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": 77130.21727915127,
-      "required_pages_per_second_upper_estimate": 0.005336047398132581
+      "required_token_positions_per_second": 77191.17828672851,
+      "required_pages_per_second_upper_estimate": 0.005340264822085799
     },
     {
       "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": 9641.277159893909,
-      "required_pages_per_second_upper_estimate": 0.00403660141408617
+      "required_token_positions_per_second": 9648.897285841063,
+      "required_pages_per_second_upper_estimate": 0.004039791801693919
     },
     {
       "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": 19282.554319787818,
-      "required_pages_per_second_upper_estimate": 0.004016047549658227
+      "required_token_positions_per_second": 19297.794571682127,
+      "required_pages_per_second_upper_estimate": 0.004019221692215341
     },
     {
       "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": 28923.83147968173,
-      "required_pages_per_second_upper_estimate": 0.0017339395330734364
+      "required_token_positions_per_second": 28946.691857523194,
+      "required_pages_per_second_upper_estimate": 0.0017353099778192563
     },
     {
       "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": 28923.83147968173,
-      "required_pages_per_second_upper_estimate": 0.0022487006750558625
+      "required_token_positions_per_second": 28946.691857523194,
+      "required_pages_per_second_upper_estimate": 0.002250477968880861
     },
     {
       "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": 28923.83147968173,
-      "required_pages_per_second_upper_estimate": 0.3710038542307273
+      "required_token_positions_per_second": 28946.691857523194,
+      "required_pages_per_second_upper_estimate": 0.37129708261211625
     }
   ],
   "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": 0.10385574158598843,
-    "producer_p95_seconds": 0.29335949400046957,
-    "gpu_update_seconds": 0.9558899559839447,
-    "producer_batches_per_second": 9.62874064282753,
-    "gpu_batches_per_second": 1.0461455251620995,
-    "producer_headroom": 9.204016469253226,
+    "producer_batch_seconds": 0.09843288482809953,
+    "producer_p95_seconds": 0.29711581299852696,
+    "gpu_update_seconds": 0.9551350508750566,
+    "producer_batches_per_second": 10.159206465871364,
+    "gpu_batches_per_second": 1.0469723617449072,
+    "producer_headroom": 9.70341418463025,
     "sustainable": true,
-    "buffer_seconds": 60.22106722698852,
+    "buffer_seconds": 60.17350820512856,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355148.61055341875,
-    "upload_bytes_per_second": 5804464.322986536,
+    "generation_bytes_per_second": 380438.7147900888,
+    "upload_bytes_per_second": 5733960.791234751,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.20662438823625,
+    "estimated_full_shards_per_hour": 81.6332920339298,
     "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": 1.330655336001655,
-      "bytes": 11531295,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.5732356919979793,
       "bytes": 10605987,
@@ -562,6 +556,12 @@
       "bytes": 326991,
       "pages": 9,
       "shards": 1
+    },
+    {
+      "seconds": 1.214567277002061,
+      "bytes": 5649266,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -278,3 +278,4 @@
 {"step": 584900, "loss": 0.5780168175697327, "lr": 0.0003, "grad_norm": 0.17205930086431406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85612460103, "tokens_per_second": 153380.65364789363, "data_wait_s": 7.360999006778002e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 585000, "loss": 0.6551768779754639, "lr": 0.0003, "grad_norm": 0.15627566228715561, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85627188784, "tokens_per_second": 152485.23323052283, "data_wait_s": 1.150599928223528e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 585100, "loss": 0.5708022713661194, "lr": 0.0003, "grad_norm": 0.18319769625944696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85641789921, "tokens_per_second": 140559.31953142636, "data_wait_s": 9.63099955697544e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 585200, "loss": 0.5328781008720398, "lr": 0.0003, "grad_norm": 0.14028533368816135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85656318075, "tokens_per_second": 150663.712994855, "data_wait_s": 1.3040000339969993e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0fa66c61692a8a292cca03f9a590c5f33e6c81ac6d1c3455097d9a255e7aaf19
-size 597903
+oid sha256:e5989defe3cc5a1b8c0e1ed739c6c99148eaf953acd70f64be5dff1c9c32025a
+size 600021