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": 76641.01325037554,
-      "required_pages_per_second_upper_estimate": 0.005302203128312141
+      "required_token_positions_per_second": 76782.53804734116,
+      "required_pages_per_second_upper_estimate": 0.005311994142148994
     },
     {
       "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": 9580.126656296943,
-      "required_pages_per_second_upper_estimate": 0.004010998975197844
+      "required_token_positions_per_second": 9597.817255917646,
+      "required_pages_per_second_upper_estimate": 0.004018405659837304
     },
     {
       "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": 19160.253312593886,
-      "required_pages_per_second_upper_estimate": 0.003990575475154182
+      "required_token_positions_per_second": 19195.63451183529,
+      "required_pages_per_second_upper_estimate": 0.003997944445891197
     },
     {
       "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": 28740.379968890833,
-      "required_pages_per_second_upper_estimate": 0.0017229418951157601
+      "required_token_positions_per_second": 28793.451767752933,
+      "required_pages_per_second_upper_estimate": 0.0017261234684215981
     },
     {
       "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": 28740.379968890833,
-      "required_pages_per_second_upper_estimate": 0.0022344381270097886
+      "required_token_positions_per_second": 28793.451767752933,
+      "required_pages_per_second_upper_estimate": 0.0022385642259331443
     },
     {
       "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": 28740.379968890833,
-      "required_pages_per_second_upper_estimate": 0.3686507352251874
+      "required_token_positions_per_second": 28793.451767752933,
+      "required_pages_per_second_upper_estimate": 0.3693314832769325
     }
   ],
   "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": 0.0919376304297117,
-    "producer_p95_seconds": 0.24079475100006675,
-    "gpu_update_seconds": 0.9619914569649655,
-    "producer_batches_per_second": 10.876939021878767,
-    "gpu_batches_per_second": 1.039510270865554,
-    "producer_headroom": 10.463522416976241,
+    "producer_batch_seconds": 0.34768476671089843,
+    "producer_p95_seconds": 0.2460044769995875,
+    "gpu_update_seconds": 0.9602183240483944,
+    "producer_batches_per_second": 2.8761685749422115,
+    "gpu_batches_per_second": 1.0414298237757862,
+    "producer_headroom": 2.761749768711669,
     "sustainable": true,
-    "buffer_seconds": 60.60546178879282,
+    "buffer_seconds": 60.49375441504885,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 473714.705895342,
-    "upload_bytes_per_second": 3941076.4282295653,
+    "generation_bytes_per_second": 471753.7394167999,
+    "upload_bytes_per_second": 4010500.1843358036,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.6481483711738,
+    "estimated_full_shards_per_hour": 101.227370613842,
     "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": 1.375044964999688,
-      "bytes": 2500711,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.54583506400013,
       "bytes": 5956633,
@@ -562,6 +556,12 @@
       "bytes": 7829235,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.7852341269990575,
+      "bytes": 11660999,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -127,3 +127,4 @@
 {"step": 812700, "loss": 0.509462296962738, "lr": 3e-05, "grad_norm": 0.26544499288587536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118956098928, "tokens_per_second": 149816.37113361756, "data_wait_s": 1.434399928257335e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 812800, "loss": 0.4952360689640045, "lr": 3e-05, "grad_norm": 0.2751130453655626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118970647257, "tokens_per_second": 150455.07692778192, "data_wait_s": 8.009999874047935e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 812900, "loss": 0.5041095614433289, "lr": 3e-05, "grad_norm": 0.2833444949896702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118985207065, "tokens_per_second": 150032.91286971106, "data_wait_s": 2.9550999897764996e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 813000, "loss": 0.5881887078285217, "lr": 3e-05, "grad_norm": 0.32019736391769926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118999865026, "tokens_per_second": 151582.16613952265, "data_wait_s": 1.4104998626862653e-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 sha256:7d476f42f38e76982f1b46d9e2e79a13c161b3da94decce3aab9a1d3051a0c02
-size 274953
+oid sha256:39414e0ba89902cd345dea681d9c242a48f41bd6445f68be4c10a87da4550864
+size 277071