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": 75796.732885317,
-      "required_pages_per_second_upper_estimate": 0.005243793853657047
+      "required_token_positions_per_second": 76788.87049226927,
+      "required_pages_per_second_upper_estimate": 0.005312432235382417
     },
     {
       "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": 9474.591610664625,
-      "required_pages_per_second_upper_estimate": 0.00396681365541395
+      "required_token_positions_per_second": 9598.608811533659,
+      "required_pages_per_second_upper_estimate": 0.004018737067644168
     },
     {
       "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": 18949.18322132925,
-      "required_pages_per_second_upper_estimate": 0.003946615141436382
+      "required_token_positions_per_second": 19197.217623067318,
+      "required_pages_per_second_upper_estimate": 0.003998274166211378
     },
     {
       "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": 28423.774831993876,
-      "required_pages_per_second_upper_estimate": 0.0017039619005868403
+      "required_token_positions_per_second": 28795.826434600975,
+      "required_pages_per_second_upper_estimate": 0.001726265826073232
     },
     {
       "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": 28423.774831993876,
-      "required_pages_per_second_upper_estimate": 0.002209823470214873
+      "required_token_positions_per_second": 28795.826434600975,
+      "required_pages_per_second_upper_estimate": 0.0022387488458355236
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28423.774831993876,
-      "required_pages_per_second_upper_estimate": 0.36458966447318375
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28795.826434600975,
+      "required_pages_per_second_upper_estimate": 0.3693619429535405
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.34051346025790963,
-    "producer_p95_seconds": 0.24626814300063415,
-    "gpu_update_seconds": 0.9727068330445447,
-    "producer_batches_per_second": 2.93674147049161,
-    "gpu_batches_per_second": 1.028058985532186,
-    "producer_headroom": 2.8565884952324736,
+    "producer_batch_seconds": 0.06846497434378307,
+    "producer_p95_seconds": 0.235886106001999,
+    "gpu_update_seconds": 0.9601391390100286,
+    "producer_batches_per_second": 14.606008540640088,
+    "gpu_batches_per_second": 1.0415157130570376,
+    "producer_headroom": 14.023800464583298,
     "sustainable": true,
-    "buffer_seconds": 60.307823648761776,
+    "buffer_seconds": 60.488765757631796,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 466978.9619144702,
-    "upload_bytes_per_second": 5632460.521580394,
+    "generation_bytes_per_second": 467003.8664902739,
+    "upload_bytes_per_second": 5748099.14689726,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.20281451297359,
+    "estimated_full_shards_per_hour": 100.20815845519222,
     "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": 1.7939357480008766,
-      "bytes": 6243705,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0855320710015803,
       "bytes": 8387056,
@@ -562,6 +556,12 @@
       "bytes": 938191,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.3705294199971831,
+      "bytes": 12626897,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -185,3 +185,4 @@
 {"step": 703500, "loss": 0.49644920229911804, "lr": 8e-05, "grad_norm": 0.19843883597558742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102973085049, "tokens_per_second": 151974.40577166385, "data_wait_s": 1.4781999198021367e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 703600, "loss": 0.5154860019683838, "lr": 8e-05, "grad_norm": 0.20916800523318554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102987543188, "tokens_per_second": 150716.4353829191, "data_wait_s": 1.824400169425644e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 703700, "loss": 0.5930591225624084, "lr": 8e-05, "grad_norm": 0.22176475071326485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103002235850, "tokens_per_second": 153173.04245013505, "data_wait_s": 1.5246998373186216e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 703800, "loss": 0.5478715300559998, "lr": 8e-05, "grad_norm": 0.21695522373717682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103016893084, "tokens_per_second": 152664.82882184125, "data_wait_s": 9.547999070491642e-06, "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 sha256:6e3702b3f14675a67eca27eaa5db0ea415bede2c46b773f13ac5827651c95900
-size 398580
+oid sha256:4373c0af8ff5b0a8415c91b896b49cfb172c786d5648128386639a6b1d15e0ae
+size 400698