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": 74887.44135510734,
-      "required_pages_per_second_upper_estimate": 0.00649105361617583
+      "required_token_positions_per_second": 73782.40244838505,
+      "required_pages_per_second_upper_estimate": 0.006395271644436646
     },
     {
       "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": 9360.930169388417,
-      "required_pages_per_second_upper_estimate": 0.00373421500294735
+      "required_token_positions_per_second": 9222.800306048131,
+      "required_pages_per_second_upper_estimate": 0.003679112935235412
     },
     {
       "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": 18721.860338776834,
-      "required_pages_per_second_upper_estimate": 0.004330972895887658
+      "required_token_positions_per_second": 18445.600612096263,
+      "required_pages_per_second_upper_estimate": 0.004267065070125239
     },
     {
       "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": 28082.790508165253,
-      "required_pages_per_second_upper_estimate": 0.0017100946029324265
+      "required_token_positions_per_second": 27668.400918144394,
+      "required_pages_per_second_upper_estimate": 0.0016848604510343091
     },
     {
       "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": 28082.790508165253,
-      "required_pages_per_second_upper_estimate": 0.00216189095145815
+      "required_token_positions_per_second": 27668.400918144394,
+      "required_pages_per_second_upper_estimate": 0.0021299900937145413
     },
     {
       "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": 28082.790508165253,
-      "required_pages_per_second_upper_estimate": 0.3486942710575916
+      "required_token_positions_per_second": 27668.400918144394,
+      "required_pages_per_second_upper_estimate": 0.3435489392222754
     }
   ],
   "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": 0.2340558968514017,
-    "producer_p95_seconds": 0.32403773199985153,
-    "gpu_update_seconds": 0.9845175461448948,
-    "producer_batches_per_second": 4.2724836821132675,
-    "gpu_batches_per_second": 1.015725929838153,
-    "producer_headroom": 4.206335150658259,
+    "producer_batch_seconds": 0.16116494035941287,
+    "producer_p95_seconds": 0.2648009529984847,
+    "gpu_update_seconds": 0.9992626636355043,
+    "producer_batches_per_second": 6.204823442182317,
+    "gpu_batches_per_second": 1.0007378804305698,
+    "producer_headroom": 6.200248400223121,
     "sustainable": true,
-    "buffer_seconds": 60.05557031483858,
+    "buffer_seconds": 60.955022481765766,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361323.8430396921,
-    "upload_bytes_per_second": 5792735.62686599,
+    "generation_bytes_per_second": 385655.9625487236,
+    "upload_bytes_per_second": 5835358.89627474,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.53168552773545,
+    "estimated_full_shards_per_hour": 82.75279195162088,
     "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": 1.4256255729997065,
-      "bytes": 11482907,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 0.8028865010001027,
       "bytes": 1091530,
@@ -562,6 +556,12 @@
       "bytes": 5919036,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1784255179991305,
+      "bytes": 13145531,
+      "pages": 35,
+      "shards": 5
     }
   ],
   "cache_key": {

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

@@ -185,3 +185,4 @@
 {"step": 133500, "loss": 0.6138676777482033, "lr": 0.0003, "grad_norm": 0.13419845753725104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19537352418, "tokens_per_second": 148410.90869555785, "data_wait_s": 1.2267002603039145e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 133600, "loss": 0.5971824824810028, "lr": 0.0003, "grad_norm": 0.13473001862666656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19551872648, "tokens_per_second": 147372.69474714156, "data_wait_s": 1.354100095340982e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 133700, "loss": 0.5746112670749426, "lr": 0.0003, "grad_norm": 0.12353973311740744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19566523091, "tokens_per_second": 149479.2568496617, "data_wait_s": 1.4605000615119934e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 133800, "loss": 0.5904359146952629, "lr": 0.0003, "grad_norm": 0.12785646706601875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19581087503, "tokens_per_second": 147423.86574178244, "data_wait_s": 1.207899913424626e-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 sha256:2e3c7f8e530d4ed08ae429b9d23a0e5121cbd5b026fa80561546bff4d2a11acb
-size 398580
+oid sha256:6cfc76e4e43718a3f4b31ad41d14134702f2129b58730be36b1f646878aa9bae
+size 400698