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": 75832.43085801014,
-      "required_pages_per_second_upper_estimate": 0.006572962911233446
+      "required_token_positions_per_second": 75104.38345404137,
+      "required_pages_per_second_upper_estimate": 0.006509857607476709
     },
     {
       "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": 9479.053857251267,
-      "required_pages_per_second_upper_estimate": 0.0037813363081423594
+      "required_token_positions_per_second": 9388.04793175517,
+      "required_pages_per_second_upper_estimate": 0.0037450326838021266
     },
     {
       "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": 18958.107714502534,
-      "required_pages_per_second_upper_estimate": 0.004385624568449993
+      "required_token_positions_per_second": 18776.09586351034,
+      "required_pages_per_second_upper_estimate": 0.00434351933002212
     },
     {
       "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": 28437.1615717538,
-      "required_pages_per_second_upper_estimate": 0.0017316739414636882
+      "required_token_positions_per_second": 28164.143795265514,
+      "required_pages_per_second_upper_estimate": 0.001715048591289122
     },
     {
       "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": 28437.1615717538,
-      "required_pages_per_second_upper_estimate": 0.002189171416895085
+      "required_token_positions_per_second": 28164.143795265514,
+      "required_pages_per_second_upper_estimate": 0.0021681537526994405
     },
     {
       "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": 28437.1615717538,
-      "required_pages_per_second_upper_estimate": 0.3530943736637049
+      "required_token_positions_per_second": 28164.143795265514,
+      "required_pages_per_second_upper_estimate": 0.34970440661143964
     }
   ],
   "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": 0.0517421977343453,
-    "producer_p95_seconds": 0.16660348799996427,
-    "gpu_update_seconds": 0.9722489331516946,
-    "producer_batches_per_second": 19.3265853362897,
-    "gpu_batches_per_second": 1.0285431702746601,
-    "producer_headroom": 18.790251974672845,
+    "producer_batch_seconds": 0.1975333914999986,
+    "producer_p95_seconds": 0.1758200740000575,
+    "gpu_update_seconds": 0.9816737267421466,
+    "producer_batches_per_second": 5.062435228830702,
+    "gpu_batches_per_second": 1.0186683953727398,
+    "producer_headroom": 4.969659657476966,
     "sustainable": true,
-    "buffer_seconds": 60.27943385540507,
+    "buffer_seconds": 60.863771058013086,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395175.09755235974,
-    "upload_bytes_per_second": 4897807.191103762,
+    "generation_bytes_per_second": 441917.59480484016,
+    "upload_bytes_per_second": 5352685.132132118,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.79537672927947,
+    "estimated_full_shards_per_hour": 94.82522853001502,
     "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": [
@@ -208,6 +208,12 @@
       "bytes": 6249982,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0289287979999244,
+      "bytes": 7839831,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -30,3 +30,4 @@
 {"step": 43000, "loss": 0.6533548310399055, "lr": 0.0003, "grad_norm": 0.13873285402286184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6292353959, "tokens_per_second": 147173.03461526957, "data_wait_s": 1.0603000191622414e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 43100, "loss": 0.6297736801207066, "lr": 0.0003, "grad_norm": 0.13378830740682623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6307029871, "tokens_per_second": 149518.64783436124, "data_wait_s": 1.8134000129066408e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 43200, "loss": 0.6300983037799597, "lr": 0.0003, "grad_norm": 0.13309480852096092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6321573242, "tokens_per_second": 147560.62589981104, "data_wait_s": 1.5577999874949455e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 43300, "loss": 0.6022576056420803, "lr": 0.0003, "grad_norm": 0.15580596171508607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6336267358, "tokens_per_second": 150087.74574274768, "data_wait_s": 1.1791999895649496e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cebd03c256c1d1be7b5ee5789d46cb37e25383d0e7df0d2ce1756d737afd0290
-size 67941
+oid sha256:1eed39071dea2d7f3a6cdc11943f78698e19251c4814e40edd07e59b6c63a336
+size 70059