Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74276.48769289246,
-      "required_pages_per_second_upper_estimate": 0.006438097701182945
+      "required_token_positions_per_second": 74285.24964389975,
+      "required_pages_per_second_upper_estimate": 0.006438857164889294
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9284.560961611558,
-      "required_pages_per_second_upper_estimate": 0.0037037501841437523
+      "required_token_positions_per_second": 9285.656205487468,
+      "required_pages_per_second_upper_estimate": 0.003704187093301208
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18569.121923223116,
-      "required_pages_per_second_upper_estimate": 0.004295639551553609
+      "required_token_positions_per_second": 18571.312410974937,
+      "required_pages_per_second_upper_estimate": 0.004296146282343725
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27853.68288483467,
-      "required_pages_per_second_upper_estimate": 0.0016961431507063957
+      "required_token_positions_per_second": 27856.968616462404,
+      "required_pages_per_second_upper_estimate": 0.001696343234523623
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27853.68288483467,
-      "required_pages_per_second_upper_estimate": 0.002144253612403669
+      "required_token_positions_per_second": 27856.968616462404,
+      "required_pages_per_second_upper_estimate": 0.0021445065571198594
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27853.68288483467,
-      "required_pages_per_second_upper_estimate": 0.3458495211497159
+      "required_token_positions_per_second": 27856.968616462404,
+      "required_pages_per_second_upper_estimate": 0.34589031893989597
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.49753229417964917,
-    "producer_p95_seconds": 0.3124902529998508,
-    "gpu_update_seconds": 0.9926155946527754,
-    "producer_batches_per_second": 2.0099197814864245,
-    "gpu_batches_per_second": 1.007439340452643,
-    "producer_headroom": 1.9950777191045237,
+    "producer_batch_seconds": 0.08805944840622715,
+    "producer_p95_seconds": 0.3086907210017671,
+    "gpu_update_seconds": 0.9924985155657278,
+    "producer_batches_per_second": 11.355964840784589,
+    "gpu_batches_per_second": 1.0075581820190394,
+    "producer_headroom": 11.2707782472953,
     "sustainable": true,
-    "buffer_seconds": 60.5495512738193,
+    "buffer_seconds": 60.54240944950939,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 425738.8541220377,
-    "upload_bytes_per_second": 5981629.873143274,
+    "generation_bytes_per_second": 441979.5700009061,
+    "upload_bytes_per_second": 6056256.739316602,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.35364740129326,
+    "estimated_full_shards_per_hour": 94.83852696438205,
     "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": [
-    {
-      "seconds": 0.8778860090005765,
-      "bytes": 2613270,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1979163849991892,
       "bytes": 9677162,
@@ -562,6 +556,12 @@
       "bytes": 5823773,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.6930653950003034,
+      "bytes": 12802437,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -132,3 +132,4 @@
 {"step": 98200, "loss": 0.6513643469661474, "lr": 0.0003, "grad_norm": 0.12758982647775435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14369909417, "tokens_per_second": 146997.03627211528, "data_wait_s": 1.3330001820577309e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 98300, "loss": 0.5933377612382174, "lr": 0.0003, "grad_norm": 0.1321144705566266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14384422838, "tokens_per_second": 146645.75477491578, "data_wait_s": 1.4479999663308263e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 98400, "loss": 0.6232114732265472, "lr": 0.0003, "grad_norm": 0.1266015312917367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14398943189, "tokens_per_second": 147535.5966096913, "data_wait_s": 1.2119999155402184e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 98500, "loss": 0.5763075035065413, "lr": 0.0003, "grad_norm": 0.1383303369129346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14413621505, "tokens_per_second": 149235.90115195137, "data_wait_s": 1.461600186303258e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:901dc87615ddf0796910b13dba1d97ecc4d5ae39cd63b16b7a56ba14eefa2b89
-size 285543
+oid sha256:2d1de521c870a972fb3c081d42deb1fd87ebb1037adc6bf2e4c236933f36a6ca
+size 287661