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": 77465.34367746899,
-      "required_pages_per_second_upper_estimate": 0.005359232219968577
+      "required_token_positions_per_second": 76919.27870488334,
+      "required_pages_per_second_upper_estimate": 0.005321454178119801
     },
     {
       "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": 9683.167959683624,
-      "required_pages_per_second_upper_estimate": 0.004054140217178749
+      "required_token_positions_per_second": 9614.909838110418,
+      "required_pages_per_second_upper_estimate": 0.004025561967067713
     },
     {
       "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": 19366.335919367248,
-      "required_pages_per_second_upper_estimate": 0.004033497047381261
+      "required_token_positions_per_second": 19229.819676220835,
+      "required_pages_per_second_upper_estimate": 0.004005064314109296
     },
     {
       "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": 29049.50387905087,
-      "required_pages_per_second_upper_estimate": 0.0017414734015249735
+      "required_token_positions_per_second": 28844.72951433125,
+      "required_pages_per_second_upper_estimate": 0.001729197491032383
     },
     {
       "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": 29049.50387905087,
-      "required_pages_per_second_upper_estimate": 0.0022584711513323474
+      "required_token_positions_per_second": 28844.72951433125,
+      "required_pages_per_second_upper_estimate": 0.0022425508451826377
     },
     {
       "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": 29049.50387905087,
-      "required_pages_per_second_upper_estimate": 0.37261584483332527
+      "required_token_positions_per_second": 28844.72951433125,
+      "required_pages_per_second_upper_estimate": 0.3699892191522845
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.109469270648475,
-    "producer_p95_seconds": 0.23832793900146498,
-    "gpu_update_seconds": 0.9517546363309299,
-    "producer_batches_per_second": 9.134983672369346,
-    "gpu_batches_per_second": 1.0506909678476155,
-    "producer_headroom": 8.69426306298487,
+    "producer_batch_seconds": 0.5037990929060072,
+    "producer_p95_seconds": 0.2295647419996385,
+    "gpu_update_seconds": 0.9585113282571547,
+    "producer_batches_per_second": 1.9849182225236122,
+    "gpu_batches_per_second": 1.0432844876421894,
+    "producer_headroom": 1.9025666019529381,
     "sustainable": true,
-    "buffer_seconds": 60.912296725179516,
+    "buffer_seconds": 60.38621368020075,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359121.23971939844,
-    "upload_bytes_per_second": 5834603.507481017,
+    "generation_bytes_per_second": 315058.4229671021,
+    "upload_bytes_per_second": 5768108.549336495,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.05905812918152,
+    "estimated_full_shards_per_hour": 67.604203384016,
     "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": 1.0358149310013687,
-      "bytes": 6338461,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8834309400008351,
       "bytes": 840516,
@@ -562,6 +556,12 @@
       "bytes": 11950661,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.2175075589984772,
+      "bytes": 2434158,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -214,3 +214,4 @@
 {"step": 706400, "loss": 0.5500536561012268, "lr": 8e-05, "grad_norm": 0.23392699689428295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103397046820, "tokens_per_second": 153541.64396606828, "data_wait_s": 1.678199987509288e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 706500, "loss": 0.5283864736557007, "lr": 8e-05, "grad_norm": 0.20569008465081567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103411683185, "tokens_per_second": 152677.07074413158, "data_wait_s": 1.5656998584745452e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 706600, "loss": 0.5602980852127075, "lr": 8e-05, "grad_norm": 0.20655222255051234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103426180889, "tokens_per_second": 140863.6111176759, "data_wait_s": 1.373200211673975e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 706700, "loss": 0.5526378750801086, "lr": 8e-05, "grad_norm": 0.2174005814233793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103440948865, "tokens_per_second": 153102.94859574814, "data_wait_s": 1.7291000403929502e-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 sha256:9e4f9f7b8fd19f372bbbd1bbf33e8799b95640c81fbc6e6063839e4539e8c621
-size 460785
+oid sha256:c82f8384f8ae9e388fe2b4d748049285e459ba46fc19cb5e80309ba2fae7fa93
+size 462903