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": 76725.44338831179,
-      "required_pages_per_second_upper_estimate": 0.005308044201159481
+      "required_token_positions_per_second": 76463.43130144365,
+      "required_pages_per_second_upper_estimate": 0.005289917597038142
     },
     {
       "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": 9590.680423538974,
-      "required_pages_per_second_upper_estimate": 0.004015417617154361
+      "required_token_positions_per_second": 9557.928912680456,
+      "required_pages_per_second_upper_estimate": 0.004001705243487219
     },
     {
       "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": 19181.360847077947,
-      "required_pages_per_second_upper_estimate": 0.003994971617944095
+      "required_token_positions_per_second": 19115.857825360912,
+      "required_pages_per_second_upper_estimate": 0.003981329065951284
     },
     {
       "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": 28772.041270616923,
-      "required_pages_per_second_upper_estimate": 0.0017248399418102306
+      "required_token_positions_per_second": 28673.78673804137,
+      "required_pages_per_second_upper_estimate": 0.0017189497326083106
     },
     {
       "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": 28772.041270616923,
-      "required_pages_per_second_upper_estimate": 0.0022368996539556436
+      "required_token_positions_per_second": 28673.78673804137,
+      "required_pages_per_second_upper_estimate": 0.0022292607962239096
     },
     {
       "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": 28772.041270616923,
-      "required_pages_per_second_upper_estimate": 0.36905685240847247
+      "required_token_positions_per_second": 28673.78673804137,
+      "required_pages_per_second_upper_estimate": 0.3677965487620909
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07246028678132177,
-    "producer_p95_seconds": 0.19598625300022832,
-    "gpu_update_seconds": 0.9609328632596938,
-    "producer_batches_per_second": 13.800663017218032,
-    "gpu_batches_per_second": 1.0406554279013644,
-    "producer_headroom": 13.26151062801749,
+    "producer_batch_seconds": 0.08926244188275234,
+    "producer_p95_seconds": 0.1922143629999482,
+    "gpu_update_seconds": 0.9642256271411664,
+    "producer_batches_per_second": 11.202920051341595,
+    "gpu_batches_per_second": 1.0371016615321675,
+    "producer_headroom": 10.802142612317198,
     "sustainable": true,
-    "buffer_seconds": 60.53877038536071,
+    "buffer_seconds": 60.746214509893484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342612.2677462291,
+    "generation_bytes_per_second": 356605.8230524657,
     "upload_bytes_per_second": 3992013.6727115987,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.51661705293803,
+    "estimated_full_shards_per_hour": 76.51930826836087,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -73,3 +73,4 @@
 {"step": 807300, "loss": 0.5453002452850342, "lr": 3e-05, "grad_norm": 0.26718484432894285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118165450149, "tokens_per_second": 152626.63814740037, "data_wait_s": 1.6449999748147093e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 807400, "loss": 0.5405596494674683, "lr": 3e-05, "grad_norm": 0.29513250157788784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118179960612, "tokens_per_second": 150440.13406134685, "data_wait_s": 1.453700133424718e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 807500, "loss": 0.5008872747421265, "lr": 3e-05, "grad_norm": 0.26396072337872695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118194447543, "tokens_per_second": 150078.94842320183, "data_wait_s": 1.4904000636306591e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 807600, "loss": 0.5238015055656433, "lr": 3e-05, "grad_norm": 0.274179345818706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118209036743, "tokens_per_second": 151056.4577194445, "data_wait_s": 1.3855000361218117e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d13ab851b8c2e940682b17b2692be7f76932bd64e4d9fad9116cf5aeea4a4453
-size 159798
+oid sha256:c047ce3dfb13585131d736aa38f9216b8415872f0e6ae1f24d17844977eb5b34
+size 161916