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": 75001.03398213441,
-      "required_pages_per_second_upper_estimate": 0.006500899537188656
+      "required_token_positions_per_second": 75521.2357243708,
+      "required_pages_per_second_upper_estimate": 0.006545989305766444
     },
     {
       "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": 9375.129247766801,
-      "required_pages_per_second_upper_estimate": 0.003739879227607628
+      "required_token_positions_per_second": 9440.15446554635,
+      "required_pages_per_second_upper_estimate": 0.0037658187591935333
     },
     {
       "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": 18750.258495533602,
-      "required_pages_per_second_upper_estimate": 0.004337542309662847
+      "required_token_positions_per_second": 18880.3089310927,
+      "required_pages_per_second_upper_estimate": 0.004367627189119951
     },
     {
       "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": 28125.387743300405,
-      "required_pages_per_second_upper_estimate": 0.0017126885510617368
+      "required_token_positions_per_second": 28320.463396639047,
+      "required_pages_per_second_upper_estimate": 0.001724567634867206
     },
     {
       "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": 28125.387743300405,
-      "required_pages_per_second_upper_estimate": 0.002165170204535548
+      "required_token_positions_per_second": 28320.463396639047,
+      "required_pages_per_second_upper_estimate": 0.0021801876683335266
     },
     {
       "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": 28125.387743300405,
-      "required_pages_per_second_upper_estimate": 0.349223186154195
+      "required_token_positions_per_second": 28320.463396639047,
+      "required_pages_per_second_upper_estimate": 0.3516453728924475
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.38092782378124923,
-    "producer_p95_seconds": 0.17707849700036604,
-    "gpu_update_seconds": 0.9830264475762074,
-    "producer_batches_per_second": 2.6251692251660192,
-    "gpu_batches_per_second": 1.0172666284469185,
-    "producer_headroom": 2.5806107777013367,
+    "producer_batch_seconds": 0.08075839533596252,
+    "producer_p95_seconds": 0.17233299800045643,
+    "gpu_update_seconds": 0.9762552120980177,
+    "producer_batches_per_second": 12.382613545500824,
+    "gpu_batches_per_second": 1.024322316140012,
+    "producer_headroom": 12.088591013190694,
     "sustainable": true,
-    "buffer_seconds": 60.94763974972486,
+    "buffer_seconds": 60.5278231500771,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418044.437239725,
+    "generation_bytes_per_second": 401568.55318997055,
     "upload_bytes_per_second": 6063844.855986559,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.7026046552068,
+    "estimated_full_shards_per_hour": 86.1672634770807,
     "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": [

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -49,3 +49,4 @@
 {"step": 44900, "loss": 0.6635061390697956, "lr": 0.0003, "grad_norm": 0.1308439027483963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6570181205, "tokens_per_second": 149633.56542232283, "data_wait_s": 8.383000022149645e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 45000, "loss": 0.624427679926157, "lr": 0.0003, "grad_norm": 0.12727347341567702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6584864111, "tokens_per_second": 149770.12722589058, "data_wait_s": 7.5930001912638545e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 45100, "loss": 0.6282200925052166, "lr": 0.0003, "grad_norm": 0.13956308911328189, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6599624209, "tokens_per_second": 138713.12959167766, "data_wait_s": 1.050100036081858e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 45200, "loss": 0.6508970856666565, "lr": 0.0003, "grad_norm": 0.12931241013622158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6614419603, "tokens_per_second": 151655.39440942844, "data_wait_s": 1.0557000678090844e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d9c9f059a09952e63ef0fef22e72bd46265fc9bd6f23564397646ca5ba8208ee
-size 108966
+oid sha256:6715fceb630722a617cc7efe6ef1af7b1105a5f7e6c124608ddc50c65c53fb19
+size 111084