345935, "data_wait_s": 1.6787000276963226e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 689900, "loss": 0.5641610622406006, "lr": 8e-05, "grad_norm": 0.19524551050481168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100982806168, "tokens_per_second": 150606.61389982823, "data_wait_s": 1.5345999599958304e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 690000, "loss": 0.5735451579093933, "lr": 8e-05, "grad_norm": 0.20603365382569494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100997233939, "tokens_per_second": 150022.19957227245, "data_wait_s": 1.6033000065363012e-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:c157ce16d56d0e510e7be13bb2fdb976f83ab9a3400626dd8da7525f81cb0ee0
-size 103947
+oid sha256:3f59b0d28157eec06ba5c4093124b26f3997e5024b21810d9aeaaa2bf5a691cf
+size 106065