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": 75479.87047918388,
-      "required_pages_per_second_upper_estimate": 0.006542403871152897
+      "required_token_positions_per_second": 75453.18884867674,
+      "required_pages_per_second_upper_estimate": 0.006540091175044512
     },
     {
       "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": 9434.983809897985,
-      "required_pages_per_second_upper_estimate": 0.0037637561073472097
+      "required_token_positions_per_second": 9431.648606084593,
+      "required_pages_per_second_upper_estimate": 0.0037624256446803068
     },
     {
       "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": 18869.96761979597,
-      "required_pages_per_second_upper_estimate": 0.0043652349087523155
+      "required_token_positions_per_second": 18863.297212169186,
+      "required_pages_per_second_upper_estimate": 0.004363691827343023
     },
     {
       "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": 28304.951429693956,
-      "required_pages_per_second_upper_estimate": 0.0017236230374652508
+      "required_token_positions_per_second": 28294.94581825378,
+      "required_pages_per_second_upper_estimate": 0.00172301374822393
     },
     {
       "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": 28304.951429693956,
-      "required_pages_per_second_upper_estimate": 0.002178993514178221
+      "required_token_positions_per_second": 28294.94581825378,
+      "required_pages_per_second_upper_estimate": 0.002178223254512251
     },
     {
       "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": 28304.951429693956,
-      "required_pages_per_second_upper_estimate": 0.35145276617820326
+      "required_token_positions_per_second": 28294.94581825378,
+      "required_pages_per_second_upper_estimate": 0.3513285299707436
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.4320394664140679,
-    "producer_p95_seconds": 0.2868722480016004,
-    "gpu_update_seconds": 0.9767902293941135,
-    "producer_batches_per_second": 2.3146033585774246,
-    "gpu_batches_per_second": 1.0237612640948335,
-    "producer_headroom": 2.2608819455812283,
+    "producer_batch_seconds": 0.12129263924245492,
+    "producer_p95_seconds": 0.3188847279998299,
+    "gpu_update_seconds": 0.9771356403221519,
+    "producer_batches_per_second": 8.244523379535627,
+    "gpu_batches_per_second": 1.0233993713199427,
+    "producer_headroom": 8.056017631613496,
     "sustainable": true,
-    "buffer_seconds": 60.56099422243504,
+    "buffer_seconds": 60.58240969997342,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291784.4538759097,
-    "upload_bytes_per_second": 3992982.7179019772,
+    "generation_bytes_per_second": 319276.5680587107,
+    "upload_bytes_per_second": 3995345.486508961,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.61015140731781,
+    "estimated_full_shards_per_hour": 68.509319127283,
     "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": 1.2597373989992775,
-      "bytes": 7034037,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.1449642560000939,
       "bytes": 7450954,
@@ -562,6 +556,12 @@
       "bytes": 6017424,
       "pages": 23,
       "shards": 2
+    },
+    {
+      "seconds": 1.0887724580024951,
+      "bytes": 6603634,
+      "pages": 44,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -142,3 +142,4 @@
 {"step": 294400, "loss": 0.6025479212403297, "lr": 0.0003, "grad_norm": 0.1530220006729937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43092112910, "tokens_per_second": 150859.8605852468, "data_wait_s": 9.805000445339829e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 294500, "loss": 0.6064490936696529, "lr": 0.0003, "grad_norm": 0.15015753211878693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43106871866, "tokens_per_second": 151812.81842149483, "data_wait_s": 1.2907999916933477e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 294600, "loss": 0.6203990988433361, "lr": 0.0003, "grad_norm": 0.19392050858719598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43121617038, "tokens_per_second": 150742.44589553765, "data_wait_s": 1.0978001228068024e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 294700, "loss": 0.6058078296482563, "lr": 0.0003, "grad_norm": 0.16915365988143374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43136199010, "tokens_per_second": 148780.71310533409, "data_wait_s": 1.2674001482082531e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:70a44edfb8a3fe082d5249f0b1dd0250bc0dbb59252a0f64266842db65351e96
-size 306723
+oid sha256:d053344affdad29104b00a04f93c1fed5eef7c7734d50408cccc45d8051e3064
+size 308841