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": 77056.88309970897,
-      "required_pages_per_second_upper_estimate": 0.005330973969440022
+      "required_token_positions_per_second": 76196.68355215853,
+      "required_pages_per_second_upper_estimate": 0.005271463368802546
     },
     {
       "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": 9632.110387463621,
-      "required_pages_per_second_upper_estimate": 0.004032763477893584
+      "required_token_positions_per_second": 9524.585444019816,
+      "required_pages_per_second_upper_estimate": 0.003987745029449812
     },
     {
       "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": 19264.220774927242,
-      "required_pages_per_second_upper_estimate": 0.00401222915575186
+      "required_token_positions_per_second": 19049.170888039633,
+      "required_pages_per_second_upper_estimate": 0.003967439935560063
     },
     {
       "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": 28896.331162390863,
-      "required_pages_per_second_upper_estimate": 0.0017322909310424023
+      "required_token_positions_per_second": 28573.756332059453,
+      "required_pages_per_second_upper_estimate": 0.001712953062507281
     },
     {
       "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": 28896.331162390863,
-      "required_pages_per_second_upper_estimate": 0.0022465626463476097
+      "required_token_positions_per_second": 28573.756332059453,
+      "required_pages_per_second_upper_estimate": 0.0022214838721460717
     },
     {
       "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": 28896.331162390863,
-      "required_pages_per_second_upper_estimate": 0.3706511096880602
+      "required_token_positions_per_second": 28573.756332059453,
+      "required_pages_per_second_upper_estimate": 0.3665134661184368
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06365825606248698,
-    "producer_p95_seconds": 0.13126471700002185,
-    "gpu_update_seconds": 0.9567996658338555,
-    "producer_batches_per_second": 15.708881484569721,
-    "gpu_batches_per_second": 1.0451508666952714,
-    "producer_headroom": 15.030252555059949,
+    "producer_batch_seconds": 0.3942281194296786,
+    "producer_p95_seconds": 0.13253080799995587,
+    "gpu_update_seconds": 0.9676011679633187,
+    "producer_batches_per_second": 2.5366024154915143,
+    "gpu_batches_per_second": 1.0334836636306226,
+    "producer_headroom": 2.454419459888165,
     "sustainable": true,
-    "buffer_seconds": 60.27837894753289,
+    "buffer_seconds": 60.95887358168908,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 61,
+    "upload_pages_per_commit": 62,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438392.07081361074,
+    "generation_bytes_per_second": 319756.1152226966,
     "upload_bytes_per_second": 5376576.787066994,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.06873315149538,
+    "estimated_full_shards_per_hour": 68.61221878538774,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -6,3 +6,4 @@
 {"step": 440800, "loss": 0.7072717547416687, "lr": 0.0003, "grad_norm": 0.17982194413959549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64520273027, "tokens_per_second": 153151.86996610765, "data_wait_s": 1.6521000134162023e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 440900, "loss": 0.601512610912323, "lr": 0.0003, "grad_norm": 0.1487570383763204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64534927260, "tokens_per_second": 151652.9609486939, "data_wait_s": 1.371000007566181e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 441000, "loss": 0.5863040685653687, "lr": 0.0003, "grad_norm": 0.15476957256013993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64549342442, "tokens_per_second": 149155.8113959861, "data_wait_s": 1.2566999885166297e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 441100, "loss": 0.6048697233200073, "lr": 0.0003, "grad_norm": 0.1563407855641162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64563846639, "tokens_per_second": 145515.66358179986, "data_wait_s": 1.3674000001628883e-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 sha256:4c1d6a8c1a7e3bd5bf2dea6486b51feafe69915f6e78e25a1e9db9ca005e6711
-size 17109
+oid sha256:58b4938abbe1deda572f02d29b7465a52a61f1c81037c32b8841a324921757c7
+size 19227