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": 76849.91544973657,
-      "required_pages_per_second_upper_estimate": 0.00531665546718382
+      "required_token_positions_per_second": 76784.68372596375,
+      "required_pages_per_second_upper_estimate": 0.005312142585174765
     },
     {
       "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": 9606.23943121707,
-      "required_pages_per_second_upper_estimate": 0.004021931848760154
+      "required_token_positions_per_second": 9598.08546574547,
+      "required_pages_per_second_upper_estimate": 0.004018517953691355
     },
     {
       "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": 19212.47886243414,
-      "required_pages_per_second_upper_estimate": 0.00400145267990554
+      "required_token_positions_per_second": 19196.17093149094,
+      "required_pages_per_second_upper_estimate": 0.003998056167959131
     },
     {
       "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": 28818.718293651214,
-      "required_pages_per_second_upper_estimate": 0.0017276381580694491
+      "required_token_positions_per_second": 28794.25639723641,
+      "required_pages_per_second_upper_estimate": 0.0017261717047305346
     },
     {
       "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": 28818.718293651214,
-      "required_pages_per_second_upper_estimate": 0.002240528587186033
+      "required_token_positions_per_second": 28794.25639723641,
+      "required_pages_per_second_upper_estimate": 0.0022386267823362935
     },
     {
       "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": 28818.718293651214,
-      "required_pages_per_second_upper_estimate": 0.3696555751420738
+      "required_token_positions_per_second": 28794.25639723641,
+      "required_pages_per_second_upper_estimate": 0.36934180420000273
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.3344160623516075,
-    "producer_p95_seconds": 0.2671034220002184,
-    "gpu_update_seconds": 0.9593764621409577,
-    "producer_batches_per_second": 2.990286988513706,
-    "gpu_batches_per_second": 1.0423436882831023,
-    "producer_headroom": 2.868810951826418,
+    "producer_batch_seconds": 0.26036417546086454,
+    "producer_p95_seconds": 0.2663148799983901,
+    "gpu_update_seconds": 0.9601914916147505,
+    "producer_batches_per_second": 3.840774170370879,
+    "gpu_batches_per_second": 1.0414589264046734,
+    "producer_headroom": 3.6878786796038203,
     "sustainable": true,
-    "buffer_seconds": 60.440717114880336,
+    "buffer_seconds": 60.49206397172928,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 30532.238588401204,
-    "upload_bytes_per_second": 5001075.567203917,
+    "generation_bytes_per_second": 29635.538210217743,
+    "upload_bytes_per_second": 4978971.241546107,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 6.55150764693286,
+    "estimated_full_shards_per_hour": 6.359096619891159,
     "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": 1.0472652070020558,
-      "bytes": 7166131,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1100580310012447,
       "bytes": 6769711,
@@ -562,6 +556,12 @@
       "bytes": 1135014,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.1626808530018025,
+      "bytes": 5979744,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 711200, "loss": 0.5821572542190552, "lr": 8e-05, "grad_norm": 0.21605494536219572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104099538817, "tokens_per_second": 149477.3624187527, "data_wait_s": 1.6649999452056363e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 711300, "loss": 0.5673778653144836, "lr": 8e-05, "grad_norm": 0.2223724790667016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104114177680, "tokens_per_second": 152747.48639292983, "data_wait_s": 1.548200089018792e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 711400, "loss": 0.5594947338104248, "lr": 8e-05, "grad_norm": 0.21541486985756092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104128815256, "tokens_per_second": 152396.34985180062, "data_wait_s": 1.97020017367322e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 711500, "loss": 0.5800469517707825, "lr": 8e-05, "grad_norm": 0.21757968886907136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104143551438, "tokens_per_second": 130549.11382430035, "data_wait_s": 1.709700154606253e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15b945927d8391dd12ef14f9611a7d45f4a8ff1d7e8f00696dbd95b65e8380f3
-size 563232
+oid sha256:cfee32c29b307ef6e595167ab99893aeb9538426d29aa519d4fbabd4ddf10464
+size 565350