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": 76109.71383536066,
-      "required_pages_per_second_upper_estimate": 0.006596997096942908
+      "required_token_positions_per_second": 76262.76048944978,
+      "required_pages_per_second_upper_estimate": 0.006610262793026155
     },
     {
       "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": 9513.714229420082,
-      "required_pages_per_second_upper_estimate": 0.0037951628488192444
+      "required_token_positions_per_second": 9532.845061181222,
+      "required_pages_per_second_upper_estimate": 0.0038027944236401876
     },
     {
       "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": 19027.428458840164,
-      "required_pages_per_second_upper_estimate": 0.004401660702649072
+      "required_token_positions_per_second": 19065.690122362445,
+      "required_pages_per_second_upper_estimate": 0.004410511865122672
     },
     {
       "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": 28541.142688260243,
-      "required_pages_per_second_upper_estimate": 0.0017380058459121744
+      "required_token_positions_per_second": 28598.535183543667,
+      "required_pages_per_second_upper_estimate": 0.0017415007477597836
     },
     {
       "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": 28541.142688260243,
-      "required_pages_per_second_upper_estimate": 0.0021971761710819038
+      "required_token_positions_per_second": 28598.535183543667,
+      "required_pages_per_second_upper_estimate": 0.002201594404241417
     },
     {
       "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": 28541.142688260243,
-      "required_pages_per_second_upper_estimate": 0.3543854711283043
+      "required_token_positions_per_second": 28598.535183543667,
+      "required_pages_per_second_upper_estimate": 0.3550980938393989
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10184002737503306,
-    "producer_p95_seconds": 0.3187774100006209,
-    "gpu_update_seconds": 0.9687068349709902,
-    "producer_batches_per_second": 9.81932179100296,
-    "gpu_batches_per_second": 1.0323040613520054,
-    "producer_headroom": 9.512044133724151,
+    "producer_batch_seconds": 0.07972038933598924,
+    "producer_p95_seconds": 0.30753542900129105,
+    "gpu_update_seconds": 0.9667628017503977,
+    "producer_batches_per_second": 12.543842401288382,
+    "gpu_batches_per_second": 1.0343798894510874,
+    "producer_headroom": 12.126920224584994,
     "sustainable": true,
-    "buffer_seconds": 60.05982376820139,
+    "buffer_seconds": 60.90605651027506,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280325.63733321265,
-    "upload_bytes_per_second": 3724626.803835945,
+    "generation_bytes_per_second": 301765.5289537398,
+    "upload_bytes_per_second": 3666102.0889741345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.151356124852036,
+    "estimated_full_shards_per_hour": 64.75185777148386,
     "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": 1.2704945900004532,
-      "bytes": 9798343,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.7868170459987596,
       "bytes": 2512813,
@@ -562,6 +556,12 @@
       "bytes": 7326926,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.7280326340005558,
+      "bytes": 1168675,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -301,3 +301,4 @@
 {"step": 70100, "loss": 0.6081743184477091, "lr": 0.0003, "grad_norm": 0.13843627369027564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10259436304, "tokens_per_second": 139876.96378341693, "data_wait_s": 1.1140997230540961e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 70200, "loss": 0.5846885461360216, "lr": 0.0003, "grad_norm": 0.12082818025227184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10274102827, "tokens_per_second": 149451.26097893732, "data_wait_s": 1.1459000234026462e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 70300, "loss": 0.6057273186743259, "lr": 0.0003, "grad_norm": 0.1319615624256686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10288636575, "tokens_per_second": 149223.92229449312, "data_wait_s": 8.087998139671981e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 70400, "loss": 0.6433261074125767, "lr": 0.0003, "grad_norm": 0.13315337529539414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10303344006, "tokens_per_second": 130383.0157648406, "data_wait_s": 1.4115001249592751e-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 sha256:01f6da07cbf8ee6b296c692cb22695651ae43737dee6345c0db8a113526ddee7
-size 646617
+oid sha256:b96df1a3f8f0e506940e7d858d7cd01912452b2dc8cbb0753f35a592298d52d5
+size 648735