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": 76084.91165876322,
-      "required_pages_per_second_upper_estimate": 0.006594847307135983
+      "required_token_positions_per_second": 75090.18674026288,
+      "required_pages_per_second_upper_estimate": 0.006508627072307632
     },
     {
       "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": 9510.613957345402,
-      "required_pages_per_second_upper_estimate": 0.0037939261039354566
+      "required_token_positions_per_second": 9386.27334253286,
+      "required_pages_per_second_upper_estimate": 0.003744324773628874
     },
     {
       "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": 19021.227914690804,
-      "required_pages_per_second_upper_estimate": 0.004400226315885967
+      "required_token_positions_per_second": 18772.54668506572,
+      "required_pages_per_second_upper_estimate": 0.004342698290052365
     },
     {
       "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": 28531.841872036206,
-      "required_pages_per_second_upper_estimate": 0.001737439475001742
+      "required_token_positions_per_second": 28158.82002759858,
+      "required_pages_per_second_upper_estimate": 0.001714724401769856
     },
     {
       "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": 28531.841872036206,
-      "required_pages_per_second_upper_estimate": 0.00219646016850267
+      "required_token_positions_per_second": 28158.82002759858,
+      "required_pages_per_second_upper_estimate": 0.0021677439143273076
     },
     {
       "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": 28531.841872036206,
-      "required_pages_per_second_upper_estimate": 0.35426998611863125
+      "required_token_positions_per_second": 28158.82002759858,
+      "required_pages_per_second_upper_estimate": 0.34963830323452055
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0923238096562784,
-    "producer_p95_seconds": 0.24979910800175276,
-    "gpu_update_seconds": 0.9690226142427051,
-    "producer_batches_per_second": 10.831442113610786,
-    "gpu_batches_per_second": 1.0319676603022354,
-    "producer_headroom": 10.495912352949654,
+    "producer_batch_seconds": 0.08711004778135134,
+    "producer_p95_seconds": 0.24109585899714148,
+    "gpu_update_seconds": 0.9818593241088255,
+    "producer_batches_per_second": 11.479731965134814,
+    "gpu_batches_per_second": 1.018475840118583,
+    "producer_headroom": 11.271481868237748,
     "sustainable": true,
-    "buffer_seconds": 60.07940208304772,
+    "buffer_seconds": 60.87527809474718,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360308.6133553488,
-    "upload_bytes_per_second": 6428109.525620583,
+    "generation_bytes_per_second": 369594.8922087617,
+    "upload_bytes_per_second": 6345393.411090532,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.31384087081288,
+    "estimated_full_shards_per_hour": 79.3064601392473,
     "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": 1.5646253450013319,
-      "bytes": 15407220,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.8243736469994474,
       "bytes": 1000268,
@@ -562,6 +556,12 @@
       "bytes": 6011206,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9119941530007054,
+      "bytes": 5747516,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -194,3 +194,4 @@
 {"step": 59400, "loss": 0.6136912349611521, "lr": 0.0003, "grad_norm": 0.12284896022984469, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8693908288, "tokens_per_second": 149322.3416025906, "data_wait_s": 8.126000466290861e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 59500, "loss": 0.6615736372768879, "lr": 0.0003, "grad_norm": 0.13941057722251005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8708364498, "tokens_per_second": 147979.99047645446, "data_wait_s": 8.458999218419194e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 59600, "loss": 0.5707687921822071, "lr": 0.0003, "grad_norm": 0.14786385360195126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8722874790, "tokens_per_second": 141383.39511530905, "data_wait_s": 8.326001989189535e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 59700, "loss": 0.6407595612108707, "lr": 0.0003, "grad_norm": 0.13687393807649878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8737571646, "tokens_per_second": 149532.1105406038, "data_wait_s": 1.1885000276379287e-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:f8379e15299ffaebe06f7bb1fd99744e943a09bf09bc2ec4c61cde3797ddad0c
-size 417642
+oid sha256:b446fff40a230efda7be113db198126a7bcc7281c1288349715078e247cbba73
+size 419760