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": 77291.83183031957,
-      "required_pages_per_second_upper_estimate": 0.005347228267779827
+      "required_token_positions_per_second": 77334.79032492285,
+      "required_pages_per_second_upper_estimate": 0.005350200236113922
     },
     {
       "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": 9661.478978789946,
-      "required_pages_per_second_upper_estimate": 0.00404505949379599
+      "required_token_positions_per_second": 9666.848790615357,
+      "required_pages_per_second_upper_estimate": 0.004047307721872868
     },
     {
       "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": 19322.957957579893,
-      "required_pages_per_second_upper_estimate": 0.004024462561894734
+      "required_token_positions_per_second": 19333.697581230714,
+      "required_pages_per_second_upper_estimate": 0.0040266993422782785
     },
     {
       "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": 28984.436936369835,
-      "required_pages_per_second_upper_estimate": 0.0017375727376627143
+      "required_token_positions_per_second": 29000.546371846074,
+      "required_pages_per_second_upper_estimate": 0.001738538473721831
     },
     {
       "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": 28984.436936369835,
-      "required_pages_per_second_upper_estimate": 0.0022534124827381314
+      "required_token_positions_per_second": 29000.546371846074,
+      "required_pages_per_second_upper_estimate": 0.0022546649204884905
     },
     {
       "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": 28984.436936369835,
-      "required_pages_per_second_upper_estimate": 0.37178123595605284
+      "required_token_positions_per_second": 29000.546371846074,
+      "required_pages_per_second_upper_estimate": 0.37198787049737786
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08685246953115211,
-    "producer_p95_seconds": 0.20435118999921542,
-    "gpu_update_seconds": 0.9538912230965968,
-    "producer_batches_per_second": 11.513777390536045,
-    "gpu_batches_per_second": 1.0483375628027285,
-    "producer_headroom": 10.982891197520372,
+    "producer_batch_seconds": 0.06206607628897132,
+    "producer_p95_seconds": 0.2096787360005692,
+    "gpu_update_seconds": 0.9533613486275854,
+    "producer_batches_per_second": 16.111861096940206,
+    "gpu_batches_per_second": 1.0489202246761455,
+    "producer_headroom": 15.360425624279243,
     "sustainable": true,
-    "buffer_seconds": 60.0951470550856,
+    "buffer_seconds": 60.061764963537875,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 170,
+    "upload_pages_per_commit": 174,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 658358.5569485612,
-    "upload_bytes_per_second": 5098289.182768334,
+    "generation_bytes_per_second": 667909.1343064612,
+    "upload_bytes_per_second": 5052014.96645505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.63421025916399,
+    "estimated_full_shards_per_hour": 71.65887604663551,
     "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": 1.6566883329996926,
-      "bytes": 13442976,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 1.1581910269997024,
       "bytes": 10184602,
@@ -562,6 +556,12 @@
       "bytes": 2480365,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.959426125000391,
+      "bytes": 5985197,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 453600, "loss": 0.5883230566978455, "lr": 0.0003, "grad_norm": 0.15073408393150328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66393277437, "tokens_per_second": 152126.32827518907, "data_wait_s": 1.2428999980329536e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 453700, "loss": 0.5846161842346191, "lr": 0.0003, "grad_norm": 0.1697625851453006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66407863997, "tokens_per_second": 151910.43086955216, "data_wait_s": 1.4471999747911468e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 453800, "loss": 0.5488110184669495, "lr": 0.0003, "grad_norm": 0.14646073473339455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66422405460, "tokens_per_second": 151530.03326908365, "data_wait_s": 1.438100116502028e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 453900, "loss": 0.5898391008377075, "lr": 0.0003, "grad_norm": 0.1594489517557798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66437003480, "tokens_per_second": 151346.54094109903, "data_wait_s": 1.332300053036306e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9fd6dd88a9522d805e48494fe9fcd17c774f7d17186603b45a95402414b4c9b2
-size 289779
+oid sha256:436af99938f0fccd6012938309f2ecda3f43b921fb15287b8ac62e721578899e
+size 291897