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": 74709.49412040575,
-      "required_pages_per_second_upper_estimate": 0.00647562960087504
+      "required_token_positions_per_second": 74498.01413780737,
+      "required_pages_per_second_upper_estimate": 0.006457299051974528
     },
     {
       "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": 9338.68676505072,
-      "required_pages_per_second_upper_estimate": 0.003725341776388511
+      "required_token_positions_per_second": 9312.251767225922,
+      "required_pages_per_second_upper_estimate": 0.0037147964605177605
     },
     {
       "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": 18677.37353010144,
-      "required_pages_per_second_upper_estimate": 0.004320681655641697
+      "required_token_positions_per_second": 18624.503534451844,
+      "required_pages_per_second_upper_estimate": 0.004308451112628307
     },
     {
       "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": 28016.060295152158,
-      "required_pages_per_second_upper_estimate": 0.001706031083066299
+      "required_token_positions_per_second": 27936.755301677767,
+      "required_pages_per_second_upper_estimate": 0.001701201825044853
     },
     {
       "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": 28016.060295152158,
-      "required_pages_per_second_upper_estimate": 0.0021567538749393497
+      "required_token_positions_per_second": 27936.755301677767,
+      "required_pages_per_second_upper_estimate": 0.0021506487570114186
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28016.060295152158,
-      "required_pages_per_second_upper_estimate": 0.3478657051436254
+      "required_token_positions_per_second": 27936.755301677767,
+      "required_pages_per_second_upper_estimate": 0.34688100254141285
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.12254149739027298,
-    "producer_p95_seconds": 0.3854070729939849,
-    "gpu_update_seconds": 0.9868625248777093,
-    "producer_batches_per_second": 8.16050090211626,
-    "gpu_batches_per_second": 1.013312365999427,
-    "producer_headroom": 8.053292524529278,
+    "producer_batch_seconds": 0.12385072914054263,
+    "producer_p95_seconds": 0.4090459990038653,
+    "gpu_update_seconds": 0.9896639642449664,
+    "producer_batches_per_second": 8.074235872000605,
+    "gpu_batches_per_second": 1.0104439851590628,
+    "producer_headroom": 7.990780281333032,
     "sustainable": true,
-    "buffer_seconds": 60.198614017540265,
+    "buffer_seconds": 60.36950181894295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 439594.67213366413,
-    "upload_bytes_per_second": 3822118.1611488024,
+    "generation_bytes_per_second": 437466.7438165025,
+    "upload_bytes_per_second": 3782392.912432931,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.3267833996529,
+    "estimated_full_shards_per_hour": 93.870179518426,
     "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": 1.3272996070008958,
-      "bytes": 12496214,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 1.0267054780015314,
       "bytes": 5908271,
@@ -562,6 +556,12 @@
       "bytes": 6086180,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.1158472850002,
+      "bytes": 7287693,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -373,3 +373,4 @@
 {"step": 352300, "loss": 0.5755482465028763, "lr": 0.0003, "grad_norm": 0.1562120319513462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51567315290, "tokens_per_second": 148395.2394847214, "data_wait_s": 8.713002898730338e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 352400, "loss": 0.5896095335483551, "lr": 0.0003, "grad_norm": 0.17177140047777942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51582091963, "tokens_per_second": 149442.81288878046, "data_wait_s": 1.4563003787770867e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 352500, "loss": 0.5629769563674927, "lr": 0.0003, "grad_norm": 0.14707417422876704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51596780825, "tokens_per_second": 148714.76398987274, "data_wait_s": 1.1992997315246612e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 352600, "loss": 0.6171035524457693, "lr": 0.0003, "grad_norm": 0.1642221486516797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51611234360, "tokens_per_second": 146443.17700148126, "data_wait_s": 1.555299968458712e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e5777981913e34a48b56fecf11f3e912c7e7c213b4355f0b42917402217027c9
-size 799896
+oid sha256:f42d0c58e3544842ed7da1daae39987b7547c81c0e9354fac4d66a522b14eac8
+size 802014