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": 76481.21522314135,
-      "required_pages_per_second_upper_estimate": 0.005291147930005573
+      "required_token_positions_per_second": 77122.45039180358,
+      "required_pages_per_second_upper_estimate": 0.005335510066844205
     },
     {
       "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": 9560.151902892669,
-      "required_pages_per_second_upper_estimate": 0.0040026359630677575
+      "required_token_positions_per_second": 9640.306298975447,
+      "required_pages_per_second_upper_estimate": 0.004036194934893491
     },
     {
       "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": 19120.303805785337,
-      "required_pages_per_second_upper_estimate": 0.003982255046425303
+      "required_token_positions_per_second": 19280.612597950894,
+      "required_pages_per_second_upper_estimate": 0.0040156431402062435
     },
     {
       "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": 28680.455708678008,
-      "required_pages_per_second_upper_estimate": 0.0017193495272150485
+      "required_token_positions_per_second": 28920.91889692634,
+      "required_pages_per_second_upper_estimate": 0.0017337649281845074
     },
     {
       "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": 28680.455708678008,
-      "required_pages_per_second_upper_estimate": 0.002229779279357206
+      "required_token_positions_per_second": 28920.91889692634,
+      "required_pages_per_second_upper_estimate": 0.0022484742345577286
     },
     {
       "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": 28680.455708678008,
-      "required_pages_per_second_upper_estimate": 0.3678820911568349
+      "required_token_positions_per_second": 28920.91889692634,
+      "required_pages_per_second_upper_estimate": 0.3709664947464289
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.4653944427341514,
-    "producer_p95_seconds": 0.25380113900246215,
-    "gpu_update_seconds": 0.9640014189744687,
-    "producer_batches_per_second": 2.148714956983775,
-    "gpu_batches_per_second": 1.0373428714076247,
-    "producer_headroom": 2.0713642675040234,
+    "producer_batch_seconds": 0.1026370459688053,
+    "producer_p95_seconds": 0.25766067299991846,
+    "gpu_update_seconds": 0.955986222240621,
+    "producer_batches_per_second": 9.74307074566363,
+    "gpu_batches_per_second": 1.0460401800103567,
+    "producer_headroom": 9.314241395170084,
     "sustainable": true,
-    "buffer_seconds": 60.732089395391526,
+    "buffer_seconds": 60.22713200115913,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299462.88439957035,
-    "upload_bytes_per_second": 3474194.6067738794,
+    "generation_bytes_per_second": 300157.0440886023,
+    "upload_bytes_per_second": 3518262.3282989003,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.25776385298093,
+    "estimated_full_shards_per_hour": 64.40671436303666,
     "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": 1.2981046400000196,
-      "bytes": 6199716,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.6966266469989932,
       "bytes": 7696834,
@@ -562,6 +556,12 @@
       "bytes": 1215408,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.4002689310000278,
+      "bytes": 11880783,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -196,3 +196,4 @@
 {"step": 862100, "loss": 0.5545560717582703, "lr": 3e-05, "grad_norm": 0.3294269875792365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126186842363, "tokens_per_second": 152316.9902996179, "data_wait_s": 8.769999112701043e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 862200, "loss": 0.5252000689506531, "lr": 3e-05, "grad_norm": 0.2944070911137944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126201412226, "tokens_per_second": 130448.41422557177, "data_wait_s": 1.3577999197877944e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 862300, "loss": 0.48036885261535645, "lr": 3e-05, "grad_norm": 0.29004395380190934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126216023550, "tokens_per_second": 151791.42873416562, "data_wait_s": 1.180099934572354e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 862400, "loss": 0.5745303630828857, "lr": 3e-05, "grad_norm": 0.32823499344767804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126230716829, "tokens_per_second": 152779.29174583525, "data_wait_s": 9.913997928379104e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e2945d7ea2b2fbd998c6296dbd67b7e1a1132820db7e67543610f87d5a94a281
-size 422661
+oid sha256:94c91bebed073435f59de6b72ad24929a4b8053995778ccfdf0397ba99a7fc05
+size 424779