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": 75408.82680906644,
-      "required_pages_per_second_upper_estimate": 0.006536245985885652
+      "required_token_positions_per_second": 75609.23453817287,
+      "required_pages_per_second_upper_estimate": 0.006553616819915852
     },
     {
       "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": 9426.103351133304,
-      "required_pages_per_second_upper_estimate": 0.003760213559571288
+      "required_token_positions_per_second": 9451.154317271608,
+      "required_pages_per_second_upper_estimate": 0.0037702067645091785
     },
     {
       "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": 18852.20670226661,
-      "required_pages_per_second_upper_estimate": 0.004361126233063371
+      "required_token_positions_per_second": 18902.308634543217,
+      "required_pages_per_second_upper_estimate": 0.004372716433331671
     },
     {
       "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": 28278.310053399913,
-      "required_pages_per_second_upper_estimate": 0.0017220007174254424
+      "required_token_positions_per_second": 28353.462951814825,
+      "required_pages_per_second_upper_estimate": 0.0017265771346421255
     },
     {
       "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": 28278.310053399913,
-      "required_pages_per_second_upper_estimate": 0.002176942587283056
+      "required_token_positions_per_second": 28353.462951814825,
+      "required_pages_per_second_upper_estimate": 0.0021827280654395736
     },
     {
       "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": 28278.310053399913,
-      "required_pages_per_second_upper_estimate": 0.351121969447582
+      "required_token_positions_per_second": 28353.462951814825,
+      "required_pages_per_second_upper_estimate": 0.3520551169253241
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.09610981756247838,
-    "producer_p95_seconds": 0.2500162529977388,
-    "gpu_update_seconds": 0.9777104766087629,
-    "producer_batches_per_second": 10.404764314009098,
-    "gpu_batches_per_second": 1.0227976726490131,
-    "producer_headroom": 10.172847076451683,
+    "producer_batch_seconds": 0.07958586630476816,
+    "producer_p95_seconds": 0.2136629560009169,
+    "gpu_update_seconds": 0.9751189844777084,
+    "producer_batches_per_second": 12.565045107011517,
+    "gpu_batches_per_second": 1.0255158764400616,
+    "producer_headroom": 12.25241402466567,
     "sustainable": true,
-    "buffer_seconds": 60.6180495497433,
+    "buffer_seconds": 60.45737703761792,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378660.3806257372,
-    "upload_bytes_per_second": 6214729.267162323,
+    "generation_bytes_per_second": 395825.6496807589,
+    "upload_bytes_per_second": 6251150.393481496,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.25170291976058,
+    "estimated_full_shards_per_hour": 84.93497007195545,
     "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": 0.7185119169989775,
-      "bytes": 873551,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 0.8877478260001226,
       "bytes": 6165145,
@@ -562,6 +556,12 @@
       "bytes": 7802740,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1620692619981128,
+      "bytes": 6079962,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -174,3 +174,4 @@
 {"step": 57400, "loss": 0.6162293180823326, "lr": 0.0003, "grad_norm": 0.13353620542181208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8401222730, "tokens_per_second": 149766.70442513688, "data_wait_s": 1.0446001397212967e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 57500, "loss": 0.6358325146138668, "lr": 0.0003, "grad_norm": 0.129646001522135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8416017932, "tokens_per_second": 150498.3857282842, "data_wait_s": 8.213999535655603e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 57600, "loss": 0.6794849261641502, "lr": 0.0003, "grad_norm": 0.13734259327304033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8430742090, "tokens_per_second": 150774.07416051565, "data_wait_s": 1.3688000763067976e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 57700, "loss": 0.6338352151215076, "lr": 0.0003, "grad_norm": 0.1444835370872305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8445307267, "tokens_per_second": 149257.57544293418, "data_wait_s": 1.0380997991887853e-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 sha256:80d26d36cb7e422963c518255ab5603da97b77f637f4f0226222df3a098101e9
-size 375282
+oid sha256:2e8ec79a6026afdddc43333a82b2342f63bb8a264f7929f4b729698a38d0216d
+size 377400