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": 75903.09098390253,
-      "required_pages_per_second_upper_estimate": 0.006579087551859346
+      "required_token_positions_per_second": 76000.48524185155,
+      "required_pages_per_second_upper_estimate": 0.006587529439294878
     },
     {
       "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": 9487.886372987816,
-      "required_pages_per_second_upper_estimate": 0.0037848597307275473
+      "required_token_positions_per_second": 9500.060655231444,
+      "required_pages_per_second_upper_estimate": 0.003789716233936271
     },
     {
       "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": 18975.77274597563,
-      "required_pages_per_second_upper_estimate": 0.004389711062587358
+      "required_token_positions_per_second": 19000.121310462888,
+      "required_pages_per_second_upper_estimate": 0.0043953436744613855
     },
     {
       "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": 28463.659118963446,
-      "required_pages_per_second_upper_estimate": 0.0017332875030668697
+      "required_token_positions_per_second": 28500.181965694333,
+      "required_pages_per_second_upper_estimate": 0.001735511552812213
     },
     {
       "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": 28463.659118963446,
-      "required_pages_per_second_upper_estimate": 0.002191211271429189
+      "required_token_positions_per_second": 28500.181965694333,
+      "required_pages_per_second_upper_estimate": 0.002194022901272243
     },
     {
       "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": 28463.659118963446,
-      "required_pages_per_second_upper_estimate": 0.3534233845184629
+      "required_token_positions_per_second": 28500.181965694333,
+      "required_pages_per_second_upper_estimate": 0.3538768760407556
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1775142985234055,
-    "producer_p95_seconds": 0.2408735289991455,
-    "gpu_update_seconds": 0.9713438417894759,
-    "producer_batches_per_second": 5.633349022124821,
-    "gpu_batches_per_second": 1.0295015595689903,
-    "producer_headroom": 5.471918881291711,
+    "producer_batch_seconds": 0.08882844120317657,
+    "producer_p95_seconds": 0.2087490030007757,
+    "gpu_update_seconds": 0.9700990693069923,
+    "producer_batches_per_second": 11.257655616321221,
+    "gpu_batches_per_second": 1.030822553736051,
+    "producer_headroom": 10.92104123597185,
     "sustainable": true,
-    "buffer_seconds": 60.2233181909475,
+    "buffer_seconds": 60.14614229703352,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 534273.3576819414,
-    "upload_bytes_per_second": 5972312.997826691,
+    "generation_bytes_per_second": 538524.4832146277,
+    "upload_bytes_per_second": 6058163.920097129,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.6426253113144,
+    "estimated_full_shards_per_hour": 115.5548178894913,
     "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": [
@@ -424,6 +424,12 @@
       "bytes": 914039,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 1.1632707989992923,
+      "bytes": 10884064,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -97,3 +97,4 @@
 {"step": 124700, "loss": 0.6310654245316982, "lr": 0.0003, "grad_norm": 0.14385225200419247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18248395216, "tokens_per_second": 148602.9556673551, "data_wait_s": 1.4315999578684568e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 124800, "loss": 0.682103905826807, "lr": 0.0003, "grad_norm": 0.16158304146214011, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18262972049, "tokens_per_second": 148817.31949435032, "data_wait_s": 1.6299998605973087e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 124900, "loss": 0.6229158714413643, "lr": 0.0003, "grad_norm": 0.17464007645231128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18277769906, "tokens_per_second": 151074.6458839549, "data_wait_s": 1.8365999494562857e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 125000, "loss": 0.6308581009507179, "lr": 0.0003, "grad_norm": 0.1471120359130923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18292497399, "tokens_per_second": 150460.7163588629, "data_wait_s": 1.7464999473304488e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b0ac61678c6cd4f820eaaec2323955805cdc376463d1756cab52e84aa7391a0a
-size 210630
+oid sha256:592ad85b7cd81390450e7fb5c535122b53b01138f3914e1aaf9badfbb248b09a
+size 212748