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": 77430.1918528613,
-      "required_pages_per_second_upper_estimate": 0.005356800335178747
+      "required_token_positions_per_second": 77500.10312092103,
+      "required_pages_per_second_upper_estimate": 0.00536163695891961
     },
     {
       "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": 9678.773981607663,
-      "required_pages_per_second_upper_estimate": 0.00405230055031504
+      "required_token_positions_per_second": 9687.51289011513,
+      "required_pages_per_second_upper_estimate": 0.004055959348817956
     },
     {
       "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": 19357.547963215326,
-      "required_pages_per_second_upper_estimate": 0.004031666747868804
+      "required_token_positions_per_second": 19375.02578023026,
+      "required_pages_per_second_upper_estimate": 0.004035306916231984
     },
     {
       "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": 29036.321944822987,
-      "required_pages_per_second_upper_estimate": 0.00174068316469566
+      "required_token_positions_per_second": 29062.538670345388,
+      "required_pages_per_second_upper_estimate": 0.001742254817360106
     },
     {
       "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": 29036.321944822987,
-      "required_pages_per_second_upper_estimate": 0.0022574463139273305
+      "required_token_positions_per_second": 29062.538670345388,
+      "required_pages_per_second_upper_estimate": 0.0022594845490215083
     },
     {
       "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": 29036.321944822987,
-      "required_pages_per_second_upper_estimate": 0.3724467611347082
+      "required_token_positions_per_second": 29062.538670345388,
+      "required_pages_per_second_upper_estimate": 0.37278304114038285
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08475807853912443,
-    "producer_p95_seconds": 0.2233907280024141,
-    "gpu_update_seconds": 0.9521867147133448,
-    "producer_batches_per_second": 11.798285393390541,
-    "gpu_batches_per_second": 1.050214190712637,
-    "producer_headroom": 11.234170607982982,
+    "producer_batch_seconds": 0.06208423249995576,
+    "producer_p95_seconds": 0.21391431799929705,
+    "gpu_update_seconds": 0.9513277664284454,
+    "producer_batches_per_second": 16.10714926693686,
+    "gpu_batches_per_second": 1.0511624229725618,
+    "producer_headroom": 15.323178335644615,
     "sustainable": true,
-    "buffer_seconds": 60.93994974165407,
+    "buffer_seconds": 60.884977051420506,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356228.72999961517,
-    "upload_bytes_per_second": 4288190.16309335,
+    "generation_bytes_per_second": 411496.09024404164,
+    "upload_bytes_per_second": 4284925.904424894,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.43839287749616,
+    "estimated_full_shards_per_hour": 88.29748182764946,
     "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": 0.9324971470014134,
-      "bytes": 1088704,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.2869174759998714,
       "bytes": 6108290,
@@ -562,6 +556,12 @@
       "bytes": 9925526,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.0260383269996964,
+      "bytes": 1162438,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -177,3 +177,4 @@
 {"step": 702700, "loss": 0.5287814736366272, "lr": 8e-05, "grad_norm": 0.22470357132468263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102855798182, "tokens_per_second": 152124.93571930652, "data_wait_s": 1.7035999917425215e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 702800, "loss": 0.5411754250526428, "lr": 8e-05, "grad_norm": 0.21744803486100425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102870373097, "tokens_per_second": 151645.50629856766, "data_wait_s": 1.3174001651350409e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 702900, "loss": 0.5611929893493652, "lr": 8e-05, "grad_norm": 0.22294607376503367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102884966095, "tokens_per_second": 152282.90303154627, "data_wait_s": 1.485399843659252e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 703000, "loss": 0.5514082312583923, "lr": 8e-05, "grad_norm": 0.20385894709825972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102899601217, "tokens_per_second": 152508.72064532407, "data_wait_s": 2.1604002540698275e-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:c5a733df0af79aca87e0d351cab012fb69a00a396e6d810ba080d7832e504676
-size 381636
+oid sha256:ba4d7b5a4ef1065dcf866d9081f8a8dab066290c872e3f88a949a2de07c1f907
+size 383754