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": 76790.28607373817,
-      "required_pages_per_second_upper_estimate": 0.005312530168593037
+      "required_token_positions_per_second": 76995.99828305747,
+      "required_pages_per_second_upper_estimate": 0.005326761816551834
     },
     {
       "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": 9598.785759217271,
-      "required_pages_per_second_upper_estimate": 0.004018811151944208
+      "required_token_positions_per_second": 9624.499785382184,
+      "required_pages_per_second_upper_estimate": 0.00402957707772953
     },
     {
       "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": 19197.571518434543,
-      "required_pages_per_second_upper_estimate": 0.003998347873283522
+      "required_token_positions_per_second": 19248.99957076437,
+      "required_pages_per_second_upper_estimate": 0.00400905898033488
     },
     {
       "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": 28796.357277651816,
-      "required_pages_per_second_upper_estimate": 0.0017262976493035796
+      "required_token_positions_per_second": 28873.499356146553,
+      "required_pages_per_second_upper_estimate": 0.0017309221991202053
     },
     {
       "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": 28796.357277651816,
-      "required_pages_per_second_upper_estimate": 0.002238790116547795
+      "required_token_positions_per_second": 28873.499356146553,
+      "required_pages_per_second_upper_estimate": 0.002244787573838617
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28796.357277651816,
-      "required_pages_per_second_upper_estimate": 0.3693687520382219
+      "required_token_positions_per_second": 28873.499356146553,
+      "required_pages_per_second_upper_estimate": 0.37035824779244175
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5149998101641415,
-    "producer_p95_seconds": 0.1910665900013555,
-    "gpu_update_seconds": 0.9601214394383477,
-    "producer_batches_per_second": 1.9417482885698123,
-    "gpu_batches_per_second": 1.041534913109513,
-    "producer_headroom": 1.8643141618485963,
+    "producer_batch_seconds": 0.07293816497649175,
+    "producer_p95_seconds": 0.1842724639991502,
+    "gpu_update_seconds": 0.9575562580402756,
+    "producer_batches_per_second": 13.71024346886577,
+    "gpu_batches_per_second": 1.0443250635180321,
+    "producer_headroom": 13.128329432868233,
     "sustainable": true,
-    "buffer_seconds": 60.48765068461591,
+    "buffer_seconds": 60.32604425653736,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391830.6602257355,
-    "upload_bytes_per_second": 3604341.4727554587,
+    "generation_bytes_per_second": 406606.6539260375,
+    "upload_bytes_per_second": 3681734.0148414355,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.07773833350228,
+    "estimated_full_shards_per_hour": 87.24832261405795,
     "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": 7479884,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.613565400999505,
+      "bytes": 11743615,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -92,3 +92,4 @@
 {"step": 694200, "loss": 0.5563333630561829, "lr": 8e-05, "grad_norm": 0.1970245352823422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101612206431, "tokens_per_second": 150376.8800478434, "data_wait_s": 1.5105999409570359e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 694300, "loss": 0.5458000302314758, "lr": 8e-05, "grad_norm": 0.19918026706141426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101626681522, "tokens_per_second": 150671.45998568556, "data_wait_s": 1.5998000890249386e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 694400, "loss": 0.5513448119163513, "lr": 8e-05, "grad_norm": 0.19856764535692467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101641298787, "tokens_per_second": 152277.84862259548, "data_wait_s": 2.0280000171624124e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 694500, "loss": 0.525731086730957, "lr": 8e-05, "grad_norm": 0.18994030377386611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101655905838, "tokens_per_second": 152045.6538247751, "data_wait_s": 1.7819998902268708e-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 sha256:9e8a3053defc95b887eedef858fef0d4e780e0ec25969dae58bb8cc6b79b4afc
-size 200040
+oid sha256:7f1348a2646eab3dfe580f7ce11a28bb2014a35929655c5185f22df4d036c3db
+size 202158