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": 75457.4414506224,
-      "required_pages_per_second_upper_estimate": 0.006540459779803041
+      "required_token_positions_per_second": 75148.621759307,
+      "required_pages_per_second_upper_estimate": 0.006513692071656175
     },
     {
       "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": 9432.1801813278,
-      "required_pages_per_second_upper_estimate": 0.00376263769799258
+      "required_token_positions_per_second": 9393.577719913375,
+      "required_pages_per_second_upper_estimate": 0.003747238598976135
     },
     {
       "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": 18864.3603626556,
-      "required_pages_per_second_upper_estimate": 0.004363937768497246
+      "required_token_positions_per_second": 18787.15543982675,
+      "required_pages_per_second_upper_estimate": 0.004346077768361026
     },
     {
       "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": 28296.5405439834,
-      "required_pages_per_second_upper_estimate": 0.0017231108586539849
+      "required_token_positions_per_second": 28180.733159740128,
+      "required_pages_per_second_upper_estimate": 0.0017160587965479573
     },
     {
       "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": 28296.5405439834,
-      "required_pages_per_second_upper_estimate": 0.0021783460209134006
+      "required_token_positions_per_second": 28180.733159740128,
+      "required_pages_per_second_upper_estimate": 0.002169430847898996
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28296.5405439834,
-      "required_pages_per_second_upper_estimate": 0.35134833112710184
+      "required_token_positions_per_second": 28180.733159740128,
+      "required_pages_per_second_upper_estimate": 0.349910390997183
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08969458308587264,
-    "producer_p95_seconds": 0.2588099609965866,
-    "gpu_update_seconds": 0.9770805712813082,
-    "producer_batches_per_second": 11.148945294083264,
-    "gpu_batches_per_second": 1.0234570509253256,
-    "producer_headroom": 10.893417837126927,
+    "producer_batch_seconds": 0.12076059967978381,
+    "producer_p95_seconds": 0.2938357570019434,
+    "gpu_update_seconds": 0.9810958374744768,
+    "producer_batches_per_second": 8.28084658946429,
+    "gpu_batches_per_second": 1.0192684157892118,
+    "producer_headroom": 8.124304119688132,
     "sustainable": true,
-    "buffer_seconds": 60.57899541944111,
+    "buffer_seconds": 60.82794192341756,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 251893.77525562528,
-    "upload_bytes_per_second": 2875994.464942769,
+    "generation_bytes_per_second": 263928.11995117454,
+    "upload_bytes_per_second": 2808159.7402229635,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.05054038287705,
+    "estimated_full_shards_per_hour": 56.6328306093352,
     "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": 1.2669327539988444,
-      "bytes": 12406942,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 1.4563389440008905,
       "bytes": 6496961,
@@ -562,6 +556,12 @@
       "bytes": 9943938,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.3160915080006816,
+      "bytes": 2528581,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -158,3 +158,4 @@
 {"step": 296000, "loss": 0.6096544731408358, "lr": 0.0003, "grad_norm": 0.16613912715231427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43326700565, "tokens_per_second": 151365.7632719637, "data_wait_s": 1.3999000657349825e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 296100, "loss": 0.6223481129854918, "lr": 0.0003, "grad_norm": 0.17809722039278386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43341269048, "tokens_per_second": 100965.20529502559, "data_wait_s": 9.744999260874465e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 296200, "loss": 0.5966150555759668, "lr": 0.0003, "grad_norm": 0.2085681188079925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43355805065, "tokens_per_second": 148752.84224579704, "data_wait_s": 1.4280001778388396e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 296300, "loss": 0.6124036759138107, "lr": 0.0003, "grad_norm": 0.17343848056638297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43370467936, "tokens_per_second": 149906.69870673746, "data_wait_s": 1.152500044554472e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b2eac56339ec11ba017cff3d23ac053b17173c8bc8c4c2e6020fbf06e7475b5e
-size 341394
+oid sha256:d94a9a690b6fe1b419d9ed0ca8aa19713855e2346ec1fd45d805713ef310f631
+size 343512